├── .gitignore ├── LICENSE ├── README.md ├── install.py ├── javascript └── temporaljavascript.js ├── readme_img ├── 1.png ├── 2.png ├── 3.png ├── 4.png └── 5.png ├── requirements.txt ├── scripts ├── Berry_Method.py ├── Ebsynth_Processing.py ├── TemporalKitImg2ImgTab.py ├── berry_utility.py ├── optical_flow_raft.py ├── optical_flow_simple.py ├── sd-TemporalKit-UI.py └── stable_diffusion_processing.py ├── style.css └── temp_file.txt /.gitignore: -------------------------------------------------------------------------------- 1 | ### Example user template template 2 | ### Example user template 3 | 4 | # IntelliJ project files 5 | .idea 6 | *.iml 7 | out 8 | gen 9 | ### Python template 10 | # Byte-compiled / optimized / DLL files 11 | __pycache__/ 12 | *.py[cod] 13 | *$py.class 14 | 15 | # C extensions 16 | *.so 17 | 18 | # Distribution / packaging 19 | .Python 20 | build/ 21 | develop-eggs/ 22 | dist/ 23 | downloads/ 24 | eggs/ 25 | .eggs/ 26 | lib/ 27 | lib64/ 28 | parts/ 29 | sdist/ 30 | var/ 31 | wheels/ 32 | share/python-wheels/ 33 | *.egg-info/ 34 | .installed.cfg 35 | *.egg 36 | MANIFEST 37 | 38 | # PyInstaller 39 | # Usually these files are written by a python script from a template 40 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 41 | *.manifest 42 | *.spec 43 | 44 | # Installer logs 45 | pip-log.txt 46 | pip-delete-this-directory.txt 47 | 48 | # Unit test / coverage reports 49 | htmlcov/ 50 | .tox/ 51 | .nox/ 52 | .coverage 53 | .coverage.* 54 | .cache 55 | nosetests.xml 56 | coverage.xml 57 | *.cover 58 | *.py,cover 59 | .hypothesis/ 60 | .pytest_cache/ 61 | cover/ 62 | 63 | # Translations 64 | *.mo 65 | *.pot 66 | 67 | # Django stuff: 68 | *.log 69 | local_settings.py 70 | db.sqlite3 71 | db.sqlite3-journal 72 | 73 | # Flask stuff: 74 | instance/ 75 | .webassets-cache 76 | 77 | # Scrapy stuff: 78 | .scrapy 79 | 80 | # Sphinx documentation 81 | docs/_build/ 82 | 83 | # PyBuilder 84 | .pybuilder/ 85 | target/ 86 | 87 | # Jupyter Notebook 88 | .ipynb_checkpoints 89 | 90 | # IPython 91 | profile_default/ 92 | ipython_config.py 93 | 94 | # pyenv 95 | # For a library or package, you might want to ignore these files since the code is 96 | # intended to run in multiple environments; otherwise, check them in: 97 | # .python-version 98 | 99 | # pipenv 100 | # According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control. 101 | # However, in case of collaboration, if having platform-specific dependencies or dependencies 102 | # having no cross-platform support, pipenv may install dependencies that don't work, or not 103 | # install all needed dependencies. 104 | #Pipfile.lock 105 | 106 | # poetry 107 | # Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control. 108 | # This is especially recommended for binary packages to ensure reproducibility, and is more 109 | # commonly ignored for libraries. 110 | # https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control 111 | #poetry.lock 112 | 113 | # pdm 114 | # Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control. 115 | #pdm.lock 116 | # pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it 117 | # in version control. 118 | # https://pdm.fming.dev/#use-with-ide 119 | .pdm.toml 120 | 121 | # PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm 122 | __pypackages__/ 123 | 124 | # Celery stuff 125 | celerybeat-schedule 126 | celerybeat.pid 127 | 128 | # SageMath parsed files 129 | *.sage.py 130 | 131 | # Environments 132 | .env 133 | .venv 134 | env/ 135 | venv/ 136 | ENV/ 137 | env.bak/ 138 | venv.bak/ 139 | 140 | # Spyder project settings 141 | .spyderproject 142 | .spyproject 143 | 144 | # Rope project settings 145 | .ropeproject 146 | 147 | # mkdocs documentation 148 | /site 149 | 150 | # mypy 151 | .mypy_cache/ 152 | .dmypy.json 153 | dmypy.json 154 | 155 | # Pyre type checker 156 | .pyre/ 157 | 158 | # pytype static type analyzer 159 | .pytype/ 160 | 161 | # Cython debug symbols 162 | cython_debug/ 163 | 164 | # PyCharm 165 | # JetBrains specific template is maintained in a separate JetBrains.gitignore that can 166 | # be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore 167 | # and can be added to the global gitignore or merged into this file. For a more nuclear 168 | # option (not recommended) you can uncomment the following to ignore the entire idea folder. 169 | #.idea/ 170 | 171 | ### PyCharm+all template 172 | # Covers JetBrains IDEs: IntelliJ, RubyMine, PhpStorm, AppCode, PyCharm, CLion, Android Studio, WebStorm and Rider 173 | # Reference: https://intellij-support.jetbrains.com/hc/en-us/articles/206544839 174 | 175 | # User-specific stuff 176 | .idea/**/workspace.xml 177 | .idea/**/tasks.xml 178 | .idea/**/usage.statistics.xml 179 | .idea/**/dictionaries 180 | .idea/**/shelf 181 | 182 | # AWS User-specific 183 | .idea/**/aws.xml 184 | 185 | # Generated files 186 | .idea/**/contentModel.xml 187 | 188 | # Sensitive or high-churn files 189 | .idea/**/dataSources/ 190 | .idea/**/dataSources.ids 191 | .idea/**/dataSources.local.xml 192 | .idea/**/sqlDataSources.xml 193 | .idea/**/dynamic.xml 194 | .idea/**/uiDesigner.xml 195 | .idea/**/dbnavigator.xml 196 | 197 | # Gradle 198 | .idea/**/gradle.xml 199 | .idea/**/libraries 200 | 201 | # Gradle and Maven with auto-import 202 | # When using Gradle or Maven with auto-import, you should exclude module files, 203 | # since they will be recreated, and may cause churn. Uncomment if using 204 | # auto-import. 205 | # .idea/artifacts 206 | # .idea/compiler.xml 207 | # .idea/jarRepositories.xml 208 | # .idea/modules.xml 209 | # .idea/*.iml 210 | # .idea/modules 211 | # *.iml 212 | # *.ipr 213 | 214 | # CMake 215 | cmake-build-*/ 216 | 217 | # Mongo Explorer plugin 218 | .idea/**/mongoSettings.xml 219 | 220 | # File-based project format 221 | *.iws 222 | 223 | # IntelliJ 224 | out/ 225 | 226 | # mpeltonen/sbt-idea plugin 227 | .idea_modules/ 228 | 229 | # JIRA plugin 230 | atlassian-ide-plugin.xml 231 | 232 | # Cursive Clojure plugin 233 | .idea/replstate.xml 234 | 235 | # SonarLint plugin 236 | .idea/sonarlint/ 237 | 238 | # Crashlytics plugin (for Android Studio and IntelliJ) 239 | com_crashlytics_export_strings.xml 240 | crashlytics.properties 241 | crashlytics-build.properties 242 | fabric.properties 243 | 244 | # Editor-based Rest Client 245 | .idea/httpRequests 246 | 247 | # Android studio 3.1+ serialized cache file 248 | .idea/caches/build_file_checksums.ser 249 | 250 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | GNU GENERAL PUBLIC LICENSE 2 | Version 3, 29 June 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | Preamble 9 | 10 | The GNU General Public License is a free, copyleft license for 11 | software and other kinds of works. 12 | 13 | The licenses for most software and other practical works are designed 14 | to take away your freedom to share and change the works. By contrast, 15 | the GNU General Public License is intended to guarantee your freedom to 16 | share and change all versions of a program--to make sure it remains free 17 | software for all its users. We, the Free Software Foundation, use the 18 | GNU General Public License for most of our software; it applies also to 19 | any other work released this way by its authors. You can apply it to 20 | your programs, too. 21 | 22 | When we speak of free software, we are referring to freedom, not 23 | price. Our General Public Licenses are designed to make sure that you 24 | have the freedom to distribute copies of free software (and charge for 25 | them if you wish), that you receive source code or can get it if you 26 | want it, that you can change the software or use pieces of it in new 27 | free programs, and that you know you can do these things. 28 | 29 | To protect your rights, we need to prevent others from denying you 30 | these rights or asking you to surrender the rights. Therefore, you have 31 | certain responsibilities if you distribute copies of the software, or if 32 | you modify it: responsibilities to respect the freedom of others. 33 | 34 | For example, if you distribute copies of such a program, whether 35 | gratis or for a fee, you must pass on to the recipients the same 36 | freedoms that you received. You must make sure that they, too, receive 37 | or can get the source code. And you must show them these terms so they 38 | know their rights. 39 | 40 | Developers that use the GNU GPL protect your rights with two steps: 41 | (1) assert copyright on the software, and (2) offer you this License 42 | giving you legal permission to copy, distribute and/or modify it. 43 | 44 | For the developers' and authors' protection, the GPL clearly explains 45 | that there is no warranty for this free software. For both users' and 46 | authors' sake, the GPL requires that modified versions be marked as 47 | changed, so that their problems will not be attributed erroneously to 48 | authors of previous versions. 49 | 50 | Some devices are designed to deny users access to install or run 51 | modified versions of the software inside them, although the manufacturer 52 | can do so. This is fundamentally incompatible with the aim of 53 | protecting users' freedom to change the software. The systematic 54 | pattern of such abuse occurs in the area of products for individuals to 55 | use, which is precisely where it is most unacceptable. Therefore, we 56 | have designed this version of the GPL to prohibit the practice for those 57 | products. If such problems arise substantially in other domains, we 58 | stand ready to extend this provision to those domains in future versions 59 | of the GPL, as needed to protect the freedom of users. 60 | 61 | Finally, every program is threatened constantly by software patents. 62 | States should not allow patents to restrict development and use of 63 | software on general-purpose computers, but in those that do, we wish to 64 | avoid the special danger that patents applied to a free program could 65 | make it effectively proprietary. To prevent this, the GPL assures that 66 | patents cannot be used to render the program non-free. 67 | 68 | The precise terms and conditions for copying, distribution and 69 | modification follow. 70 | 71 | TERMS AND CONDITIONS 72 | 73 | 0. Definitions. 74 | 75 | "This License" refers to version 3 of the GNU General Public License. 76 | 77 | "Copyright" also means copyright-like laws that apply to other kinds of 78 | works, such as semiconductor masks. 79 | 80 | "The Program" refers to any copyrightable work licensed under this 81 | License. Each licensee is addressed as "you". "Licensees" and 82 | "recipients" may be individuals or organizations. 83 | 84 | To "modify" a work means to copy from or adapt all or part of the work 85 | in a fashion requiring copyright permission, other than the making of an 86 | exact copy. The resulting work is called a "modified version" of the 87 | earlier work or a work "based on" the earlier work. 88 | 89 | A "covered work" means either the unmodified Program or a work based 90 | on the Program. 91 | 92 | To "propagate" a work means to do anything with it that, without 93 | permission, would make you directly or secondarily liable for 94 | infringement under applicable copyright law, except executing it on a 95 | computer or modifying a private copy. Propagation includes copying, 96 | distribution (with or without modification), making available to the 97 | public, and in some countries other activities as well. 98 | 99 | To "convey" a work means any kind of propagation that enables other 100 | parties to make or receive copies. Mere interaction with a user through 101 | a computer network, with no transfer of a copy, is not conveying. 102 | 103 | An interactive user interface displays "Appropriate Legal Notices" 104 | to the extent that it includes a convenient and prominently visible 105 | feature that (1) displays an appropriate copyright notice, and (2) 106 | tells the user that there is no warranty for the work (except to the 107 | extent that warranties are provided), that licensees may convey the 108 | work under this License, and how to view a copy of this License. If 109 | the interface presents a list of user commands or options, such as a 110 | menu, a prominent item in the list meets this criterion. 111 | 112 | 1. Source Code. 113 | 114 | The "source code" for a work means the preferred form of the work 115 | for making modifications to it. "Object code" means any non-source 116 | form of a work. 117 | 118 | A "Standard Interface" means an interface that either is an official 119 | standard defined by a recognized standards body, or, in the case of 120 | interfaces specified for a particular programming language, one that 121 | is widely used among developers working in that language. 122 | 123 | The "System Libraries" of an executable work include anything, other 124 | than the work as a whole, that (a) is included in the normal form of 125 | packaging a Major Component, but which is not part of that Major 126 | Component, and (b) serves only to enable use of the work with that 127 | Major Component, or to implement a Standard Interface for which an 128 | implementation is available to the public in source code form. A 129 | "Major Component", in this context, means a major essential component 130 | (kernel, window system, and so on) of the specific operating system 131 | (if any) on which the executable work runs, or a compiler used to 132 | produce the work, or an object code interpreter used to run it. 133 | 134 | The "Corresponding Source" for a work in object code form means all 135 | the source code needed to generate, install, and (for an executable 136 | work) run the object code and to modify the work, including scripts to 137 | control those activities. However, it does not include the work's 138 | System Libraries, or general-purpose tools or generally available free 139 | programs which are used unmodified in performing those activities but 140 | which are not part of the work. For example, Corresponding Source 141 | includes interface definition files associated with source files for 142 | the work, and the source code for shared libraries and dynamically 143 | linked subprograms that the work is specifically designed to require, 144 | such as by intimate data communication or control flow between those 145 | subprograms and other parts of the work. 146 | 147 | The Corresponding Source need not include anything that users 148 | can regenerate automatically from other parts of the Corresponding 149 | Source. 150 | 151 | The Corresponding Source for a work in source code form is that 152 | same work. 153 | 154 | 2. Basic Permissions. 155 | 156 | All rights granted under this License are granted for the term of 157 | copyright on the Program, and are irrevocable provided the stated 158 | conditions are met. This License explicitly affirms your unlimited 159 | permission to run the unmodified Program. The output from running a 160 | covered work is covered by this License only if the output, given its 161 | content, constitutes a covered work. This License acknowledges your 162 | rights of fair use or other equivalent, as provided by copyright law. 163 | 164 | You may make, run and propagate covered works that you do not 165 | convey, without conditions so long as your license otherwise remains 166 | in force. You may convey covered works to others for the sole purpose 167 | of having them make modifications exclusively for you, or provide you 168 | with facilities for running those works, provided that you comply with 169 | the terms of this License in conveying all material for which you do 170 | not control copyright. Those thus making or running the covered works 171 | for you must do so exclusively on your behalf, under your direction 172 | and control, on terms that prohibit them from making any copies of 173 | your copyrighted material outside their relationship with you. 174 | 175 | Conveying under any other circumstances is permitted solely under 176 | the conditions stated below. Sublicensing is not allowed; section 10 177 | makes it unnecessary. 178 | 179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 180 | 181 | No covered work shall be deemed part of an effective technological 182 | measure under any applicable law fulfilling obligations under article 183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or 184 | similar laws prohibiting or restricting circumvention of such 185 | measures. 186 | 187 | When you convey a covered work, you waive any legal power to forbid 188 | circumvention of technological measures to the extent such circumvention 189 | is effected by exercising rights under this License with respect to 190 | the covered work, and you disclaim any intention to limit operation or 191 | modification of the work as a means of enforcing, against the work's 192 | users, your or third parties' legal rights to forbid circumvention of 193 | technological measures. 194 | 195 | 4. Conveying Verbatim Copies. 196 | 197 | You may convey verbatim copies of the Program's source code as you 198 | receive it, in any medium, provided that you conspicuously and 199 | appropriately publish on each copy an appropriate copyright notice; 200 | keep intact all notices stating that this License and any 201 | non-permissive terms added in accord with section 7 apply to the code; 202 | keep intact all notices of the absence of any warranty; and give all 203 | recipients a copy of this License along with the Program. 204 | 205 | You may charge any price or no price for each copy that you convey, 206 | and you may offer support or warranty protection for a fee. 207 | 208 | 5. Conveying Modified Source Versions. 209 | 210 | You may convey a work based on the Program, or the modifications to 211 | produce it from the Program, in the form of source code under the 212 | terms of section 4, provided that you also meet all of these conditions: 213 | 214 | a) The work must carry prominent notices stating that you modified 215 | it, and giving a relevant date. 216 | 217 | b) The work must carry prominent notices stating that it is 218 | released under this License and any conditions added under section 219 | 7. This requirement modifies the requirement in section 4 to 220 | "keep intact all notices". 221 | 222 | c) You must license the entire work, as a whole, under this 223 | License to anyone who comes into possession of a copy. This 224 | License will therefore apply, along with any applicable section 7 225 | additional terms, to the whole of the work, and all its parts, 226 | regardless of how they are packaged. This License gives no 227 | permission to license the work in any other way, but it does not 228 | invalidate such permission if you have separately received it. 229 | 230 | d) If the work has interactive user interfaces, each must display 231 | Appropriate Legal Notices; however, if the Program has interactive 232 | interfaces that do not display Appropriate Legal Notices, your 233 | work need not make them do so. 234 | 235 | A compilation of a covered work with other separate and independent 236 | works, which are not by their nature extensions of the covered work, 237 | and which are not combined with it such as to form a larger program, 238 | in or on a volume of a storage or distribution medium, is called an 239 | "aggregate" if the compilation and its resulting copyright are not 240 | used to limit the access or legal rights of the compilation's users 241 | beyond what the individual works permit. Inclusion of a covered work 242 | in an aggregate does not cause this License to apply to the other 243 | parts of the aggregate. 244 | 245 | 6. Conveying Non-Source Forms. 246 | 247 | You may convey a covered work in object code form under the terms 248 | of sections 4 and 5, provided that you also convey the 249 | machine-readable Corresponding Source under the terms of this License, 250 | in one of these ways: 251 | 252 | a) Convey the object code in, or embodied in, a physical product 253 | (including a physical distribution medium), accompanied by the 254 | Corresponding Source fixed on a durable physical medium 255 | customarily used for software interchange. 256 | 257 | b) Convey the object code in, or embodied in, a physical product 258 | (including a physical distribution medium), accompanied by a 259 | written offer, valid for at least three years and valid for as 260 | long as you offer spare parts or customer support for that product 261 | model, to give anyone who possesses the object code either (1) a 262 | copy of the Corresponding Source for all the software in the 263 | product that is covered by this License, on a durable physical 264 | medium customarily used for software interchange, for a price no 265 | more than your reasonable cost of physically performing this 266 | conveying of source, or (2) access to copy the 267 | Corresponding Source from a network server at no charge. 268 | 269 | c) Convey individual copies of the object code with a copy of the 270 | written offer to provide the Corresponding Source. This 271 | alternative is allowed only occasionally and noncommercially, and 272 | only if you received the object code with such an offer, in accord 273 | with subsection 6b. 274 | 275 | d) Convey the object code by offering access from a designated 276 | place (gratis or for a charge), and offer equivalent access to the 277 | Corresponding Source in the same way through the same place at no 278 | further charge. You need not require recipients to copy the 279 | Corresponding Source along with the object code. If the place to 280 | copy the object code is a network server, the Corresponding Source 281 | may be on a different server (operated by you or a third party) 282 | that supports equivalent copying facilities, provided you maintain 283 | clear directions next to the object code saying where to find the 284 | Corresponding Source. Regardless of what server hosts the 285 | Corresponding Source, you remain obligated to ensure that it is 286 | available for as long as needed to satisfy these requirements. 287 | 288 | e) Convey the object code using peer-to-peer transmission, provided 289 | you inform other peers where the object code and Corresponding 290 | Source of the work are being offered to the general public at no 291 | charge under subsection 6d. 292 | 293 | A separable portion of the object code, whose source code is excluded 294 | from the Corresponding Source as a System Library, need not be 295 | included in conveying the object code work. 296 | 297 | A "User Product" is either (1) a "consumer product", which means any 298 | tangible personal property which is normally used for personal, family, 299 | or household purposes, or (2) anything designed or sold for incorporation 300 | into a dwelling. In determining whether a product is a consumer product, 301 | doubtful cases shall be resolved in favor of coverage. For a particular 302 | product received by a particular user, "normally used" refers to a 303 | typical or common use of that class of product, regardless of the status 304 | of the particular user or of the way in which the particular user 305 | actually uses, or expects or is expected to use, the product. A product 306 | is a consumer product regardless of whether the product has substantial 307 | commercial, industrial or non-consumer uses, unless such uses represent 308 | the only significant mode of use of the product. 309 | 310 | "Installation Information" for a User Product means any methods, 311 | procedures, authorization keys, or other information required to install 312 | and execute modified versions of a covered work in that User Product from 313 | a modified version of its Corresponding Source. The information must 314 | suffice to ensure that the continued functioning of the modified object 315 | code is in no case prevented or interfered with solely because 316 | modification has been made. 317 | 318 | If you convey an object code work under this section in, or with, or 319 | specifically for use in, a User Product, and the conveying occurs as 320 | part of a transaction in which the right of possession and use of the 321 | User Product is transferred to the recipient in perpetuity or for a 322 | fixed term (regardless of how the transaction is characterized), the 323 | Corresponding Source conveyed under this section must be accompanied 324 | by the Installation Information. But this requirement does not apply 325 | if neither you nor any third party retains the ability to install 326 | modified object code on the User Product (for example, the work has 327 | been installed in ROM). 328 | 329 | The requirement to provide Installation Information does not include a 330 | requirement to continue to provide support service, warranty, or updates 331 | for a work that has been modified or installed by the recipient, or for 332 | the User Product in which it has been modified or installed. Access to a 333 | network may be denied when the modification itself materially and 334 | adversely affects the operation of the network or violates the rules and 335 | protocols for communication across the network. 336 | 337 | Corresponding Source conveyed, and Installation Information provided, 338 | in accord with this section must be in a format that is publicly 339 | documented (and with an implementation available to the public in 340 | source code form), and must require no special password or key for 341 | unpacking, reading or copying. 342 | 343 | 7. Additional Terms. 344 | 345 | "Additional permissions" are terms that supplement the terms of this 346 | License by making exceptions from one or more of its conditions. 347 | Additional permissions that are applicable to the entire Program shall 348 | be treated as though they were included in this License, to the extent 349 | that they are valid under applicable law. If additional permissions 350 | apply only to part of the Program, that part may be used separately 351 | under those permissions, but the entire Program remains governed by 352 | this License without regard to the additional permissions. 353 | 354 | When you convey a copy of a covered work, you may at your option 355 | remove any additional permissions from that copy, or from any part of 356 | it. (Additional permissions may be written to require their own 357 | removal in certain cases when you modify the work.) You may place 358 | additional permissions on material, added by you to a covered work, 359 | for which you have or can give appropriate copyright permission. 360 | 361 | Notwithstanding any other provision of this License, for material you 362 | add to a covered work, you may (if authorized by the copyright holders of 363 | that material) supplement the terms of this License with terms: 364 | 365 | a) Disclaiming warranty or limiting liability differently from the 366 | terms of sections 15 and 16 of this License; or 367 | 368 | b) Requiring preservation of specified reasonable legal notices or 369 | author attributions in that material or in the Appropriate Legal 370 | Notices displayed by works containing it; or 371 | 372 | c) Prohibiting misrepresentation of the origin of that material, or 373 | requiring that modified versions of such material be marked in 374 | reasonable ways as different from the original version; or 375 | 376 | d) Limiting the use for publicity purposes of names of licensors or 377 | authors of the material; or 378 | 379 | e) Declining to grant rights under trademark law for use of some 380 | trade names, trademarks, or service marks; or 381 | 382 | f) Requiring indemnification of licensors and authors of that 383 | material by anyone who conveys the material (or modified versions of 384 | it) with contractual assumptions of liability to the recipient, for 385 | any liability that these contractual assumptions directly impose on 386 | those licensors and authors. 387 | 388 | All other non-permissive additional terms are considered "further 389 | restrictions" within the meaning of section 10. If the Program as you 390 | received it, or any part of it, contains a notice stating that it is 391 | governed by this License along with a term that is a further 392 | restriction, you may remove that term. If a license document contains 393 | a further restriction but permits relicensing or conveying under this 394 | License, you may add to a covered work material governed by the terms 395 | of that license document, provided that the further restriction does 396 | not survive such relicensing or conveying. 397 | 398 | If you add terms to a covered work in accord with this section, you 399 | must place, in the relevant source files, a statement of the 400 | additional terms that apply to those files, or a notice indicating 401 | where to find the applicable terms. 402 | 403 | Additional terms, permissive or non-permissive, may be stated in the 404 | form of a separately written license, or stated as exceptions; 405 | the above requirements apply either way. 406 | 407 | 8. Termination. 408 | 409 | You may not propagate or modify a covered work except as expressly 410 | provided under this License. Any attempt otherwise to propagate or 411 | modify it is void, and will automatically terminate your rights under 412 | this License (including any patent licenses granted under the third 413 | paragraph of section 11). 414 | 415 | However, if you cease all violation of this License, then your 416 | license from a particular copyright holder is reinstated (a) 417 | provisionally, unless and until the copyright holder explicitly and 418 | finally terminates your license, and (b) permanently, if the copyright 419 | holder fails to notify you of the violation by some reasonable means 420 | prior to 60 days after the cessation. 421 | 422 | Moreover, your license from a particular copyright holder is 423 | reinstated permanently if the copyright holder notifies you of the 424 | violation by some reasonable means, this is the first time you have 425 | received notice of violation of this License (for any work) from that 426 | copyright holder, and you cure the violation prior to 30 days after 427 | your receipt of the notice. 428 | 429 | Termination of your rights under this section does not terminate the 430 | licenses of parties who have received copies or rights from you under 431 | this License. If your rights have been terminated and not permanently 432 | reinstated, you do not qualify to receive new licenses for the same 433 | material under section 10. 434 | 435 | 9. Acceptance Not Required for Having Copies. 436 | 437 | You are not required to accept this License in order to receive or 438 | run a copy of the Program. Ancillary propagation of a covered work 439 | occurring solely as a consequence of using peer-to-peer transmission 440 | to receive a copy likewise does not require acceptance. However, 441 | nothing other than this License grants you permission to propagate or 442 | modify any covered work. These actions infringe copyright if you do 443 | not accept this License. Therefore, by modifying or propagating a 444 | covered work, you indicate your acceptance of this License to do so. 445 | 446 | 10. Automatic Licensing of Downstream Recipients. 447 | 448 | Each time you convey a covered work, the recipient automatically 449 | receives a license from the original licensors, to run, modify and 450 | propagate that work, subject to this License. You are not responsible 451 | for enforcing compliance by third parties with this License. 452 | 453 | An "entity transaction" is a transaction transferring control of an 454 | organization, or substantially all assets of one, or subdividing an 455 | organization, or merging organizations. If propagation of a covered 456 | work results from an entity transaction, each party to that 457 | transaction who receives a copy of the work also receives whatever 458 | licenses to the work the party's predecessor in interest had or could 459 | give under the previous paragraph, plus a right to possession of the 460 | Corresponding Source of the work from the predecessor in interest, if 461 | the predecessor has it or can get it with reasonable efforts. 462 | 463 | You may not impose any further restrictions on the exercise of the 464 | rights granted or affirmed under this License. For example, you may 465 | not impose a license fee, royalty, or other charge for exercise of 466 | rights granted under this License, and you may not initiate litigation 467 | (including a cross-claim or counterclaim in a lawsuit) alleging that 468 | any patent claim is infringed by making, using, selling, offering for 469 | sale, or importing the Program or any portion of it. 470 | 471 | 11. Patents. 472 | 473 | A "contributor" is a copyright holder who authorizes use under this 474 | License of the Program or a work on which the Program is based. The 475 | work thus licensed is called the contributor's "contributor version". 476 | 477 | A contributor's "essential patent claims" are all patent claims 478 | owned or controlled by the contributor, whether already acquired or 479 | hereafter acquired, that would be infringed by some manner, permitted 480 | by this License, of making, using, or selling its contributor version, 481 | but do not include claims that would be infringed only as a 482 | consequence of further modification of the contributor version. For 483 | purposes of this definition, "control" includes the right to grant 484 | patent sublicenses in a manner consistent with the requirements of 485 | this License. 486 | 487 | Each contributor grants you a non-exclusive, worldwide, royalty-free 488 | patent license under the contributor's essential patent claims, to 489 | make, use, sell, offer for sale, import and otherwise run, modify and 490 | propagate the contents of its contributor version. 491 | 492 | In the following three paragraphs, a "patent license" is any express 493 | agreement or commitment, however denominated, not to enforce a patent 494 | (such as an express permission to practice a patent or covenant not to 495 | sue for patent infringement). To "grant" such a patent license to a 496 | party means to make such an agreement or commitment not to enforce a 497 | patent against the party. 498 | 499 | If you convey a covered work, knowingly relying on a patent license, 500 | and the Corresponding Source of the work is not available for anyone 501 | to copy, free of charge and under the terms of this License, through a 502 | publicly available network server or other readily accessible means, 503 | then you must either (1) cause the Corresponding Source to be so 504 | available, or (2) arrange to deprive yourself of the benefit of the 505 | patent license for this particular work, or (3) arrange, in a manner 506 | consistent with the requirements of this License, to extend the patent 507 | license to downstream recipients. "Knowingly relying" means you have 508 | actual knowledge that, but for the patent license, your conveying the 509 | covered work in a country, or your recipient's use of the covered work 510 | in a country, would infringe one or more identifiable patents in that 511 | country that you have reason to believe are valid. 512 | 513 | If, pursuant to or in connection with a single transaction or 514 | arrangement, you convey, or propagate by procuring conveyance of, a 515 | covered work, and grant a patent license to some of the parties 516 | receiving the covered work authorizing them to use, propagate, modify 517 | or convey a specific copy of the covered work, then the patent license 518 | you grant is automatically extended to all recipients of the covered 519 | work and works based on it. 520 | 521 | A patent license is "discriminatory" if it does not include within 522 | the scope of its coverage, prohibits the exercise of, or is 523 | conditioned on the non-exercise of one or more of the rights that are 524 | specifically granted under this License. You may not convey a covered 525 | work if you are a party to an arrangement with a third party that is 526 | in the business of distributing software, under which you make payment 527 | to the third party based on the extent of your activity of conveying 528 | the work, and under which the third party grants, to any of the 529 | parties who would receive the covered work from you, a discriminatory 530 | patent license (a) in connection with copies of the covered work 531 | conveyed by you (or copies made from those copies), or (b) primarily 532 | for and in connection with specific products or compilations that 533 | contain the covered work, unless you entered into that arrangement, 534 | or that patent license was granted, prior to 28 March 2007. 535 | 536 | Nothing in this License shall be construed as excluding or limiting 537 | any implied license or other defenses to infringement that may 538 | otherwise be available to you under applicable patent law. 539 | 540 | 12. No Surrender of Others' Freedom. 541 | 542 | If conditions are imposed on you (whether by court order, agreement or 543 | otherwise) that contradict the conditions of this License, they do not 544 | excuse you from the conditions of this License. If you cannot convey a 545 | covered work so as to satisfy simultaneously your obligations under this 546 | License and any other pertinent obligations, then as a consequence you may 547 | not convey it at all. For example, if you agree to terms that obligate you 548 | to collect a royalty for further conveying from those to whom you convey 549 | the Program, the only way you could satisfy both those terms and this 550 | License would be to refrain entirely from conveying the Program. 551 | 552 | 13. Use with the GNU Affero General Public License. 553 | 554 | Notwithstanding any other provision of this License, you have 555 | permission to link or combine any covered work with a work licensed 556 | under version 3 of the GNU Affero General Public License into a single 557 | combined work, and to convey the resulting work. The terms of this 558 | License will continue to apply to the part which is the covered work, 559 | but the special requirements of the GNU Affero General Public License, 560 | section 13, concerning interaction through a network will apply to the 561 | combination as such. 562 | 563 | 14. Revised Versions of this License. 564 | 565 | The Free Software Foundation may publish revised and/or new versions of 566 | the GNU General Public License from time to time. Such new versions will 567 | be similar in spirit to the present version, but may differ in detail to 568 | address new problems or concerns. 569 | 570 | Each version is given a distinguishing version number. If the 571 | Program specifies that a certain numbered version of the GNU General 572 | Public License "or any later version" applies to it, you have the 573 | option of following the terms and conditions either of that numbered 574 | version or of any later version published by the Free Software 575 | Foundation. If the Program does not specify a version number of the 576 | GNU General Public License, you may choose any version ever published 577 | by the Free Software Foundation. 578 | 579 | If the Program specifies that a proxy can decide which future 580 | versions of the GNU General Public License can be used, that proxy's 581 | public statement of acceptance of a version permanently authorizes you 582 | to choose that version for the Program. 583 | 584 | Later license versions may give you additional or different 585 | permissions. However, no additional obligations are imposed on any 586 | author or copyright holder as a result of your choosing to follow a 587 | later version. 588 | 589 | 15. Disclaimer of Warranty. 590 | 591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT 593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY 594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, 595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM 597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF 598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 599 | 600 | 16. Limitation of Liability. 601 | 602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING 603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS 604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY 605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE 606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF 607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), 609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF 610 | SUCH DAMAGES. 611 | 612 | 17. Interpretation of Sections 15 and 16. 613 | 614 | If the disclaimer of warranty and limitation of liability provided 615 | above cannot be given local legal effect according to their terms, 616 | reviewing courts shall apply local law that most closely approximates 617 | an absolute waiver of all civil liability in connection with the 618 | Program, unless a warranty or assumption of liability accompanies a 619 | copy of the Program in return for a fee. 620 | 621 | END OF TERMS AND CONDITIONS 622 | 623 | How to Apply These Terms to Your New Programs 624 | 625 | If you develop a new program, and you want it to be of the greatest 626 | possible use to the public, the best way to achieve this is to make it 627 | free software which everyone can redistribute and change under these terms. 628 | 629 | To do so, attach the following notices to the program. It is safest 630 | to attach them to the start of each source file to most effectively 631 | state the exclusion of warranty; and each file should have at least 632 | the "copyright" line and a pointer to where the full notice is found. 633 | 634 | 635 | Copyright (C) 636 | 637 | This program is free software: you can redistribute it and/or modify 638 | it under the terms of the GNU General Public License as published by 639 | the Free Software Foundation, either version 3 of the License, or 640 | (at your option) any later version. 641 | 642 | This program is distributed in the hope that it will be useful, 643 | but WITHOUT ANY WARRANTY; without even the implied warranty of 644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 645 | GNU General Public License for more details. 646 | 647 | You should have received a copy of the GNU General Public License 648 | along with this program. If not, see . 649 | 650 | Also add information on how to contact you by electronic and paper mail. 651 | 652 | If the program does terminal interaction, make it output a short 653 | notice like this when it starts in an interactive mode: 654 | 655 | Copyright (C) 656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. 657 | This is free software, and you are welcome to redistribute it 658 | under certain conditions; type `show c' for details. 659 | 660 | The hypothetical commands `show w' and `show c' should show the appropriate 661 | parts of the General Public License. Of course, your program's commands 662 | might be different; for a GUI interface, you would use an "about box". 663 | 664 | You should also get your employer (if you work as a programmer) or school, 665 | if any, to sign a "copyright disclaimer" for the program, if necessary. 666 | For more information on this, and how to apply and follow the GNU GPL, see 667 | . 668 | 669 | The GNU General Public License does not permit incorporating your program 670 | into proprietary programs. If your program is a subroutine library, you 671 | may consider it more useful to permit linking proprietary applications with 672 | the library. If this is what you want to do, use the GNU Lesser General 673 | Public License instead of this License. But first, please read 674 | . 675 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | TemporalKit 2 | === 3 | An all in one solution for adding Temporal Stability to a Stable Diffusion Render via an automatic1111 extension 4 | 5 |

6 | 7 | --- 8 | 9 | ***You must install FFMPEG to path before running this*** 10 | 11 | You can find a demonstration run through with a single batch here:
12 | https://twitter.com/CiaraRowles1/status/1645923461343363072 13 | 14 | And a batch demonstration here:
15 | https://mobile.twitter.com/CiaraRowles1/status/1646458056803250178 16 | 17 | Ebsynth tutorial:
18 | https://twitter.com/CiaraRowles1/status/1648462374125576192
19 | **NOTE: EBSYNTH DOES NOT REGISTER THE KEYFRAMES IF YOU USE ABOVE 20.** 20 | 21 | Ebsynth split frames tutorial:
22 | https://www.youtube.com/watch?v=z3YNHiuvxyg&ab_channel=CiaraRowles 23 | 24 | Example results you can get: 25 | 26 | https://user-images.githubusercontent.com/13116982/234425054-9a1bbf30-93a8-4f5b-9e80-4376ab3c510a.mp4 27 | 28 |

29 | 30 | --- 31 | 32 | The values in the extension are as follows 33 | --- 34 | 35 | | Variable | Description | 36 | |---------------------| --- | 37 | | `FPS` | The fps the video is extracted and produced at. | 38 | | `batch_Size` | This is the number of frames between each keyframe, so for example if you had an fps of 30, and a batch size of 10, it would make 3 keyframes a second and estimate the rest. | 39 | | `per side` | This is the square root of the number of frames per plate, so for example a per side value of 2 would make 4 plates, 3, 9 plates, 4 16 plates. | 40 | | `Resolution` | The size of each plate, it is strongly reccomended you set this to a multiple of your per side variable | 41 | | `batch settings` | Only open this drop down if you want to generate a folder of plates. | 42 | | `Max Frames` | When generating a folder of plates, this gets how many frames at the above fps you want to get, and then divides them into plates in groups of (per side * per side * batch size) | 43 | | `Border Frames` | Every batch generated plate will contain this many frames from the next plate and blend between them. | 44 | | `Batch Folder` | If you're generating a batch of plates, just specify a empty folder and on clicking run, it will populate it with the relevant folders and files, all you need to do is go to img2img batch processing in original sd, enter the newly create input folder as the input, the newly created output folder as the output, generate, move back to the temporal-kit Batch-Warp Tab, put in the whole folder directory and click read and it will set everything up. | 45 | | `Output Resolution` | The maximum resolution on any side of the output video. | 46 | 47 |

48 | 49 | --- 50 | 51 | FAQ: 52 | --- 53 | 54 | > **Q:** My video has smearing. 55 | > 56 | > **A:** Use a higher fps and/or lower batchnumber, the closer together the keyframes the less artifacts. 57 | 58 | > **Q:** Stable diffusion cannot be turned on after installing this Extension. `ModuleNotFoundError: No module named 'tqdm.auto'` 59 | > 60 | > **A:** Because the dependency currently used by this plug-in uses an old version of tqdm, an error occurs. The short-term solution is to manually install a new version(4.66.1) of tqdm.
61 | > ```bash 62 | > pip install tqdm==4.66.1 63 | > ``` 64 | > [More information](https://github.com/CiaraStrawberry/TemporalKit/issues/104#issuecomment-1722527970) 65 | 66 |

67 | 68 | --- 69 | 70 | Step-by-Step Tutorial 71 | --- 72 | Written with reference to this [web page](https://stable-diffusion-art.com/video-to-video/#Method_5_Temporal_Kit) teaching and my [own(cocomine)](https://github.com/cocomine/) experience 73 | 74 | 75 | 76 | ### Step 1: Install Extensions on WebUI 77 | 78 | Open `Extensions` tab > `Install from URL` > Paste follow link in `URL for extension’s git repository` > Click `Install` 79 | 80 | ``` 81 | https://github.com/CiaraStrawberry/TemporalKit 82 | ``` 83 | 84 | ### Step 2: Install FFMPEG 85 | 86 | #### Ubuntu 87 | 88 | ```bash 89 | sudo apt install ffmpeg 90 | ``` 91 | 92 | #### Arch Linux 93 | ```bash 94 | sudo pacman -S ffmpeg 95 | ``` 96 | 97 | #### Windows 98 | Download and install from https://ffmpeg.org/download.html 99 | Make sure to add ffmpeg to your PATH. 100 | Learn more: https://www.wikihow.com/Install-FFmpeg-on-Windows 101 | 102 | ### Step 4: Prepare your video 103 | Create a folder in your desired location. This folder will be used to store the files that need to be processed. 104 | Prepare the video you need to use (referred to as the original video in subsequent teaching), and understand the format of the original video, such as resolution and frame. 105 | 106 | ### Step 5: Extract frames from the original video 107 | 1. Open `Temporal-Kit` Tab on Top. 108 | 2. Open `Pre-Process` Tab. 109 | 2. Drag & Drop the original video into the `Input Video`. 110 | 3. Set `fps` to the frame rate of the original video. 111 | 4. Set `frames per keyframe` to the number of frames between each keyframe. For example, if the original video is 30fps and you set it to 10, then 3 keyframes will be generated per second, and the rest will be estimated. 112 | 5. Set `Side` to the square root of the number of frames per plate. For example, if you set it to 2, 4 plates(2x2) will be generated, 3, 9 plates(3x3), 4, 16 plates(4x4). 113 | 6. Set `Height Resolution` to the size of each plate. It is strongly recommended that you set this to a multiple of your side variable.
For example, if you want to generate 4 plates and set the side to 2, each plate high 512, then you need to set the height resolution to 1024(512x2). 114 | 7. Set `Target Folder` to the folder you created in [step 4](#step-4-prepare-your-video). 115 | 8. Open `Batch Settings` Tab. 116 | 9. Tick the `Batch Run`. 117 | 10. Open `EBsynth` Tab. 118 | 11. Tick the `Split Video` 119 | 120 | When you complete the above steps you should see a structure similar to this in the folder you specified (depending on the length of your video) 121 | 122 | ![folder structure](/readme_img/1.png) 123 | 124 | You will see a structure like this in the video clips divided into folders named by numbers. 125 | 126 | ![folder structure](/readme_img/2.png) 127 | 128 | > If you encounter out of memory issue in the next **img2img** step, reduce the `side` or `Height Resolution` parameters. 129 | 130 | ### Step 6: Perform Img2img on keyframes 131 | Let’s deal with the folder named `0` first.
132 | Go to the **Img2img** page. Switch to the **Batch** tab. Set the following parameters: 133 | 134 | **Input directory**: The name of your [target directory](#step-4-prepare-your-video) with `\input` appended. E.g. `YOUR_FOLDER_PATH_IN_SETP_4\0\input`
135 | **Output directory**: Similarly but with `\output` appended. E.g. `YOUR_FOLDER_PATH_IN_SETP_4\0\output` 136 | 137 | Enter a **prompt** and a **negative prompt** like txt2img.
138 | **Sampling method:** DPM++2M Karras
139 | **Sampling steps:** 20
140 | **CFG scale:** 7
141 | **Denoising strength:** 0.5 (adjust accordingly)
142 | > The above parameters can be changed as needed. 143 | 144 | 145 | ##### Control Net (option, which would give better results if available) 146 | In ControlNet (Unit 0) section, set: 147 | + Enable: Yes 148 | + Pixel Perfect: Yes 149 | + ControlType: Tile 150 | + Preprocessor: tile_resample 151 | + Model: control_xxxx_tile 152 | 153 | Press **Generate**. After it is done, you will find the image in the batch output folder. 154 | 155 | > Make sure to open the image in full size and inspect the details in full size. Make sure they look sharp and have a consistent style. 156 | 157 | > If you want to obtain high-resolution images, please put the output images back into img2img, and adjust resizd by `1.5-2.0`, Denoising strength `0.3-0.4`, and then generate. 158 | 159 | ### Step 7: Prepare EbSynth data 160 | Go to `Temporal-Kit` page and switch to the `Ebsynth-Process` tab. 161 | 162 | **Input Folder:** Put in the same [target folder](#step-4-prepare-your-video) path you put in the Pre-Processing page. E.g. `YOUR_FOLDER_PATH_IN_SETP_4\0` 163 | 164 | Click read last_settings. If your input folder is correct, the video and the settings will be populated. 165 | 166 | Click prepare ebsynth. After it is done, you should see the keys folder populated with your stylized keyframes, and the frames folder populated with your images. 167 | 168 | ![folder structure](/readme_img/3.png) 169 | ![folder structure](/readme_img/4.png) 170 | 171 | > Please note that this program does not generate `.ebs` files. When your images are imported into the program, they will be automatically populated. 172 | 173 | ### Step 8: Process with EbSynth 174 | Now open the **EbSynth** program. 175 | 176 | Open the File Explorer and navigate to the folder your creation in [step4](#step-4-prepare-your-video). You should folder like the ones showed below. We need the **keys** folder and the **frames** folder for EbSynth. 177 | 178 | Drag the **keys** folder from the File Explorer and drop it to the **Keyframes** field in EbSynth.
179 | Drag the **frames** folder from the File Explorer and drop it to the **frames** field in EbSynth. 180 | 181 | ![folder structure](/readme_img/5.png) 182 | 183 | Click **Run All** and wait for them to complete. 184 | When it is done, you should see a series of `out_#####` directories generated in the [target project folder](#step-4-prepare-your-video). 185 | 186 | > Please download the program from the official website.
187 | > https://ebsynth.com/ 188 | 189 | ### Step 9: Generate the final video 190 | Now go back to **AUTOMATIC1111**(webUI). You should still be on the **Temporal Kit** page and **Ebsynth-Process** tab. 191 | 192 | Click **recombine ebsynth** and you are done! 193 | 194 | ![](https://stable-diffusion-art.com/wp-content/uploads/2023/06/temporalkit_ebsynth_ds0.5.gif) 195 | 196 | Look how smooth the video is. With some tweaking, you can probably make it better! 197 | 198 | ### Step 10: Continue processing other split segments 199 | Repeat steps [6](#step-6-perform-img2img-on-keyframes) to [9](#step-9-generate-the-final-video). Process other folders `(1,2,3,4,...)` 200 | 201 | ### Step 11: Combine the split videos 202 | Use a video splicing program to merge segmented videos 203 | 204 |

205 | 206 | --- 207 | 208 | TODO 209 | --- 210 | 211 | - set up diffusion based upscaling for the plates output 212 | - get the img2img button working with batch processing. 213 | - add a check to see if the output folder was added. 214 | - fix that weird shutdown error it gives after running 215 | - hook up to the api. 216 | - flowmaps from game engine export\import support 217 | 218 | _Thanks to RAFT for the optical flow system._ 219 | -------------------------------------------------------------------------------- /install.py: -------------------------------------------------------------------------------- 1 | import launch 2 | 3 | if not launch.is_installed("ffmpeg"): 4 | launch.run_pip("install ffmpeg-python", "Install \"ffmpeg-python\" requirements for TemporalKit extension") 5 | 6 | if not launch.is_installed("moviepy"): 7 | launch.run_pip("install moviepy", "Install \"moviepy\" requirements for TemporalKit extension") 8 | 9 | if not launch.is_installed("imageio_ffmpeg"): 10 | launch.run_pip("install imageio_ffmpeg", "Install \"imageio_ffmpeg\" requirements for TemporalKit extension") 11 | 12 | if not launch.is_installed("scenedetect"): 13 | launch.run_pip("install scenedetect", "Install \"scenedetect\" requirements for TemporalKit extension") 14 | -------------------------------------------------------------------------------- /javascript/temporaljavascript.js: -------------------------------------------------------------------------------- 1 | function switch_to_temporal_kit() { 2 | gradioApp().querySelector('#tabs').querySelectorAll('button')[6].click(); 3 | // gradioApp().getElementById('TemporalKit').querySelectorAll('button')[0].click(); 4 | } 5 | 6 | function switch_to_temporal_kit_final2() { 7 | // Get the current image source 8 | // const gallery = document.getElementById('img2img_gallery'); 9 | // const firstImageSource = gallery.getElementsByTagName('img')[0].src; 10 | //firstImageSource = document.getElementById('svelte-1tkea93').src; 11 | // Switch to the "temporal-kit" tab and the "final" subtab 12 | 13 | switch_to_temporal_kit_final(); 14 | const tabList = document.querySelector("#TemporalKit-Tab"); 15 | 16 | if (tabList) { 17 | const firstChild = tabList.firstElementChild; 18 | 19 | if (firstChild) { 20 | const secondTab = firstChild.querySelector(":nth-child(2)"); 21 | 22 | if (secondTab) { 23 | secondTab.click(); 24 | } else { 25 | console.error("Second tab element not found."); 26 | } 27 | } else { 28 | console.error("First child element not found."); 29 | } 30 | } else { 31 | console.error("Tab list element not found."); 32 | } 33 | document.getElementById("read_last_settings").click(); 34 | document.getElementById("read_last_image").click(); 35 | 36 | // Paste the image source into the "final" subtab's image element 37 | // document.getElementById('output_image').src = firstImageSource; 38 | } 39 | 40 | -------------------------------------------------------------------------------- /readme_img/1.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/CiaraStrawberry/TemporalKit/f46d255cfeb9c4b0c682bae5660a04ac7b7cd8dc/readme_img/1.png -------------------------------------------------------------------------------- /readme_img/2.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/CiaraStrawberry/TemporalKit/f46d255cfeb9c4b0c682bae5660a04ac7b7cd8dc/readme_img/2.png -------------------------------------------------------------------------------- /readme_img/3.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/CiaraStrawberry/TemporalKit/f46d255cfeb9c4b0c682bae5660a04ac7b7cd8dc/readme_img/3.png -------------------------------------------------------------------------------- /readme_img/4.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/CiaraStrawberry/TemporalKit/f46d255cfeb9c4b0c682bae5660a04ac7b7cd8dc/readme_img/4.png -------------------------------------------------------------------------------- /readme_img/5.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/CiaraStrawberry/TemporalKit/f46d255cfeb9c4b0c682bae5660a04ac7b7cd8dc/readme_img/5.png -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- 1 | ffmpeg-python 2 | moviepy 3 | imageio_ffmpeg 4 | scenedetect -------------------------------------------------------------------------------- /scripts/Berry_Method.py: -------------------------------------------------------------------------------- 1 | import cv2 2 | import base64 3 | from PIL import Image 4 | import numpy as np 5 | import math 6 | import scripts.berry_utility as utilityb 7 | import scripts.stable_diffusion_processing as sdprocess 8 | from moviepy.editor import * 9 | import os 10 | import subprocess 11 | import json 12 | import tempfile 13 | import uuid 14 | from torchvision.io import write_jpeg 15 | import re 16 | import gradio as gr 17 | from io import BytesIO 18 | import shutil 19 | 20 | resolution = 1400 21 | smol_resolution = 512 22 | prompt = "cyborg humans photo realistic" 23 | fill_in_denoise = 0 24 | edge_denoise = 0.4 # this is a factor of fill in denoise 25 | initial_denoise = 0.85 26 | frames_limit = 50 27 | seed = 5434536443 28 | diffuse = False 29 | check_edges = False 30 | 31 | def split_into_batches(frames, batch_size, max_batches): 32 | groups = [frames[i:i+batch_size] for i in range(0, len(frames), batch_size)][:max_batches] 33 | 34 | # Add any remaining images to the last group 35 | if len(frames) > max_batches * batch_size: 36 | groups[-1] += frames[max_batches*batch_size:] 37 | 38 | return groups 39 | 40 | def create_square_texture(frames, max_size, side_length=3): 41 | 42 | original_height, original_width = frames[0].shape[:2] 43 | # Calculate the average aspect ratio of the input frames 44 | big_frame_width = original_width * side_length 45 | big_frame_height = original_height * side_length 46 | 47 | texture_aspect_ratio = float(big_frame_width) / float(big_frame_height) 48 | _smol_frame_height = max_size 49 | _smol_frame_width = int(_smol_frame_height * texture_aspect_ratio) 50 | 51 | 52 | actual_texture_width, actual_texture_height = utilityb.resize_to_nearest_multiple(_smol_frame_width, _smol_frame_height, side_length) 53 | 54 | frames_per_row = side_length 55 | frame_width = int (actual_texture_width / side_length) 56 | frame_height = int(actual_texture_height / side_length) 57 | print (f"generating square of width {actual_texture_width} and height {actual_texture_height}") 58 | 59 | texture = np.zeros((actual_texture_height, actual_texture_width, 3), dtype=np.uint8) 60 | 61 | for i, frame in enumerate(frames): 62 | if frame is not None and not frame.size == 0: 63 | resized_frame = cv2.resize(frame, (frame_width, frame_height), interpolation=cv2.INTER_AREA) 64 | row, col = i // frames_per_row, i % frames_per_row 65 | texture[row * frame_height:(row + 1) * frame_height, col * frame_width:(col + 1) * frame_width] = resized_frame 66 | #truth be told i am not entirely sure why this is needed 67 | fixed_texture = cv2.resize(texture, (actual_texture_width, actual_texture_height), interpolation=cv2.INTER_AREA) 68 | 69 | return fixed_texture 70 | 71 | def split_frames_into_big_batches(frames, batch_size, border,ebsynth,returnframe_locations=False): 72 | """ 73 | Splits an array of numpy frames into batches of a given size, adding a certain number of border 74 | frames from the next batch to each batch. 75 | 76 | Parameters: 77 | frames (numpy.ndarray): The input frames to be split. 78 | batch_size (int): The number of frames per batch. 79 | border (int): The number of border frames from the next batch to add to each batch. 80 | 81 | Returns: 82 | List[numpy.ndarray]: A list of batches, each containing `batch_size` + `border` frames (except for the last batch). 83 | """ 84 | num_frames = len(frames) 85 | num_batches = int(np.ceil(num_frames / batch_size)) 86 | print(f"frames num = {len(frames)} while num batches = {num_batches}") 87 | batches = [] 88 | 89 | frame_locations = [] 90 | for i in range(num_batches): 91 | start_idx = i * batch_size 92 | end_idx = start_idx + batch_size 93 | if ebsynth == False: 94 | # Add border frames if not the last batch and if available 95 | if i < num_batches - 1: 96 | end_idx += min(border, num_frames - end_idx) 97 | else: 98 | # Combine the last batch with the previous batch if the number of frames in the last batch is smaller than the border size 99 | if end_idx - start_idx < border and len(batches) > 0: 100 | batches[-1] = np.concatenate((batches[-1], frames[start_idx:end_idx])) 101 | break 102 | else: 103 | if i < num_batches - 1: 104 | end_idx = end_idx + border 105 | 106 | end_idx = min(end_idx, num_frames) 107 | batches.append(frames[start_idx:end_idx]) 108 | print (f"batch {i} has {len(batches[i])} frames") 109 | frame_locations.append((start_idx,end_idx)) 110 | 111 | if returnframe_locations == False: 112 | return batches 113 | else: 114 | return batches,frame_locations 115 | 116 | def split_square_texture(texture, num_frames,max_frames, _smol_resolution,ebsynth=False): 117 | 118 | texture_height, texture_width = texture.shape[:2] 119 | texture_aspect_ratio = float(texture_width) / float(texture_height) 120 | 121 | frames_per_row = int(math.ceil(math.sqrt(max_frames))) 122 | frame_height = int (texture_height / frames_per_row) 123 | frame_width = int(texture_width / frames_per_row) 124 | 125 | _smol_frame_height = _smol_resolution 126 | _smol_frame_width = int(_smol_frame_height * texture_aspect_ratio) 127 | 128 | if ebsynth == False: 129 | _smol_frame_resized_width, _smol_frame_resized_height = utilityb.resize_to_nearest_multiple_of_8(_smol_frame_width, _smol_frame_height) 130 | else: 131 | _smol_frame_resized_width, _smol_frame_resized_height = _smol_frame_width, _smol_frame_height 132 | #_smol_frame_resized_width, _smol_frame_resized_height = _smol_frame_width, _smol_frame_height 133 | frames = [] 134 | 135 | for i in range(num_frames): 136 | row, col = i // frames_per_row, i % frames_per_row 137 | frame = texture[row * frame_height:(row + 1) * frame_height, col * frame_width:(col + 1) * frame_width] 138 | 139 | if not frame.size == 0: 140 | resized_frame = cv2.resize(frame, (_smol_frame_resized_width, _smol_frame_resized_height), interpolation=cv2.INTER_AREA) 141 | frames.append(resized_frame) 142 | else: 143 | print("frame size 0") 144 | frames.append(np.zeros((_smol_frame_resized_width, _smol_frame_resized_height, 3), dtype=np.uint8)) 145 | 146 | return frames 147 | 148 | def save_square_texture(texture, file_path): 149 | # Check if the input has the correct data type and convert if necessary 150 | if texture.dtype != np.uint8: 151 | texture = (texture * 255).astype(np.uint8) 152 | 153 | # Check if the input has the intended shape (3 channels for an RGB image) 154 | if texture.ndim != 3 or texture.shape[2] != 3: 155 | raise ValueError("Invalid texture shape. Expected a 3-channel RGB image.") 156 | 157 | # Convert the NumPy array to a PIL Image 158 | image = Image.fromarray(texture) 159 | 160 | # Save the image to the specified file path 161 | print(f'saved to {file_path} at size {image.size}') 162 | image.save(file_path, format="PNG") 163 | 164 | 165 | def convert_video_to_bytes(input_file): 166 | # Read the uploaded video file 167 | print(f"reading video file... {input_file}") 168 | with open(input_file, "rb") as f: 169 | video_bytes = f.read() 170 | 171 | # Return the processed video bytes (or any other output you want) 172 | return video_bytes 173 | 174 | 175 | 176 | def generate_square_from_video(video_path, fps, batch_size,resolution,size_size): 177 | video_data = convert_video_to_bytes(video_path) 178 | frames_limit = (size_size * size_size) * batch_size 179 | frames = utilityb.extract_frames_movpie(video_data, fps, frames_limit) 180 | print(len(frames)) 181 | number_of_batches = size_size * size_size 182 | batches = split_into_batches(frames, batch_size,number_of_batches) 183 | print("Number of batches:", len(batches)) 184 | first_frames = [batch[0] for batch in batches] 185 | 186 | square_texture = create_square_texture(first_frames, resolution,side_length=size_size) 187 | #save_square_texture(square_texture, "./result/original.png") 188 | 189 | return square_texture 190 | 191 | def generate_squares_to_folder (video_path, fps, batch_size,resolution,size_size,max_frames,output_folder,border,ebsynth_mode,max_frames_to_save): 192 | fps = int(fps) 193 | if ebsynth_mode == False: 194 | if border >= (batch_size * size_size * size_size) / 2: 195 | raise Exception("too many border frames, reduce border or increase batch size") 196 | 197 | 198 | input_folder_loc = os.path.join(output_folder, "input") 199 | output_folder_loc = os.path.join(output_folder, "output") 200 | debug_result = os.path.join(output_folder, "result") 201 | if not os.path.exists(output_folder): 202 | os.makedirs(output_folder) 203 | if not os.path.exists(input_folder_loc): 204 | os.makedirs(input_folder_loc) 205 | if not os.path.exists(output_folder_loc): 206 | os.makedirs(output_folder_loc) 207 | if not os.path.exists(debug_result): 208 | os.makedirs(debug_result) 209 | frames_loc = os.path.join(output_folder, "frames") 210 | keys_loc = os.path.join(output_folder, "keys") 211 | 212 | if ebsynth_mode == True: 213 | if not os.path.exists(frames_loc): 214 | os.makedirs(frames_loc) 215 | if not os.path.exists(keys_loc): 216 | os.makedirs(keys_loc) 217 | 218 | video_data = convert_video_to_bytes(video_path) 219 | per_batch_limmit = ((size_size * size_size) * batch_size) + border 220 | #if ebsynth_mode == False: 221 | # per_batch_limmit = per_batch_limmit + border 222 | frames = utilityb.extract_frames_movpie(video_data, fps, max_frames,False) 223 | 224 | bigbatches = split_frames_into_big_batches(frames, per_batch_limmit,border,ebsynth=ebsynth_mode) 225 | square_textures = [] 226 | height = 0 227 | width = 0 228 | for i in range(len(bigbatches)): 229 | batches = split_into_batches(bigbatches[i], batch_size, size_size * size_size) 230 | print("Number of batches:", len(batches)) 231 | if ebsynth_mode == False: 232 | keyframes = [batch[0] for batch in batches] 233 | else: 234 | keyframes = [batch[int(len(batch)/2)] for batch in batches] 235 | #for batch in batches: 236 | #print (f"framenum = {int(len(batch)/2)} out of batch length {len(batch)} and size {len(frames)}") 237 | square_texture = create_square_texture(keyframes, resolution,side_length=size_size) 238 | save_square_texture(square_texture, os.path.join(input_folder_loc, f"input{i}.png")) 239 | square_textures.append(square_texture) 240 | height = square_texture.shape[0] 241 | width = square_texture.shape[1] 242 | 243 | batch_settings_loc = os.path.join(output_folder, "batch_settings.txt") 244 | with open(batch_settings_loc, "w") as f: 245 | f.write(str(fps) + "\n") 246 | f.write(str(size_size) + "\n") 247 | f.write(str(batch_size) + "\n") 248 | f.write(str(video_path) + "\n") 249 | f.write(str(max_frames_to_save) + "\n") 250 | f.write(str(border) + "\n") 251 | #return list of urls 252 | 253 | 254 | 255 | return square_textures 256 | 257 | 258 | 259 | def merge_image_batches(image_batches, border): 260 | merged_batches = [] 261 | height, width = image_batches[0][0].shape[:2] 262 | 263 | for i in range(len(image_batches) - 1): 264 | current_batch = image_batches[i] 265 | next_batch = image_batches[i + 1] 266 | for i in range(len(current_batch)): 267 | current_batch[i] = cv2.resize(current_batch[i], (width, height)) 268 | for i in range(len(next_batch)): 269 | next_batch[i] = cv2.resize(next_batch[i], (width, height)) 270 | 271 | # If it's not the first batch, remove the blended images from the current batch 272 | if i > 0: 273 | current_batch = current_batch[border:] 274 | 275 | # Copy all images except the border ones from the current batch 276 | for j in range(len(current_batch) - border): 277 | merged_batches.append(current_batch[j]) 278 | 279 | # Blend the border images between the current and next batch 280 | for j in range(border): 281 | try: 282 | alpha = float(j) / float(border) 283 | blended_image = cv2.addWeighted(current_batch[len(current_batch) - border + j], 1 - alpha, next_batch[j], alpha, 0) 284 | merged_batches.append(blended_image) 285 | except IndexError: 286 | print ("merge failed") 287 | 288 | # Add remaining images from the last batch 289 | merged_batches.extend(image_batches[-1][border:]) 290 | 291 | return merged_batches 292 | 293 | def process_video_batch (video_path_old, fps, per_side, batch_size, fillindenoise, edgedenoise, _smol_resolution,square_textures,max_frames,output_folder,border): 294 | video_path = os.path.join (output_folder, "input_video.mp4") 295 | per_batch_limmit = (((per_side * per_side) * batch_size)) + border 296 | video_data = convert_video_to_bytes(video_path) 297 | frames = utilityb.extract_frames_movpie(video_data, fps, max_frames) 298 | print(f"splitting into batches with per_batch_limmit = {per_batch_limmit} and border {border}" ) 299 | bigbatches = split_frames_into_big_batches(frames, per_batch_limmit,border,ebsynth=False) 300 | bigprocessedbatches = [] 301 | for i , batch in enumerate(bigbatches): 302 | if i < len(square_textures): 303 | new_batch = process_video(batch, per_side, batch_size, fillindenoise, edgedenoise, _smol_resolution,square_textures[i]) 304 | bigprocessedbatches.append(new_batch) 305 | for a, image in enumerate(new_batch): 306 | Image.fromarray(image).save(os.path.join(output_folder, f"result/output{a + (len(new_batch) * i)}.png")) 307 | 308 | just_frame_groups = [] 309 | print (f"bigprocessedbatches len = {len(bigprocessedbatches)}") 310 | for i in range(len(bigprocessedbatches)): 311 | newgroup = [] 312 | for b in range(len(bigprocessedbatches[i])): 313 | newgroup.append(bigprocessedbatches[i][b]) 314 | just_frame_groups.append(newgroup) 315 | 316 | combined = merge_image_batches(just_frame_groups, border) 317 | 318 | save_loc = os.path.join(output_folder, "blended.mp4") 319 | generated_vid = utilityb.pil_images_to_video(combined,save_loc, fps) 320 | return generated_vid 321 | 322 | 323 | def process_video_single(video_path, fps, per_side, batch_size, fillindenoise, edgedenoise, _smol_resolution,square_texture): 324 | 325 | extension_path = os.path.abspath(__file__) 326 | extension_dir = os.path.dirname(os.path.dirname(extension_path)) 327 | output_folder = os.path.join(extension_dir, "result") 328 | extension_path = os.path.abspath(__file__) 329 | frames_limit = (per_side * per_side) * batch_size 330 | extension_dir = os.path.dirname(os.path.dirname(extension_path)) 331 | extension_save_folder = os.path.join(extension_dir, "result") 332 | if not os.path.exists(extension_save_folder): 333 | os.makedirs(extension_save_folder) 334 | utilityb.delete_folder_contents(extension_save_folder) 335 | #rerun the generatesquarefromvideo function to get the unaltered square texture 336 | video_data = convert_video_to_bytes(video_path) 337 | frames = utilityb.extract_frames_movpie(video_data, fps, frames_limit) 338 | processed_frames = process_video(frames,per_side,batch_size,fillindenoise,edgedenoise,_smol_resolution,square_texture) 339 | output_video_path = os.path.join(output_folder, "output.mp4") 340 | generated_video = utilityb.pil_images_to_video(processed_frames, output_video_path, fps) 341 | return generated_video 342 | 343 | 344 | def process_video(frames, per_side, batch_size, fillindenoise, edgedenoise, _smol_resolution,square_texture): 345 | 346 | frame_count = 0 347 | print(len(frames)) 348 | batches = split_into_batches(frames, batch_size, per_side * per_side) 349 | print("Number of batches:", len(batches)) 350 | first_frames = [batch[0] for batch in batches] 351 | #actuallyprocessthevideo 352 | debug = False 353 | 354 | frame_count = 0 355 | global resolution 356 | print(len(frames)) 357 | 358 | 359 | if debug is False: 360 | #result_texture = sdprocess.square_Image_request(encoded_square_texture, prompt, initial_denoise, resolution, seed) 361 | result_texture = square_texture 362 | #save_square_texture(encoded_returns, "./result/processed.png") 363 | 364 | else: 365 | f = open("./result/processed.png", "rb") 366 | bytes = f.read() 367 | result_texture = base64.b64encode(bytes).decode("utf-8") 368 | resolution_get = Image.open("./result/processed.png") 369 | resolution= resolution_get.height 370 | # this is stupid and inefficiant i dont care 371 | #its not really encoded anymore is it 372 | encoded_returns = result_texture 373 | #encoded_returns = cv2.cvtColor(utilityb.base64_to_texture(result_texture), cv2.COLOR_BGR2RGB) 374 | 375 | new_frames = split_square_texture(encoded_returns,len(first_frames), per_side * per_side,_smol_resolution,False) 376 | 377 | if check_edges: 378 | for i, image in enumerate(new_frames): 379 | image = utilityb.check_edges(image) 380 | #:( 381 | 382 | # Save first frames 383 | #for idx, first_frame in enumerate(first_frames): 384 | # save_square_texture(first_frame, os.path.join(output_folder, f"first_frame_{idx}.png")) 385 | # save_square_texture(new_frames[idx], os.path.join(output_folder, f"first_frame_processed_{idx}.png")) 386 | """ 387 | Turns out merging each frame backwards and forwards doesn't actually work, you'd think it did because each frame is conceptually closer to it's origin, but it breaks the flowmap in all sorts of weird ways if you tell it to go backwards, very very annoying 388 | last_processed = None 389 | for i, batch in enumerate(batches): 390 | 391 | encoded_batch = [] 392 | for b, image in enumerate(batch): 393 | encoded_batch.append(utilityb.texture_to_base64(image)) 394 | encoded_new_frame = utilityb.texture_to_base64(new_frames[i]) 395 | 396 | processed_batch_before,all_flow_before = sdprocess.batch_sd_run(encoded_batch, encoded_new_frame, frame_count, seed, False, fillindenoise, edgedenoise, _smol_resolution,False,encoded_new_frame,False) 397 | 398 | if i < len(batches) - 1: 399 | 400 | encoded_batch_next = [] 401 | for b, image in enumerate(batches[i+1]): 402 | encoded_batch_next.append(utilityb.texture_to_base64(image)) 403 | encoded_ext_frame = utilityb.texture_to_base64(new_frames[i + 1]) 404 | encoded_batch.insert(0,utilityb.texture_to_base64(first_frames[i])) 405 | processed_batch_after,all_flow_after = sdprocess.batch_sd_run(encoded_batch_next, encoded_new_frame, frame_count, seed, True, fillindenoise, edgedenoise, _smol_resolution,True,encoded_new_frame,False) 406 | processed_batch_after.append(encoded_ext_frame) 407 | 408 | if last_processed is not None: 409 | print (len(last_processed)) 410 | print (len(processed_batch_before)) 411 | blended_frames = blend_batches(last_processed, processed_batch_before, resolution=_smol_resolution) 412 | for b, blended_frame in enumerate(blended_frames): 413 | savepath = os.path.join(output_folder, f"frame_{frame_count + b}.png") 414 | #save_square_texture(blended_frame, savepath) 415 | save_square_texture(cv2.cvtColor(utilityb.base64_to_texture(processed_batch_before[b]), cv2.COLOR_BGR2RGB), savepath) 416 | 417 | 418 | 419 | #save_square_texture(cv2.cvtColor(utilityb.base64_to_texture(last_processed[b]), cv2.COLOR_BGR2RGB), f"./debug/before_{frame_count + b}.png") 420 | #save_square_texture(cv2.cvtColor(utilityb.base64_to_texture(processed_batch_before[b]), cv2.COLOR_BGR2RGB), f"./debug/after_{frame_count + b}.png") 421 | #for c, flow in enumerate(all_flow_before): 422 | #write_jpeg(flow, f"./debug/before_flow_{frame_count + c + 1}.png") 423 | else: 424 | for b, frame in enumerate(processed_batch_before): 425 | savepath = os.path.join(output_folder, f"frame_{frame_count + b}.png") 426 | save_square_texture(cv2.cvtColor(utilityb.base64_to_texture(frame), cv2.COLOR_BGR2RGB), savepath) 427 | 428 | 429 | frame_count += len(batch) 430 | last_processed = processed_batch_after 431 | 432 | 433 | 434 | 435 | """ 436 | 437 | output_pil_images = [] 438 | last_processed = None 439 | for i, batch in enumerate(batches): 440 | 441 | encoded_new_frame = utilityb.texture_to_base64(new_frames[i]) 442 | 443 | 444 | encoded_batch = [] 445 | for b, image in enumerate(batches[i]): 446 | encoded_batch.append(utilityb.texture_to_base64(image)) 447 | 448 | processed_batch,all_flow_after = sdprocess.batch_sd_run(encoded_batch, encoded_new_frame, frame_count, seed, False, fillindenoise, edgedenoise, _smol_resolution,True,encoded_new_frame,False) 449 | print (f"number {i} processed batch length {len(processed_batch)} and batch length {len(batch)} and num batches {len(batches)}") 450 | if last_processed is not None: 451 | encoded_batch.insert(0,utilityb.texture_to_base64(batches[i - 1][-1])) 452 | processed_batch_from_before,all_flow_after = sdprocess.batch_sd_run(encoded_batch, last_processed, frame_count, seed, True, fillindenoise, edgedenoise, _smol_resolution,True,last_processed,False) 453 | if not min(len (processed_batch_from_before),len(processed_batch)) > 1: 454 | output_pil_images.append(cv2.cvtColor(utilityb.base64_to_texture(processed_batch[0]), cv2.COLOR_BGR2RGB)) 455 | continue 456 | blended_frames = blend_batches(processed_batch_from_before, processed_batch, resolution=_smol_resolution) 457 | print (f"blended frames {len(blended_frames)}") 458 | for b, blended_frame in enumerate(blended_frames): 459 | #savepath = os.path.join(output_folder, f"frame_{frame_count + b}.png") 460 | #save_square_texture(blended_frame, savepath) 461 | output_pil_images.append(blended_frame) 462 | else: 463 | for b, frame in enumerate(processed_batch): 464 | #savepath = os.path.join(output_folder, f"frame_{frame_count + b}.png") 465 | #save_square_texture(cv2.cvtColor(utilityb.base64_to_texture(frame), cv2.COLOR_BGR2RGB), savepath) 466 | output_pil_images.append(cv2.cvtColor(utilityb.base64_to_texture(frame), cv2.COLOR_BGR2RGB)) 467 | 468 | 469 | frame_count += len(batch) 470 | last_processed = processed_batch[-1] 471 | print (f"output pil images {len(output_pil_images)}") 472 | return output_pil_images 473 | 474 | 475 | def image_folder_to_video(folder_path, output_file, fps=24): 476 | """ 477 | Turns a folder of images into a video using MoviePy. 478 | 479 | :param folder_path: str, path to the folder containing the images 480 | :param output_file: str, path to the output video file (e.g., 'output.mp4') 481 | :param fps: int, frames per second (default: 24) 482 | """ 483 | # Get a list of image file names 484 | image_files = [f for f in os.listdir(folder_path) if f.lower().endswith(('.png', '.jpg', '.jpeg', '.tiff', '.bmp', '.gif'))] 485 | 486 | # Sort image files by their names 487 | image_files.sort(key=lambda x: int(re.search(r'\d+', x).group())) 488 | 489 | # Create a list of full image file paths 490 | image_paths = [os.path.join(folder_path, image) for image in image_files] 491 | 492 | # Create a clip from the image sequence 493 | clip = ImageSequenceClip(image_paths, fps=fps) 494 | 495 | # Write the clip to a video file 496 | clip.write_videofile(output_file, codec='libx264') 497 | 498 | return output_file 499 | 500 | 501 | 502 | 503 | 504 | def blend_batches(batch_before, current_batch,resolution, blend_start_ratio=0.9, blend_end_ratio=0.1): 505 | blended_frames = [] 506 | num_frames = min(len (batch_before),len(current_batch)) 507 | decoded_batch_before = [cv2.cvtColor(utilityb.base64_to_texture(frame), cv2.COLOR_BGR2RGB) for frame in batch_before] 508 | decoded_current_batch = [cv2.cvtColor(utilityb.base64_to_texture(frame), cv2.COLOR_BGR2RGB) for frame in current_batch] 509 | 510 | #target_width, target_height = resolution, resolution 511 | height, width = decoded_batch_before[0].shape[:2] 512 | # Resize the images in decoded_batch_before and decoded_current_batch 513 | decoded_batch_before = [cv2.resize(img, (width,height)) for img in decoded_batch_before] 514 | decoded_current_batch = [cv2.resize(img, (width,height)) for img in decoded_current_batch] 515 | 516 | output_folder = "moretemp" 517 | if not os.path.exists(output_folder): 518 | os.makedirs(output_folder) 519 | 520 | if not num_frames > 1: 521 | return [current_batch[0]] 522 | for i in range(num_frames): 523 | alpha = blend_start_ratio - (i / (num_frames - 1)) * (blend_start_ratio - blend_end_ratio) 524 | blended_frame = cv2.addWeighted(decoded_batch_before[i], alpha, decoded_current_batch[i], 1 - alpha, 0) 525 | print(f"blended frame {i}") 526 | 527 | # Concatenate the two input frames and the blended frame horizontally 528 | concatenated_frame = cv2.hconcat([decoded_batch_before[i], decoded_current_batch[i], blended_frame]) 529 | 530 | cv2.imwrite(os.path.join(output_folder, f"concatenated_{i}.png"), concatenated_frame) 531 | 532 | blended_frames.append(blended_frame) 533 | return blended_frames 534 | 535 | 536 | 537 | def interpolate_frames(frame1, frame2, alpha): 538 | gray1 = cv2.cvtColor(frame1, cv2.COLOR_BGR2GRAY) 539 | gray2 = cv2.cvtColor(frame2, cv2.COLOR_BGR2GRAY) 540 | 541 | flow = cv2.calcOpticalFlowFarneback(gray1, gray2, None, pyr_scale=0.5, levels=3, winsize=15, iterations=3, poly_n=5, poly_sigma=1.2, flags=0) 542 | h, w = flow.shape[:2] 543 | 544 | flow_map = -alpha * flow + np.indices((h, w)).transpose(1, 2, 0) 545 | flow_map = flow_map.astype(np.float32) # Convert flow_map to float32 data type 546 | return cv2.remap(frame1, flow_map, None, cv2.INTER_LINEAR) 547 | 548 | def interpolate_video(input_path, output_path, output_fps): 549 | clip = VideoFileClip(input_path) 550 | input_fps = clip.fps 551 | 552 | frames = [frame for frame in clip.iter_frames()] 553 | new_frames = [] 554 | 555 | if output_fps <= input_fps: 556 | raise ValueError("Output fps should be greater than input fps") 557 | 558 | frame_ratio = input_fps / output_fps 559 | 560 | for i in range(len(frames) - 1): 561 | new_frames.append(frames[i]) 562 | print(f"interpolating for frame {i}") 563 | extra_frames = int(round((i + 1) / frame_ratio) - round(i / frame_ratio)) 564 | for j in range(1, extra_frames + 1): 565 | 566 | alpha = j / (extra_frames + 1) 567 | frame1 = frames[i] # Transpose the dimensions of frame1 (HxWxC to WxHxC) 568 | frame2 = frames[i + 1] # Transpose the dimensions of frame2 (HxWxC to WxHxC) 569 | interpolated_frame = interpolate_frames(frame1, frame2, alpha) 570 | interpolated_frame = interpolated_frame.transpose(1, 0, 2) # Transpose back the dimensions of interpolated_frame (WxHxC to HxWxC) 571 | new_frames.append(interpolated_frame) 572 | 573 | new_frames.append(frames[-1]) 574 | 575 | new_clip = ImageSequenceClip(new_frames, fps=output_fps) 576 | new_clip.write_videofile(output_path) 577 | return output_path 578 | 579 | 580 | def split_videos_into_smaller_videos(max_keys,video,fps,max_frames,target_path,border_number, scenecuts = False): 581 | max_total_frames = int((max_keys / 20) * max_frames) 582 | split_frames,border_indices = divideFrames(video, max_frames,border_number) 583 | split_frames_trimmed,trimmed_borders = trim_images(split_frames,max_total_frames,border_indices ) 584 | print(f" trim_imagestransitions {border_indices}") 585 | output_files = [] 586 | print(f"frames_total_size = {len(split_frames_trimmed)}, frames batch size = {max_frames} array length = {len(split_frames_trimmed)}") 587 | 588 | for i,frames in enumerate(split_frames_trimmed): 589 | print (f"splitting video {i}") 590 | new_folder_location = os.path.join(target_path, f"{i}") 591 | if not os.path.exists(new_folder_location): 592 | os.makedirs(new_folder_location) 593 | new_video_loc = os.path.join(new_folder_location, f"input_video.mp4") 594 | output_files.append(utilityb.pil_images_to_video(frames, new_video_loc, fps)) 595 | return output_files,trimmed_borders 596 | 597 | 598 | def divideFrames(frame_groups, x, y): 599 | result = [] 600 | transitions = [] 601 | 602 | for index, group in enumerate(frame_groups): 603 | print (f"frame_groups {len(group)}") 604 | start = 0 605 | while start < len(group): 606 | end = start + x 607 | new_group = group[start:end] 608 | 609 | if end + y <= len(group): 610 | overlap_group = group[end:end+y] 611 | 612 | 613 | # Concatenate the images from new_group and overlap_group 614 | if y > 0: 615 | combined_group = np.concatenate((new_group, overlap_group), axis=0) 616 | else: 617 | combined_group = new_group 618 | print (f"overlap group size {len(overlap_group)}") 619 | transitions.append(len(result)) 620 | 621 | else: 622 | combined_group = new_group 623 | 624 | result.append(combined_group) 625 | start += x 626 | 627 | return result,transitions 628 | 629 | 630 | def trim_images(images_list_of_lists, max_images, border_indices): 631 | """ 632 | Trims the given list of lists of image arrays so that the total number of image arrays is below the specified maximum. 633 | Removes whole image arrays from the end of the list of lists if the max_images doesn't include them. 634 | 635 | Parameters: 636 | images_list_of_lists (list): List of lists of NumPy image arrays 637 | max_images (int): Maximum number of image arrays allowed 638 | 639 | Returns: 640 | list: List of lists of trimmed image arrays 641 | """ 642 | total_images = sum([len(img_list) for img_list in images_list_of_lists]) 643 | 644 | while total_images > max_images: 645 | print(f"total_images = {total_images}, max_images = {max_images}") 646 | last_list_idx = len(images_list_of_lists) - 1 647 | last_img_idx = len(images_list_of_lists[last_list_idx]) - 1 648 | 649 | if last_img_idx >= 0: 650 | total_images -= 1 651 | images_list_of_lists[last_list_idx] = images_list_of_lists[last_list_idx][:-1] 652 | 653 | if len(images_list_of_lists[last_list_idx]) == 0: 654 | images_list_of_lists.pop() 655 | if last_list_idx in border_indices: 656 | border_indices.pop() 657 | 658 | return images_list_of_lists, border_indices 659 | -------------------------------------------------------------------------------- /scripts/Ebsynth_Processing.py: -------------------------------------------------------------------------------- 1 | import os 2 | import glob 3 | #om nom nom nom 4 | import requests 5 | import json 6 | from pprint import pprint 7 | import base64 8 | import numpy as np 9 | from io import BytesIO 10 | import extensions.TemporalKit.scripts.berry_utility 11 | import scripts.optical_flow_simple as opflow 12 | from PIL import Image, ImageOps,ImageFilter 13 | import io 14 | from collections import deque 15 | import cv2 16 | import scripts.Berry_Method as bmethod 17 | import scripts.berry_utility as butility 18 | import re 19 | 20 | 21 | 22 | def sort_into_folders(video_path, fps, per_side, batch_size, _smol_resolution,square_textures,max_frames,output_folder,border): 23 | border = 0 24 | per_batch_limmit = (((per_side * per_side) * batch_size)) + border 25 | 26 | frames = [] 27 | # original_frames_directory = os.path.join(output_folder, "original_frames") 28 | # if os.path.exists(original_frames_directory): 29 | # for filename in os.listdir(original_frames_directory): 30 | # frames.append(cv2.imread(os.path.join(original_frames_directory, filename), cv2.COLOR_BGR2RGB)) 31 | # else: 32 | video_data = bmethod.convert_video_to_bytes(video_path) 33 | frames = butility.extract_frames_movpie(video_data, fps, max_frames) 34 | 35 | print(f"full frames num = {len(frames)}") 36 | 37 | 38 | output_frames_folder = os.path.join(output_folder, "frames") 39 | if not os.path.exists(output_frames_folder): 40 | os.makedirs(output_frames_folder) 41 | output_keys_folder = os.path.join(output_folder, "keys") 42 | if not os.path.exists(output_keys_folder): 43 | os.makedirs(output_keys_folder) 44 | input_folder = os.path.join(output_folder, "input") 45 | 46 | filenames = os.listdir(input_folder) 47 | img = Image.open(os.path.join(input_folder, filenames[0])) 48 | original_width, original_height = img.size 49 | height,width = frames[0].shape[:2] 50 | 51 | texture_aspect_ratio = float(width) / float(height) 52 | 53 | 54 | _smol_frame_height = _smol_resolution 55 | _smol_frame_width = int(_smol_frame_height * texture_aspect_ratio) 56 | print(f"saving size = {_smol_frame_width}x{_smol_frame_height}") 57 | 58 | 59 | for i, frame in enumerate(frames): 60 | frame_to_save = cv2.resize(frame, (_smol_frame_width, _smol_frame_height), interpolation=cv2.INTER_LINEAR) 61 | bmethod.save_square_texture(frame_to_save, os.path.join(output_frames_folder, "frames{:05d}.png".format(i))) 62 | original_frame_height,original_frame_width = frames[0].shape[:2] 63 | 64 | 65 | 66 | bigbatches,frameLocs = bmethod.split_frames_into_big_batches(frames, per_batch_limmit,border,ebsynth=True,returnframe_locations=True) 67 | bigprocessedbatches = [] 68 | 69 | last_frame_end = 0 70 | print (len(square_textures)) 71 | for a,bigbatch in enumerate(bigbatches): 72 | batches = bmethod.split_into_batches(bigbatches[a], batch_size,per_side* per_side) 73 | 74 | keyframes = [batch[int(len(batch)/2)] for batch in batches] 75 | if a < len(square_textures): 76 | resized_square_texture = cv2.resize(square_textures[a], (original_width, original_height), interpolation=cv2.INTER_LINEAR) 77 | new_frames = bmethod.split_square_texture(resized_square_texture,len(keyframes), per_side* per_side,_smol_resolution,True) 78 | new_frame_start,new_frame_end = frameLocs[a] 79 | 80 | for b in range(len(new_frames)): 81 | print (new_frame_start) 82 | inner_start = last_frame_end 83 | inner_end = inner_start + len(batches[b]) 84 | last_frame_end = inner_end 85 | frame_position = inner_start + int((inner_end - inner_start)/2) 86 | print (f"saving at frame {frame_position}") 87 | frame_to_save = cv2.resize(new_frames[b], (_smol_frame_width, _smol_frame_height), interpolation=cv2.INTER_LINEAR) 88 | bmethod.save_square_texture(frame_to_save, os.path.join(output_keys_folder, "keys{:05d}.png".format(frame_position))) 89 | 90 | just_frame_groups = [] 91 | for i in range(len(bigprocessedbatches)): 92 | newgroup = [] 93 | for b in range(len(bigprocessedbatches[i])): 94 | newgroup.append(bigprocessedbatches[i][b]) 95 | just_frame_groups.append(newgroup) 96 | 97 | return 98 | 99 | 100 | def recombine (video_path, fps, per_side, batch_size, fillindenoise, edgedenoise, _smol_resolution,square_textures,max_frames,output_folder,border): 101 | just_frame_groups = [] 102 | per_batch_limmit = (((per_side * per_side) * batch_size)) - border 103 | video_data = bmethod.convert_video_to_bytes(video_path) 104 | frames = bmethod.extract_frames_movpie(video_data, fps, max_frames) 105 | bigbatches,frameLocs = bmethod.split_frames_into_big_batches(frames, per_batch_limmit,border,returnframe_locations=True) 106 | bigprocessedbatches = [] 107 | for i in range(len(bigprocessedbatches)): 108 | newgroup = [] 109 | for b in range(len(bigprocessedbatches[i])): 110 | newgroup.append(bigprocessedbatches[i][b]) 111 | just_frame_groups.append(newgroup) 112 | 113 | combined = bmethod.merge_image_batches(just_frame_groups, border) 114 | 115 | save_loc = os.path.join(output_folder, "non_blended.mp4") 116 | generated_vid = extensions.TemporalKit.scripts.berry_utility.pil_images_to_video(combined,save_loc, fps) 117 | 118 | 119 | 120 | 121 | def crossfade_folder_of_folders(output_folder, fps,return_generated_video_path=False): 122 | """Crossfade between images in a folder of folders and save the results.""" 123 | root_folder = output_folder 124 | all_dirs = [d for d in os.listdir(root_folder) if os.path.isdir(os.path.join(root_folder, d))] 125 | dirs = [d for d in all_dirs if d.startswith("out_")] 126 | 127 | dirs.sort() 128 | 129 | output_images = [] 130 | allkeynums = getkeynums(os.path.join(root_folder, "keys")) 131 | print(allkeynums) 132 | 133 | for b in range(allkeynums[0]): 134 | current_dir = os.path.join(root_folder, dirs[0]) 135 | images_current = sorted(os.listdir(current_dir)) 136 | image1_path = os.path.join(current_dir, images_current[b]) 137 | image1 = Image.open(image1_path) 138 | output_images.append(np.array(image1)) 139 | 140 | for i in range(len(dirs) - 1): 141 | current_dir = os.path.join(root_folder, dirs[i]) 142 | next_dir = os.path.join(root_folder, dirs[i + 1]) 143 | 144 | images_current = sorted(os.listdir(current_dir)) 145 | images_next = sorted(os.listdir(next_dir)) 146 | 147 | startnum = get_num_at_index(current_dir,0) 148 | bigkeynum = allkeynums[i] 149 | keynum = bigkeynum - startnum 150 | print(f"recombining directory {dirs[i]} and {dirs[i+1]}, len {keynum}") 151 | 152 | 153 | 154 | 155 | for j in range(keynum, len(images_current) - 1): 156 | alpha = (j - keynum) / (len(images_current) - keynum) 157 | image1_path = os.path.join(current_dir, images_current[j]) 158 | next_image_index = j - keynum if j - keynum < len(images_next) else len(images_next) - 1 159 | image2_path = os.path.join(next_dir, images_next[next_image_index]) 160 | 161 | image1 = Image.open(image1_path) 162 | image2 = Image.open(image2_path) 163 | 164 | blended_image = butility.crossfade_images(image1, image2, alpha) 165 | output_images.append(np.array(blended_image)) 166 | # blended_image.save(os.path.join(output_folder, f"{dirs[i]}_{dirs[i+1]}_crossfade_{j:04}.png")) 167 | 168 | final_dir = os.path.join(root_folder, dirs[-1]) 169 | final_dir_images = sorted(os.listdir(final_dir)) 170 | 171 | # Find the index of the image with the last keyframe number in its name 172 | last_keyframe_number = allkeynums[-1] 173 | last_keyframe_index = None 174 | for index, image_name in enumerate(final_dir_images): 175 | number_in_name = int(''.join(filter(str.isdigit, image_name))) 176 | if number_in_name == last_keyframe_number: 177 | last_keyframe_index = index 178 | break 179 | 180 | if last_keyframe_index is not None: 181 | print(f"going from dir {last_keyframe_number} to end at {len(final_dir_images)}") 182 | 183 | # Iterate from the last keyframe number to the end 184 | for c in range(last_keyframe_index, len(final_dir_images)): 185 | image1_path = os.path.join(final_dir, final_dir_images[c]) 186 | image1 = Image.open(image1_path) 187 | output_images.append(np.array(image1)) 188 | else: 189 | print("Last keyframe not found in the final directory") 190 | 191 | 192 | print (f"outputting {len(output_images)} images") 193 | output_save_location = os.path.join(output_folder, "crossfade.mp4") 194 | generated_vid = extensions.TemporalKit.scripts.berry_utility.pil_images_to_video(output_images, output_save_location, fps) 195 | 196 | if return_generated_video_path == True: 197 | return generated_vid 198 | else: 199 | return output_images 200 | 201 | def getkeynums (folder_path): 202 | filenames = os.listdir(folder_path) 203 | 204 | # Filter filenames to keep only the ones starting with "keys" and ending with ".png" 205 | keys_filenames = [f for f in filenames if f.startswith("keys") and f.endswith(".png")] 206 | 207 | # Sort the filtered filenames 208 | sorted_keys_filenames = sorted(keys_filenames, key=lambda f: int(re.search(r'(\d+)', f).group(0))) 209 | 210 | # Extract the numbers from the sorted filenames 211 | return [int(re.search(r'(\d+)', f).group(0)) for f in sorted_keys_filenames] 212 | 213 | 214 | def get_num_at_index(folder_path,index): 215 | """Get the starting number of the output images in a folder.""" 216 | filenames = os.listdir(folder_path) 217 | 218 | # Filter filenames to keep only the ones starting with "keys" and ending with ".png" 219 | #keys_filenames = [f for f in filenames if f.startswith("keys") and f.endswith(".png")] 220 | 221 | # Sort the filtered filenames 222 | sorted_keys_filenames = sorted(filenames, key=lambda f: int(re.search(r'(\d+)', f).group(0))) 223 | 224 | # Extract the numbers from the sorted filenames 225 | numbers = [int(re.search(r'(\d+)', f).group(0)) for f in sorted_keys_filenames] 226 | return numbers[index] -------------------------------------------------------------------------------- /scripts/TemporalKitImg2ImgTab.py: -------------------------------------------------------------------------------- 1 | import gradio as gr 2 | from einops import rearrange 3 | from omegaconf import OmegaConf 4 | from PIL import Image, ImageOps 5 | from torch import autocast 6 | import modules.scripts as scripts 7 | import gradio as gr 8 | from modules import processing, images, shared, sd_samplers, devices 9 | import modules.generation_parameters_copypaste as parameters_copypaste 10 | import modules.images as images 11 | import os 12 | import glob 13 | lastimage = None 14 | 15 | def save_image(image, filename, directory): 16 | os.makedirs(directory, exist_ok=True) 17 | 18 | # create the full file path by joining the directory and filename 19 | file_path = os.path.join(directory, filename) 20 | 21 | # save the image to the specified file path 22 | image.save(file_path) 23 | 24 | def on_button_click(): 25 | global lastimage 26 | extension_path = os.path.abspath(__file__) 27 | extension_dir = os.path.dirname(os.path.dirname(extension_path)) 28 | extension_folder = os.path.join(extension_dir,"squares") 29 | 30 | # save the image to the parent directory with a new filename 31 | save_image(lastimage, 'last.png', extension_folder) 32 | 33 | 34 | class Script(scripts.Script): 35 | global button 36 | def title(self): 37 | return "TemporalKit" 38 | 39 | def show(self, is_img2img): 40 | if is_img2img: 41 | return True 42 | return True 43 | 44 | 45 | def ui(self, is_img2img): 46 | global lastimage 47 | savebutton = gr.Button("save", label="Save") 48 | savebutton.click( 49 | fn=on_button_click, 50 | ) 51 | movebutton = gr.Button("move", label="Move") 52 | movebutton.click( 53 | fn=None, 54 | _js="switch_to_temporal_kit", 55 | ) 56 | 57 | 58 | def run(self, p): 59 | global lastimage 60 | processed = processing.process_images(p) 61 | lastimage = processed.images[0] 62 | #registerbuttons(button) 63 | return processed 64 | 65 | -------------------------------------------------------------------------------- /scripts/berry_utility.py: -------------------------------------------------------------------------------- 1 | import os 2 | import glob 3 | from moviepy.editor import * 4 | import tempfile 5 | #om nom nom nom 6 | import requests 7 | import json 8 | from pprint import pprint 9 | import base64 10 | import numpy as np 11 | from io import BytesIO 12 | import scripts.optical_flow_simple as opflow 13 | from PIL import Image, ImageOps,ImageFilter 14 | import io 15 | from collections import deque 16 | import cv2 17 | import copy 18 | import shutil 19 | import subprocess 20 | import scenedetect 21 | 22 | window_size = 5 23 | 24 | intensity_window = deque(maxlen=window_size) 25 | 26 | def calculate_intensity(flow_map,frame_count): 27 | global min_intensity, max_intensity 28 | intensity_map = np.sqrt(np.sum(flow_map**2, axis=2)) 29 | 30 | intensity_window.append(intensity_map) 31 | 32 | min_intensity = min(im.min() for im in intensity_window) 33 | max_intensity = max(im.max() for im in intensity_window) 34 | 35 | normalized_intensity_map = (intensity_map - min_intensity) / (max_intensity - min_intensity) 36 | 37 | intensity_image = Image.fromarray((normalized_intensity_map * 255).astype(np.uint8)) 38 | intensity_image.save(f'intensitymaps/intensity_map_{frame_count:04d}.png') 39 | 40 | return normalized_intensity_map 41 | 42 | def scale_mask_intensity(mask, intensity): 43 | scaled_mask = np.clip(mask * intensity, 0, 255).astype(np.uint8) 44 | return scaled_mask 45 | 46 | def mask_to_grayscale(mask): 47 | grayscale_mask = 0.2989 * mask[:, :, 0] + 0.5870 * mask[:, :, 1] + 0.1140 * mask[:, :, 2] 48 | return grayscale_mask 49 | 50 | 51 | 52 | def replace_masked_area(flow,index, base_image_path, mask_base64_str, replacement_image_path, threshold=128): 53 | 54 | #intensity = 0.2 55 | 56 | #intensity_map = calculate_intensity(flow) 57 | 58 | 59 | # Load images 60 | base_image = Image.open(base_image_path).convert("RGBA") 61 | #mask_image = Image.open(io.BytesIO(base64.b64decode(mask_base64_str))).convert("L") 62 | mask_image = mask_base64_str # its not base64 encoded anymore 63 | #replacement_tex_unmodified = base64_to_texture(replacement_image_path) 64 | #replacement_image = Image.fromarray(np.uint8(replacement_tex_unmodified)).convert("RGBA") 65 | replacement_image = Image.open(replacement_image_path).convert("RGBA") 66 | print(mask_image) 67 | print(mask_image.size) 68 | 69 | # Resize the mask image and replacement image to match the base image size 70 | base_width, base_height = base_image.size 71 | alpha_mask = np.array(mask_image) 72 | 73 | mask_image = alpha_mask.resize((base_width, base_height)) 74 | replacement_image = replacement_image.resize((base_width, base_height)) 75 | 76 | alpha_mask_intensity = (alpha_mask).astype(np.uint8) 77 | alpha_image = Image.fromarray(alpha_mask_intensity) 78 | 79 | blended_image = Image.composite(replacement_image, base_image, alpha_image) 80 | # print(alpha_mask_intensity.size) 81 | # print (f"image{intensity_map.size}") 82 | 83 | output_image_path = os.path.join("./debug3/", f"output_image_{index}.png") 84 | # Save the output image 85 | os.makedirs("./debug3/", exist_ok=True) 86 | blended_image.save(output_image_path) 87 | 88 | 89 | return output_image_path 90 | # return base_image_path 91 | 92 | #not in use rn 93 | def invert_base64_image(base64_str: str) -> str: 94 | # Decode the base64 string 95 | img_data = base64.b64decode(base64_str) 96 | 97 | # Convert the decoded data to a PIL Image object 98 | img = Image.open(io.BytesIO(img_data)) 99 | 100 | # Invert the image colors 101 | inverted_img = ImageOps.invert(img) 102 | 103 | # Convert the inverted image back to a byte stream 104 | img_byte_arr = io.BytesIO() 105 | inverted_img.save(img_byte_arr, format='PNG') 106 | img_byte_arr = img_byte_arr.getvalue() 107 | 108 | # Encode the inverted image as a base64 string 109 | inverted_base64_str = base64.b64encode(img_byte_arr).decode('utf-8') 110 | 111 | return inverted_base64_str 112 | 113 | # hardens the mask 114 | def harden_mask(encoded_image,taper): 115 | decoded_image = base64.b64decode(encoded_image) 116 | image = Image.open(BytesIO(decoded_image)).convert("RGBA") 117 | new_image = image 118 | # Make every pixel not black solid white 119 | width, height = image.size 120 | for x in range(width): 121 | for y in range(height): 122 | r, g, b, a = image.getpixel((x, y)) 123 | # if r != 0 or g != 0 or b != 0: 124 | # image.putpixel((x, y), (255, 255, 255, a)) 125 | if r > 180 or g > 180 or b > 180: 126 | new_image.putpixel((x, y), (255, 255, 255, a)) 127 | 128 | #Taper for 3 pixels around 129 | if taper == True: 130 | new_image = new_image.filter(ImageFilter.BoxBlur(9)) 131 | 132 | 133 | for x in range(width): 134 | for y in range(height): 135 | r, g, b, a = image.getpixel((x, y)) 136 | # if r != 0 or g != 0 or b != 0: 137 | # image.putpixel((x, y), (255, 255, 255, a)) 138 | if r > 180 or g > 180 or b > 180: 139 | new_image.putpixel((x, y), (255, 255, 255, a)) 140 | 141 | 142 | for x in range(width): 143 | for y in range(height): 144 | r, g, b, a = new_image.getpixel((x, y)) 145 | # if r != 0 or g != 0 or b != 0: 146 | # image.putpixel((x, y), (255, 255, 255, a)) 147 | if r < 128 or g < 128 or b < 128: 148 | new_image.putpixel((x, y), (r + 5, g + 5, b + 5,a)) 149 | 150 | output_buffer = BytesIO() 151 | new_image.save(output_buffer, format="PNG") 152 | processed_image_base64 = base64.b64encode(output_buffer.getvalue()).decode() 153 | # combined = overlay_base64_images(encoded_image,processed_image_base64) 154 | return processed_image_base64 155 | 156 | 157 | def resize_base64_image(base64_str, new_width: int, new_height: int) -> str: 158 | # Decode the base64 string 159 | #img_data = base64.b64decode(base64_str) 160 | #img_data2 = Image.fromarray( base64_to_texture(base64_str)) 161 | # Convert the decoded data to a PIL Image object 162 | #img = Image.open(io.BytesIO(img_data)) 163 | 164 | # Resize the image 165 | #resized_img = img_data2.resize((new_width, new_height), Image.ANTIALIAS) 166 | 167 | #decoded_data = base64.b64decode(base64_str) 168 | with open(base64_str, "rb") as f: 169 | bytes = f.read() 170 | 171 | # Create a BytesIO object 172 | buffered_data = io.BytesIO(bytes) 173 | 174 | # Open the image from the BytesIO object 175 | image = Image.open(buffered_data) 176 | resized_img = image.resize((new_width, new_height), Image.ANTIALIAS) 177 | 178 | # Convert the resized image back to a byte stream 179 | img_byte_arr = io.BytesIO() 180 | resized_img.save(img_byte_arr, format='PNG') 181 | img_byte_arr = img_byte_arr.getvalue() 182 | 183 | # Encode the resized image as a base64 string 184 | resized_base64_str = base64.b64encode(img_byte_arr).decode('utf-8') 185 | 186 | return resized_base64_str 187 | 188 | def overlay_base64_images(encoded_image1, encoded_image2): 189 | decoded_image1 = base64.b64decode(encoded_image1) 190 | decoded_image2 = base64.b64decode(encoded_image2) 191 | 192 | image1 = Image.open(BytesIO(decoded_image1)) 193 | image2 = Image.open(BytesIO(decoded_image2)) 194 | 195 | # Overlay the images 196 | result = Image.blend(image1, image2,0.5) 197 | 198 | # Convert the overlaid image back to base64 199 | output_buffer = BytesIO() 200 | result.save(output_buffer, format="PNG") 201 | result_base64 = base64.b64encode(output_buffer.getvalue()).decode() 202 | 203 | return result_base64 204 | 205 | #get all images inthe server 206 | def get_image_paths(folder): 207 | image_extensions = ("*.jpg", "*.jpeg", "*.png", "*.bmp") 208 | files = [] 209 | for ext in image_extensions: 210 | files.extend(glob.glob(os.path.join(folder, ext))) 211 | return sorted(files) 212 | 213 | # convert image to base64 214 | # is this really th best way to do this? 215 | def texture_to_base64(texture): 216 | # Convert the NumPy array to a PIL Image 217 | image = Image.fromarray(texture).convert("RGBA") 218 | 219 | # Save the image to an in-memory buffer 220 | buffer = BytesIO() 221 | image.save(buffer, format="PNG") 222 | 223 | # Get the byte data from the buffer and encode it as a base64 string 224 | img_base64 = base64.b64encode(buffer.getvalue()).decode() 225 | 226 | return img_base64 227 | 228 | # thanks to https://github.com/jinnsp ❤ 229 | def base64_to_texture(base64_string): 230 | if base64_string.lower().endswith('png'): 231 | image = Image.open(base64_string) 232 | else: 233 | decoded_data = base64.b64decode(base64_string) 234 | buffer = BytesIO(decoded_data) 235 | image = Image.open(buffer) 236 | texture = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR) 237 | return texture 238 | 239 | def combine_masks(masks): 240 | """ 241 | Combine three grayscale masks into one by overlaying them. 242 | 243 | Args: 244 | mask1 (np.ndarray): First grayscale mask with shape (height, width). 245 | mask2 (np.ndarray): Second grayscale mask with shape (height, width). 246 | mask3 (np.ndarray): Third grayscale mask with shape (height, width). 247 | 248 | Returns: 249 | np.ndarray: Combined grayscale mask with the same shape as the input masks. 250 | """ 251 | combined_mask = np.maximum.reduce(masks) 252 | return combined_mask 253 | 254 | def create_hole_mask(flow_map): 255 | h, w, _ = flow_map.shape 256 | x_coords, y_coords = np.meshgrid(np.arange(w), np.arange(h)) 257 | 258 | # Compute the new coordinates of each pixel after the optical flow is applied 259 | new_x_coords = np.clip(x_coords + flow_map[..., 0], 0, w - 1).astype(int) 260 | new_y_coords = np.clip(y_coords + flow_map[..., 1], 0, h - 1).astype(int) 261 | 262 | # Create a 2D array to keep track of whether a pixel is occupied or not 263 | occupied = np.zeros((h, w), dtype=bool) 264 | 265 | # Mark the pixels that are occupied after the optical flow is applied 266 | occupied[new_y_coords, new_x_coords] = True 267 | 268 | # Create the hole mask by marking unoccupied pixels as holes (value of 1) 269 | hole_mask = np.logical_not(occupied).astype(np.uint8) 270 | 271 | 272 | 273 | expanded = filter_mask(hole_mask) * 255 274 | #expanded = hole_mask * 255 275 | #blurred_hole_mask = box_(expanded, sigma=3) 276 | toblur = Image.fromarray(expanded).convert('L') 277 | blurred_hole_mask = np.array(toblur.filter(ImageFilter.GaussianBlur(3))) 278 | 279 | #blurred_numpy = np.array( Image.fromarray(expanded).filter(ImageFilter.GaussianBlur(3))) 280 | #blurred_hole_mask[blurred_hole_mask > 150] = 255 281 | filtered_smol = filter_mask(hole_mask,4,0.4,0.3) * 255 282 | return blurred_hole_mask + filtered_smol 283 | 284 | # there are pixels all over the place that are not holes, so this only gets the holes with a high concentration 285 | def filter_mask(mask, kernel_size=4, threshold_ratio=0.3,grayscale_intensity=1.0): 286 | # Create a custom kernel 287 | kernel = np.ones((kernel_size, kernel_size), np.uint8) 288 | 289 | # Convolve the mask with the kernel 290 | conv_result = cv2.filter2D(mask, -1, kernel) 291 | 292 | # Calculate the threshold based on the ratio 293 | threshold = int(kernel.size * threshold_ratio) 294 | 295 | # Filter the mask using the calculated threshold 296 | filtered_mask = np.where(conv_result >= threshold, mask, 0) 297 | 298 | # thanks to https://github.com/jinnsp ❤ 299 | grayscale_mask = np.where(conv_result >= threshold, int(255 * grayscale_intensity), 0).astype(np.uint8) 300 | 301 | # Combine the filtered mask and grayscale mask 302 | combined_mask = np.maximum(filtered_mask, grayscale_mask) 303 | 304 | 305 | return combined_mask 306 | 307 | 308 | def resize_image(image, max_dimension_x, max_dimension_y): 309 | h, w = image.shape[:2] 310 | aspect_ratio = float(w) / float(h) 311 | if h > w: 312 | new_height = max_dimension_y 313 | new_width = int(new_height * aspect_ratio) 314 | else: 315 | new_width = max_dimension_x 316 | new_height = int(new_width / aspect_ratio) 317 | resized_image = cv2.resize(image, (new_width, new_height), interpolation=cv2.INTER_LINEAR) 318 | return resized_image 319 | 320 | #delete this later 321 | def replaced_mask_from_other_direction_debug(index,image, mask, flowmap, original, output_folder1='./debug',forwards=True): 322 | # Ensure inputs are numpy arrays 323 | image = np.array(image) 324 | mask = np.array(mask) 325 | 326 | flowmap_detached = flowmap.detach().cpu().numpy() 327 | if original is not None: 328 | original = np.array(original) 329 | original = resize_image(original, image.shape[1], image.shape[0]) 330 | 331 | 332 | 333 | # Get the height and width of the image 334 | height, width, _ = image.shape 335 | 336 | # Get the coordinates of the masked pixels 337 | masked_coords = np.argwhere(mask) 338 | 339 | # Initialize a new_image to store the result 340 | new_image = image.copy() 341 | 342 | # Initialize debug_image to store debug information 343 | 344 | 345 | for y, x in masked_coords: 346 | border_limit = 10 347 | # Skip if pixel is within 20 pixels of the border 348 | if border_limit <= y < height - border_limit and border_limit <= x < width - border_limit: 349 | 350 | # Get the opposite flow vector 351 | flow_vector = np.array([flowmap_detached[0, y, x], flowmap_detached[1, y, x]]) 352 | 353 | # Calculate the new coordinates after applying the flow 354 | new_y = int(round(y - flow_vector[1] * 2)) 355 | new_x = int(round(x - flow_vector[0] * 2)) 356 | 357 | # Check if the new coordinates are inside the image bounds 358 | if 0 <= new_y < height and 0 <= new_x < width: 359 | # Replace the pixel in the new_image with the pixel from the opposite flow direction 360 | intensity = mask[y, x] / 255 361 | weight = gaussian_weight(intensity, sigma=0.5) 362 | 363 | # Replace the pixel in the new_image with the pixel from the opposite flow direction, weighted by the Gaussian weight 364 | #new_image[y, x] = weight * image[new_y, new_x] + (1 - weight) * image[y, x] 365 | #if original is not None and not is_similar_color(image[y, x], original[y, x],50): 366 | # weird edge case where if the background is moving and the foreground is not, the foreground will be replaced by the background 367 | ### new_image[y, x] = weight * image[y, x] + (1 - weight) * original[y, x] 368 | #else: 369 | new_image[y, x] = (1 - weight) * image[new_y, new_x] + weight * image[y, x] 370 | 371 | 372 | # Replace the pixel in the new_image with the pixel from the opposite flow direction, weighted by the intensity 373 | #new_image[y, x] = intensity * image[new_y, new_x] + (1 - intensity) * image[y, x] 374 | #new_image[y, x] = image[new_y, new_x] 375 | #new_image[y, x] = image[new_y, new_x] + (1 - intensity) * image[y, x] 376 | # Update debug_image 377 | #debug_image[y, x] = [0,0,255 * (1 - weight)] # Red color for the mask 378 | #debug_image[new_y, new_x] = [0, 255 * (1 - weight),0] # Green color for the moved pixel content 379 | #debug_image[y, x] = [0,0,255 * (1 - weight)] # Blue color for the moved pixel destination 380 | 381 | 382 | #output_folder = os.path.join(output_folder1, f'newmaskimg{index}.png') 383 | # Save the debug image to the specified folder 384 | 385 | 386 | #if original is not None: 387 | #debug_image_pil = Image.fromarray(debug_image) 388 | #debug_image_pil.save(output_folder) 389 | 390 | return texture_to_base64(new_image) 391 | 392 | 393 | def is_similar_color(pixel1, pixel2, threshold): 394 | """ 395 | Returns True if pixel1 and pixel2 are similar in color within the given threshold, False otherwise. 396 | """ 397 | r1, g1, b1,a1 = pixel1 398 | r2, g2, b2,a2 = pixel2 399 | color_difference = ((r1 - r2) ** 2 + (g1 - g2) ** 2 + (b1 - b2) ** 2) ** 0.5 400 | return color_difference <= threshold 401 | 402 | def gaussian_weight(d, sigma=1.0): 403 | return np.exp(-(d ** 2) / (2 * sigma ** 2)) 404 | 405 | def avg_edge_pixels(img): 406 | height, width = img.shape[:2] 407 | edge_pixels = [] 408 | 409 | # top and bottom edges 410 | edge_pixels.extend(img[0,:]) 411 | edge_pixels.extend(img[height-1,:]) 412 | 413 | # left and right edges 414 | edge_pixels.extend(img[:,0]) 415 | edge_pixels.extend(img[:,width-1]) 416 | 417 | # calculate average of edge pixels 418 | avg_edge_pixel = np.mean(edge_pixels) 419 | 420 | return avg_edge_pixel 421 | 422 | def check_edges(image): 423 | h, w, c = image.shape 424 | threshold = 0.4 425 | 426 | def is_different(pixel1, pixel2): 427 | return np.any(np.abs(pixel1 - pixel2) > threshold * 255) 428 | 429 | for i in range(h): 430 | for j in range(w): 431 | if i < 2 or i > h - 3 or j < 2 or j > w - 3: 432 | central_i = i + 5 if i < h // 2 else i - 5 433 | central_j = j + 5 if j < w // 2 else j - 5 434 | 435 | # Ensure the central pixel is within the image boundaries 436 | central_i = max(0, min(central_i, h - 1)) 437 | central_j = max(0, min(central_j, w - 1)) 438 | 439 | if is_different(image[i, j], image[central_i, central_j]): 440 | image[i, j] = image[central_i, central_j] 441 | 442 | 443 | def resize_to_nearest_multiple_of_8(width, height): 444 | def nearest_multiple(n, factor): 445 | return round(n / factor) * factor 446 | 447 | new_width = nearest_multiple(width, 8) 448 | new_height = nearest_multiple(height, 8) 449 | return new_width, new_height 450 | 451 | 452 | 453 | def resize_to_nearest_multiple(width, height, a): 454 | def nearest_common_multiple(target, a, b): 455 | multiple = 1 456 | nearest_multiple = 0 457 | min_diff = float('inf') 458 | 459 | while True: 460 | current_multiple = a * multiple 461 | if current_multiple % b == 0: 462 | diff = abs(target - current_multiple) 463 | if diff < min_diff: 464 | min_diff = diff 465 | nearest_multiple = current_multiple 466 | else: 467 | break 468 | multiple += 1 469 | 470 | return nearest_multiple 471 | 472 | new_width = nearest_common_multiple(width, a, 8) 473 | new_height = nearest_common_multiple(height, a, 8) 474 | return int(new_width), int(new_height) 475 | 476 | 477 | def delete_folder_contents(folder_path): 478 | for filename in os.listdir(folder_path): 479 | file_path = os.path.join(folder_path, filename) 480 | try: 481 | if os.path.isfile(file_path) or os.path.islink(file_path): 482 | os.unlink(file_path) 483 | elif os.path.isdir(file_path): 484 | shutil.rmtree(file_path) 485 | except Exception as e: 486 | print(f'Failed to delete {file_path}. Reason: {e}') 487 | 488 | def blend_images(img1, img2, alpha=0.5): 489 | blended = cv2.addWeighted(img1, alpha, img2, 1-alpha, 0) 490 | return blended 491 | 492 | 493 | def pil_images_to_video(pil_images, output_file, fps=24): 494 | """ 495 | Saves an array of PIL images to a video file using MoviePy. 496 | 497 | Args: 498 | pil_images (list): A list of PIL images. 499 | output_file (str): The output file path for the video. 500 | fps (int, optional): The desired frames per second. Defaults to 24. 501 | 502 | Returns: 503 | the filepath of the video file 504 | """ 505 | # Convert PIL images to NumPy arrays 506 | np_images = [np.array(img) for img in pil_images] 507 | 508 | # Create an ImageSequenceClip instance with the array of NumPy images and the specified fps 509 | clip = ImageSequenceClip(np_images, fps=fps) 510 | 511 | # Write the video file to the specified output location 512 | clip.write_videofile(output_file,fps,codec='libx264') 513 | 514 | return output_file 515 | 516 | def copy_video(source_path, destination_path): 517 | """ 518 | Copy a video file from source_path to destination_path. 519 | 520 | :param source_path: str, path to the source video file 521 | :param destination_path: str, path to the destination video file 522 | :return: None 523 | """ 524 | try: 525 | shutil.copy(source_path, destination_path) 526 | print(f"Video copied successfully from {source_path} to {destination_path}") 527 | except IOError as e: 528 | print(f"Unable to copy video. Error: {e}") 529 | except Exception as e: 530 | print(f"Unexpected error: {e}") 531 | 532 | def crossfade_frames(frame1, frame2, alpha): 533 | """Crossfade between two video frames with a given alpha value.""" 534 | image1 = Image.fromarray(frame1) 535 | image2 = Image.fromarray(frame2) 536 | blended_image = crossfade_images(image1, image2, alpha) 537 | blended_image = blended_image.convert('RGB') 538 | #THIS FUNCTION CONSUMED 3 HOURS OF DEBUGING AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA 539 | return np.array(blended_image) 540 | 541 | 542 | 543 | 544 | def crossfade_videos(video_paths,fps, overlap_indexes, num_overlap_frames, output_path): 545 | 546 | """ 547 | a python function and i need it to input an array of video paths, 548 | an array of the indexes where the videos overlap with the next video, 549 | the number of overlapping frames and the next file and i need it to combine these 550 | clips while crossfading between the clips where there is overlapping happening and 551 | output it to the output file location 552 | """ 553 | 554 | #not video paths any more frame arrays 555 | original_frames_arrays = video_paths 556 | #for i in range(len(video_paths)): 557 | # data = convert_video_to_bytes(video_paths[i]) 558 | ## frames_list = extract_frames_movpie(data, fps) 559 | # original_frames_arrays.append(frames_list) 560 | # print (f"video {i} has {len(frames_list)} frames") 561 | new_frames_arrays = copy.deepcopy(original_frames_arrays) 562 | 563 | for index, frames_array in enumerate(original_frames_arrays): 564 | if index < len(original_frames_arrays) - 1 and index in overlap_indexes: 565 | next_array = original_frames_arrays[index+1] 566 | print (f"crossfading between video {index} and video {index+1}") 567 | first_of_next = next_array[:num_overlap_frames] 568 | last_of_current = frames_array[-num_overlap_frames:] 569 | #last_of_current = last_of_current[::-1] 570 | if len(first_of_next) != len(last_of_current): 571 | print ("crossfade frames are not the same length") 572 | while len(first_of_next) != len(last_of_current): 573 | if len(first_of_next) > len(last_of_current): 574 | first_of_next.pop() # remove the last element from array1 575 | else: 576 | last_of_current.pop() # remove the last element from array2 577 | 578 | crossfaded = [] 579 | for i in range(num_overlap_frames): 580 | alpha = 1 - (i / num_overlap_frames) # set alpha value 581 | if i > len(last_of_current) - 1 or i > len(first_of_next) - 1: 582 | 583 | print ("ran out of crossfade space at index ",str(i)) 584 | break; 585 | 586 | new_frame = crossfade_frames(last_of_current[i], first_of_next[i], alpha) 587 | #print (new_frame.shape) 588 | crossfaded.append(new_frame) 589 | print (f"crossfaded {len(crossfaded)} frames with num overlap = {num_overlap_frames}, the last of current array is of length {len(last_of_current)} and the first of next is of length {len(first_of_next)}") 590 | #saving first of next and last of current 591 | new_frames_arrays[index][-num_overlap_frames:] = crossfaded 592 | 593 | if index > 0 and index - 1 in overlap_indexes: 594 | new_frames_arrays[index] = new_frames_arrays[index][num_overlap_frames:] 595 | 596 | for arr in new_frames_arrays: 597 | print(len(arr)) 598 | #combined_arrays = np.concatenate(new_frames_arrays) 599 | output_array = [] 600 | for arr in new_frames_arrays: 601 | for frame in arr: 602 | #frame = cv2.resize(frame, (new_frames_arrays, new_height), interpolation=cv2.INTER_LINEAR) 603 | #print (frame.shape) 604 | output_array.append(Image.fromarray(frame).convert("RGB")) 605 | return pil_images_to_video(output_array, output_path, fps) 606 | 607 | 608 | 609 | 610 | def crossfade_images(image1, image2, alpha): 611 | """Crossfade between two images with a given alpha value.""" 612 | image1 = image1.convert("RGBA") 613 | image2 = image2.convert("RGBA") 614 | return Image.blend(image1, image2, alpha) 615 | 616 | 617 | def extract_frames_movpie(input_video, target_fps, max_frames=None, perform_interpolation=False): 618 | print(f"Interpolating extra frames with max frames {max_frames} and interpolating = {perform_interpolation}" ) 619 | 620 | def get_video_info(video_path): 621 | cmd = ['ffprobe', '-v', 'quiet', '-print_format', 'json', '-show_streams', video_path] 622 | 623 | try: 624 | result = subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True) 625 | except FileNotFoundError as e: 626 | print(f"Error: {e}. Please ensure that ffmpeg is installed and available in your system's PATH.") 627 | return None 628 | 629 | return json.loads(result.stdout) 630 | 631 | 632 | def interpolate_frames(frame1, frame2, ratio): 633 | flow = cv2.calcOpticalFlowFarneback(cv2.cvtColor(frame1, cv2.COLOR_BGR2GRAY), cv2.cvtColor(frame2, cv2.COLOR_BGR2GRAY), None, 0.5, 3, 15, 3, 5, 1.2, 0) 634 | return cv2.addWeighted(frame1, 1 - ratio, frame2, ratio, 0) + ratio * cv2.remap(frame1, flow * (1 - ratio), None, cv2.INTER_LINEAR) 635 | 636 | with tempfile.NamedTemporaryFile(suffix='.mp4', delete=False) as f: 637 | f.write(input_video) 638 | video_path = f.name 639 | 640 | video_info = get_video_info(video_path) 641 | video_stream = next((stream for stream in video_info['streams'] if stream['codec_type'] == 'video'), None) 642 | numerator, denominator = map(int, video_stream['avg_frame_rate'].split('/')) 643 | original_fps = float(numerator) / float(denominator) if denominator != 0 else 0.0 644 | 645 | video_clip = VideoFileClip(video_path) 646 | video_duration = video_clip.duration 647 | print (f"video duration is {video_duration}") 648 | frames = [] 649 | frame_ratio = original_fps / target_fps 650 | frame_time = 1 / target_fps 651 | current_time = 0 652 | 653 | 654 | if max_frames is None: 655 | max_frames = int(video_duration * target_fps) 656 | else: 657 | max_frames = min(max_frames, int(video_duration * target_fps)) 658 | 659 | if not perform_interpolation or target_fps <= original_fps: 660 | frame_repeat = int(target_fps / original_fps) 661 | print (f"frame repeat is {frame_repeat}, target fps is {target_fps} and original fps is {original_fps}") 662 | if frame_repeat == 0: 663 | frame_repeat = 1 664 | input_frame_time = 0 665 | input_frame_step = 1 / original_fps 666 | 667 | while len(frames) < max_frames and input_frame_time < video_duration: 668 | frame = video_clip.get_frame(input_frame_time) 669 | for _ in range(frame_repeat): 670 | frames.append(frame) 671 | if len(frames) >= max_frames: 672 | break 673 | input_frame_time += input_frame_step 674 | else: 675 | while len(frames) < max_frames: 676 | current_time = (len(frames) / target_fps) * video_duration 677 | frame1_time = current_time * frame_ratio 678 | frame2_time = min(current_time * frame_ratio + frame_ratio, video_duration) 679 | 680 | if frame2_time >= video_duration: 681 | break 682 | 683 | frame1 = video_clip.get_frame(frame1_time) 684 | frame2 = video_clip.get_frame(frame2_time) 685 | 686 | ratio = (current_time * original_fps) % 1 687 | frame = interpolate_frames(frame1, frame2, ratio) 688 | 689 | frames.append(frame) 690 | 691 | print(f"Extracted {len(frames)} frames at {target_fps} fps over a clip with a length of {len(frames) / target_fps} seconds with the old duration of {video_duration} seconds") 692 | return frames 693 | 694 | 695 | 696 | def convert_video_to_bytes(input_file): 697 | # Read the uploaded video file 698 | print(f"reading video file... {input_file}") 699 | with open(input_file, "rb") as f: 700 | video_bytes = f.read() 701 | 702 | # Return the processed video bytes (or any other output you want) 703 | return video_bytes 704 | 705 | def split_video_into_numpy_arrays(video_path, target_fps=None, perform_interpolation=False): 706 | 707 | video_manager = scenedetect.VideoManager([video_path]) 708 | scene_manager = scenedetect.SceneManager() 709 | scene_manager.add_detector(scenedetect.ContentDetector()) 710 | 711 | video_manager.set_downscale_factor() 712 | video_manager.start() 713 | 714 | scene_manager.detect_scenes(frame_source=video_manager) 715 | scene_list = scene_manager.get_scene_list(start_in_scene=True) 716 | 717 | if target_fps is not None: 718 | original_fps = video_manager.get(cv2.CAP_PROP_FPS) 719 | 720 | if len(scene_list) == 0: 721 | start_time = 0 722 | end_time = video_manager.get(cv2.CAP_PROP_FRAME_COUNT) / video_manager.get(cv2.CAP_PROP_FPS) 723 | scene_list.append((start_time, end_time)) 724 | 725 | print (f"Detected {len(scene_list)} scenes") 726 | numpy_arrays = save_scenes_as_numpy_arrays(scene_list, video_path, target_fps, original_fps if target_fps else None, perform_interpolation) 727 | 728 | print(f"Total scenes: {len(numpy_arrays)}") 729 | return numpy_arrays 730 | 731 | def save_scenes_as_numpy_arrays(scene_list, video_path, target_fps=None, original_fps=None, perform_interpolation=True): 732 | def interpolate_frames(frame1, frame2, ratio): 733 | flow = cv2.calcOpticalFlowFarneback(cv2.cvtColor(frame1, cv2.COLOR_BGR2GRAY), cv2.cvtColor(frame2, cv2.COLOR_BGR2GRAY), None, 0.5, 3, 15, 3, 5, 1.2, 0) 734 | return cv2.addWeighted(frame1, 1 - ratio, frame2, ratio, 0) + ratio * cv2.remap(frame1, flow * (1 - ratio), None, cv2.INTER_LINEAR) 735 | 736 | numpy_arrays = [] 737 | video_capture = cv2.VideoCapture(video_path) 738 | 739 | if target_fps and original_fps: 740 | frame_ratio = original_fps / target_fps 741 | frame_time = 1 / target_fps 742 | 743 | for i, (start_time, end_time) in enumerate(scene_list): 744 | start_frame = int(start_time.get_frames()) 745 | end_frame = int(end_time.get_frames()) 746 | scene_frames = [] 747 | current_time = start_frame / original_fps if target_fps else start_frame 748 | 749 | print(f"Processing scene {i + 1}: start_frame={start_frame}, end_frame={end_frame} original fps={original_fps} target fps={target_fps}") 750 | 751 | while current_time < end_frame / original_fps if target_fps else end_frame: 752 | if target_fps and original_fps and perform_interpolation: 753 | frame1_time = current_time * frame_ratio 754 | frame2_time = min((current_time + frame_time) * frame_ratio, end_frame / original_fps) 755 | else: 756 | frame1_time = current_time 757 | frame2_time = current_time 758 | 759 | video_capture.set(cv2.CAP_PROP_POS_FRAMES, frame1_time * original_fps if target_fps else frame1_time) 760 | ret, frame1 = video_capture.read() 761 | 762 | if ret: 763 | frame1 = cv2.cvtColor(frame1, cv2.COLOR_BGR2RGB) 764 | 765 | if target_fps and original_fps and perform_interpolation: 766 | video_capture.set(cv2.CAP_PROP_POS_FRAMES, frame2_time * original_fps) 767 | ret, frame2 = video_capture.read() 768 | 769 | if ret: 770 | frame2 = cv2.cvtColor(frame2, cv2.COLOR_BGR2RGB) 771 | 772 | ratio = (frame1_time * original_fps) % 1 773 | frame = interpolate_frames(frame1, frame2, ratio) 774 | else: 775 | frame = frame1 776 | 777 | scene_frames.append(frame) 778 | 779 | current_time += frame_time if target_fps else 1 780 | print(f"Scene {i + 1} has {len(scene_frames)} frames with length of {len(scene_frames)} frames with the old duration of {end_frame - start_frame} frames") 781 | numpy_arrays.append(np.array(scene_frames)) 782 | 783 | video_capture.release() 784 | return numpy_arrays 785 | -------------------------------------------------------------------------------- /scripts/optical_flow_raft.py: -------------------------------------------------------------------------------- 1 | import cv2 2 | import numpy as np 3 | import os 4 | import sys 5 | import torch 6 | from PIL import Image 7 | import matplotlib.pyplot as plt 8 | import torchvision.transforms.functional as F 9 | from torchvision.io import read_video, read_image, ImageReadMode 10 | from torchvision.models.optical_flow import Raft_Large_Weights 11 | from torchvision.models.optical_flow import raft_large 12 | from torchvision.io import write_jpeg 13 | import torchvision.transforms as T 14 | import scripts.berry_utility as utilityb 15 | import tempfile 16 | from pathlib import Path 17 | from urllib.request import urlretrieve 18 | from scipy.interpolate import LinearNDInterpolator 19 | from imageio import imread, imwrite 20 | from torchvision.utils import flow_to_image 21 | 22 | device = "cuda" if torch.cuda.is_available() else "cpu" 23 | model = raft_large(weights=Raft_Large_Weights.DEFAULT, progress=False).to(device) 24 | model = model.eval() 25 | 26 | #no clue if this works 27 | def flow_to_rgb(flow): 28 | """ 29 | Convert optical flow to RGB image 30 | 31 | :param flow: optical flow map 32 | :return: RGB image 33 | 34 | """ 35 | # forcing conversion to float32 precision 36 | flow = flow.numpy() 37 | hsv = np.zeros(flow.shape, dtype=np.uint8) 38 | hsv[..., 1] = 255 39 | 40 | mag, ang = cv2.cartToPolar(flow[..., 0], flow[..., 1]) 41 | hsv[..., 0] = ang * 180 / np.pi / 2 42 | hsv[..., 2] = cv2.normalize(mag, None, 0, 255, cv2.NORM_MINMAX) 43 | #bgr = cv2.cvtColor(hsv, cv2.COLOR_HSV2RGB) 44 | #cv2.imshow("colored flow", bgr) 45 | #cv2.waitKey(0) 46 | #cv2.destroyAllWindows() 47 | 48 | return hsv 49 | 50 | def write_flo(flow, filename): 51 | """ 52 | Write optical flow in Middlebury .flo format 53 | 54 | :param flow: optical flow map 55 | :param filename: optical flow file path to be saved 56 | :return: None 57 | 58 | from https://github.com/liruoteng/OpticalFlowToolkit/ 59 | 60 | """ 61 | # forcing conversion to float32 precision 62 | flow = flow.cpu().data.numpy() 63 | flow = flow.astype(np.float32) 64 | f = open(filename, 'wb') 65 | magic = np.array([202021.25], dtype=np.float32) 66 | (height, width) = flow.shape[0:2] 67 | w = np.array([width], dtype=np.int32) 68 | h = np.array([height], dtype=np.int32) 69 | magic.tofile(f) 70 | w.tofile(f) 71 | h.tofile(f) 72 | flow.tofile(f) 73 | f.close() 74 | 75 | 76 | #def infer_old (frameA,frameB) 77 | 78 | def infer(frameA, frameB): 79 | device = torch.device("cuda" if torch.cuda.is_available() else "cpu") 80 | model = raft_large(weights=Raft_Large_Weights.DEFAULT, progress=False).to(device) 81 | model = model.eval() 82 | 83 | # Check if both frames have the same size 84 | if frameA.size != frameB.size: 85 | raise ValueError("Both input frames must have the same size") 86 | 87 | transform = T.ToTensor() 88 | 89 | img1_batch = transform(frameA) 90 | img2_batch = transform(frameB) 91 | img1_batch = torch.stack([img1_batch]) 92 | img2_batch = torch.stack([img2_batch]) 93 | weights = Raft_Large_Weights.DEFAULT 94 | transforms = weights.transforms() 95 | 96 | def preprocess(img1_batch, img2_batch): 97 | return transforms(img1_batch, img2_batch) 98 | 99 | img1_batch, img2_batch = preprocess(img1_batch, img2_batch) 100 | 101 | return img1_batch, img2_batch 102 | 103 | 104 | 105 | def apply_flow_based_on_images (image1_path, image2_path, provided_image_path,max_dimension, index,output_folder): 106 | w,h = get_target_size(utilityb.base64_to_texture(image1_path), max_dimension) 107 | w = int(w / 8) * 8 108 | h = int(h / 8) * 8 109 | image1 = resize_image(utilityb.base64_to_texture(image1_path),h,w) 110 | h, w = image1.shape[:2] 111 | image2 = cv2.resize(utilityb.base64_to_texture(image2_path), (w,h), interpolation=cv2.INTER_LINEAR) 112 | 113 | # image1 = utilityb.base64_to_texture(image1_path),max_dimension 114 | # image2 = utilityb.base64_to_texture(image2_path),max_dimension 115 | # provided_image = read_image(provided_image_path) 116 | provided_image = utilityb.base64_to_texture(provided_image_path) 117 | provided_image = cv2.resize(provided_image, (w,h), interpolation=cv2.INTER_LINEAR) 118 | 119 | 120 | 121 | img1_batch,img2_batch = infer(image1,image2) 122 | list_of_flows = model(img1_batch.to(device), img2_batch.to(device)) 123 | predicted_flows = list_of_flows[-1] 124 | predicted_flow = list_of_flows[-1][0] 125 | flow_img = flow_to_image(predicted_flow).to("cpu") 126 | #flo_file = write_flo(predicted_flow, "flofile.flo") 127 | 128 | #write_jpeg(flow_img, f"./flow/predicted_flow{index}.jpg") 129 | #write_jpeg(flow_img, os.path.join("temp", f'flow_{index + 1}.flo')) 130 | 131 | #print(flow.shape) 132 | #warped_image = apply_flow_to_image_try3(provided_image,predicted_flow) 133 | warped_image,unused_mask,white_pixels = apply_flow_to_image_with_unused_mask(provided_image,predicted_flow) 134 | 135 | 136 | 137 | 138 | warped_image_path = os.path.join(output_folder, f'warped_provided_image_{index + 1}.png') 139 | save_image(warped_image, warped_image_path) 140 | return warped_image_path,predicted_flow,unused_mask,white_pixels,flow_img 141 | 142 | def apply_flow_to_image(image, flow): 143 | """ 144 | Apply optical flow transforms to an input image 145 | 146 | :param image: input image 147 | :param flow: optical flow map 148 | :return: warped image 149 | 150 | """ 151 | 152 | # forcing conversion to float32 precision 153 | #flow = flow.numpy() 154 | flow = flow.astype(np.float32) 155 | 156 | # Get the height and width of the input image 157 | height, width = image.shape[:2] 158 | 159 | # Create a grid of (x, y) coordinates 160 | x, y = np.meshgrid(np.arange(width), np.arange(height)) 161 | 162 | # Apply the optical flow to the coordinates 163 | x_warped = (x + flow[..., 0]).astype(np.float32) 164 | y_warped = (y + flow[..., 1]).astype(np.float32) 165 | 166 | # Remap the input image using the warped coordinates 167 | warped_image = cv2.remap(image, x_warped, y_warped, cv2.INTER_LINEAR) 168 | 169 | return warped_image 170 | 171 | def warp_image(image, flow): 172 | h, w = image.shape[:2] 173 | 174 | flow_map = np.array([[x, y] for y in range(h) for x in range(w)], dtype=np.float32) - flow.reshape(-1, 2) 175 | flow_map = flow_map.reshape(h, w, 2).astype(np.float32) # Ensure the flow_map is in the correct format 176 | 177 | # Clip the flow_map to the image bounds 178 | flow_map[:, :, 0] = np.clip(flow_map[:, :, 0], 0, w - 1) 179 | flow_map[:, :, 1] = np.clip(flow_map[:, :, 1], 0, h - 1) 180 | 181 | warped_image = cv2.remap(image, flow_map, None, cv2.INTER_LANCZOS4) 182 | return warped_image 183 | 184 | def save_image(image, file_path): 185 | cv2.imwrite(file_path, image) 186 | 187 | def resize_image(image, new_height,new_width): 188 | resized_image = cv2.resize(image, (new_width, new_height), interpolation=cv2.INTER_LINEAR) 189 | return resized_image 190 | 191 | def get_target_size (image,max_dimension): 192 | h, w = image.shape[:2] 193 | aspect_ratio = float(w) / float(h) 194 | if h > w: 195 | new_height = max_dimension 196 | new_width = int(new_height * aspect_ratio) 197 | else: 198 | new_width = max_dimension 199 | new_height = int(new_width / aspect_ratio) 200 | return new_width,new_height 201 | 202 | 203 | def apply_flow_to_image_try3(image,flow): 204 | """ 205 | Apply an optical flow tensor to a NumPy image by moving the pixels based on the flow. 206 | 207 | Args: 208 | image (np.ndarray): Input image with shape (height, width, channels). 209 | flow (np.ndarray): Optical flow tensor with shape (height, width, 2). 210 | 211 | Returns: 212 | np.ndarray: Warped image with the same shape as the input image. 213 | """ 214 | height, width, _ = image.shape 215 | x_coords, y_coords = np.meshgrid(np.arange(width), np.arange(height)) 216 | coords = np.stack([x_coords, y_coords], axis=-1).astype(np.float32) 217 | 218 | # Add the flow to the original coordinates 219 | if isinstance(flow, torch.Tensor): 220 | flow = flow.detach().cpu().numpy() 221 | flow = flow.transpose(1, 2, 0) 222 | new_coords = np.subtract(coords, flow) 223 | 224 | 225 | # Map the new coordinates to the pixel values in the original image 226 | warped_image = cv2.remap(image, new_coords, None, interpolation=cv2.INTER_LINEAR, borderMode=cv2.BORDER_REFLECT) 227 | 228 | return warped_image 229 | 230 | 231 | def apply_flow_to_image_with_unused_mask(image, flow): 232 | """ 233 | Apply an optical flow tensor to a NumPy image by moving the pixels based on the flow and create a mask where the remap meant there was nothing there. 234 | 235 | Args: 236 | image (np.ndarray): Input image with shape (height, width, channels). 237 | flow (np.ndarray): Optical flow tensor with shape (height, width, 2). 238 | 239 | Returns: 240 | tuple: Warped image with the same shape as the input image, and a mask where the remap meant there was nothing there. 241 | """ 242 | height, width, _ = image.shape 243 | x_coords, y_coords = np.meshgrid(np.arange(width), np.arange(height)) 244 | coords = np.stack([x_coords, y_coords], axis=-1).astype(np.float32) 245 | 246 | # Add the flow to the original coordinates 247 | if isinstance(flow, torch.Tensor): 248 | flow = flow.detach().cpu().numpy() 249 | flow = flow.transpose(1, 2, 0) 250 | new_coords = np.subtract(coords, flow) 251 | avg = utilityb.avg_edge_pixels(image) 252 | warped_image = cv2.remap(image, new_coords, None, interpolation=cv2.INTER_LINEAR, borderMode=cv2.BORDER_REFLECT) 253 | 254 | # Create a mask where the remap meant there was nothing there 255 | mask = utilityb.create_hole_mask(flow) 256 | white_pixels = np.sum(mask > 0) 257 | #print(f'white pixels {white_pixels}') 258 | 259 | #remove later 260 | #warped_image = warp_image2(image,flow) 261 | 262 | return warped_image, mask,white_pixels 263 | 264 | def warp_image2(image, flow): 265 | h, w = image.shape[:2] 266 | flow_map = np.array([[x, y] for y in range(h) for x in range(w)], dtype=np.float32) - flow.reshape(-1, 2) 267 | flow_map = flow_map.reshape(h, w, 2).astype(np.float32) # Ensure the flow_map is in the correct format 268 | 269 | # Clip the flow_map to the image bounds 270 | flow_map[:, :, 0] = np.clip(flow_map[:, :, 0], 0, w - 1) 271 | flow_map[:, :, 1] = np.clip(flow_map[:, :, 1], 0, h - 1) 272 | 273 | warped_image = cv2.remap(image, flow_map, None, cv2.INTER_LINEAR, borderMode=cv2.BORDER_REFLECT, borderValue=(0, 0, 0) ) 274 | return warped_image -------------------------------------------------------------------------------- /scripts/optical_flow_simple.py: -------------------------------------------------------------------------------- 1 | import os 2 | import glob 3 | import cv2 4 | import numpy as np 5 | from PIL import Image 6 | import scripts.berry_utility as utilityb 7 | 8 | def read_image(file_path): 9 | image = cv2.imread(file_path, cv2.IMREAD_COLOR) 10 | if image is None: 11 | raise ValueError(f"Image '{file_path}' not found.") 12 | return image 13 | 14 | def save_image(image, file_path): 15 | cv2.imwrite(file_path, image) 16 | 17 | def resize_image(image, max_dimension): 18 | h, w = image.shape[:2] 19 | aspect_ratio = float(w) / float(h) 20 | if h > w: 21 | new_height = max_dimension 22 | new_width = int(new_height * aspect_ratio) 23 | else: 24 | new_width = max_dimension 25 | new_height = int(new_width / aspect_ratio) 26 | resized_image = cv2.resize(image, (new_width, new_height), interpolation=cv2.INTER_LINEAR) 27 | return resized_image 28 | 29 | def flow_to_color(flow, max_flow=None): 30 | hsv = np.zeros((flow.shape[0], flow.shape[1], 3), dtype=np.float32) 31 | hsv[..., 1] = 1.0 32 | 33 | mag, ang = cv2.cartToPolar(flow[..., 0], flow[..., 1]) 34 | hsv[..., 0] = ang * 180 / np.pi / 2 35 | if max_flow is not None: 36 | hsv[..., 2] = np.clip(mag / max_flow, 0, 1) 37 | else: 38 | hsv[..., 2] = np.clip(mag / (np.max(mag) + 1e-5), 0, 1) 39 | rgb = cv2.cvtColor(hsv, cv2.COLOR_HSV2BGR) 40 | return (rgb * 255).astype(np.uint8) 41 | 42 | def save_optical_flow(flow, file_path, max_flow=None): 43 | flow_color = flow_to_color(flow, max_flow) 44 | save_image(flow_color, file_path) 45 | 46 | 47 | def compute_optical_flow(image1, image2): 48 | flow = cv2.calcOpticalFlowFarneback(image1, image2, None, pyr_scale=0.5, levels=3, winsize=15, iterations=3, poly_n=5, poly_sigma=1.2, flags=0) 49 | return flow 50 | 51 | def warp_image(image, flow): 52 | h, w = image.shape[:2] 53 | flow_map = np.array([[x, y] for y in range(h) for x in range(w)], dtype=np.float32) - flow.reshape(-1, 2) 54 | flow_map = flow_map.reshape(h, w, 2).astype(np.float32) # Ensure the flow_map is in the correct format 55 | 56 | # Clip the flow_map to the image bounds 57 | flow_map[:, :, 0] = np.clip(flow_map[:, :, 0], 0, w - 1) 58 | flow_map[:, :, 1] = np.clip(flow_map[:, :, 1], 0, h - 1) 59 | 60 | warped_image = cv2.remap(image, flow_map, None, cv2.INTER_LANCZOS4 ) 61 | return warped_image 62 | 63 | def process_image_basic (image1_path, image2_path, provided_image_path,max_dimension, index,output_folder): 64 | 65 | #image1 = read_image(image1_path) 66 | # image2 = read_image(image2_path) 67 | 68 | # image1 = resize_image(image1, max_dimension) 69 | # image2 = resize_image(image2, max_dimension) 70 | image1 = resize_image(utilityb.base64_to_texture(image1_path),max_dimension) 71 | image2 = resize_image(utilityb.base64_to_texture(image2_path),max_dimension) 72 | 73 | # provided_image = read_image(provided_image_path) 74 | provided_image = utilityb.base64_to_texture(provided_image_path) 75 | provided_image = resize_image(provided_image, max_dimension) 76 | 77 | flow = compute_optical_flow(cv2.cvtColor(image1, cv2.COLOR_BGR2GRAY), cv2.cvtColor(image2, cv2.COLOR_BGR2GRAY)) 78 | warped_image = warp_image(provided_image, flow) 79 | 80 | warped_image_path = os.path.join(output_folder, f'warped_provided_image_{index + 1}.png') 81 | save_image(warped_image, warped_image_path) 82 | print(f"Warped image saved as '{warped_image_path}'") 83 | combine_images(image1,image2,provided_image,warped_image,f"{index}.png") 84 | return warped_image_path,flow 85 | # return provided_image_path,flow 86 | #AAAAAAAAAAAAAAAAAAA 87 | 88 | def process_image(image1, image2, provided_image, output_folder, flow_output_folder, max_dimension, index): 89 | image1 = resize_image(image1, max_dimension) 90 | image2 = resize_image(image2, max_dimension) 91 | 92 | flow = compute_optical_flow(cv2.cvtColor(image1, cv2.COLOR_BGR2GRAY), cv2.cvtColor(image2, cv2.COLOR_BGR2GRAY)) 93 | warped_image = warp_image(provided_image, flow) 94 | 95 | warped_image_path = os.path.join(output_folder, f'warped_provided_image_{index + 1}.png') 96 | save_image(warped_image, warped_image_path) 97 | print(f"Warped image saved as '{warped_image_path}'") 98 | 99 | flow_image_path = os.path.join(flow_output_folder, f'optical_flow_{index + 1}.png') 100 | save_optical_flow(flow, flow_image_path) 101 | print(f"Optical flow map saved as '{flow_image_path}'") 102 | 103 | return warped_image 104 | 105 | def process_images(input_folder, output_folder, flow_output_folder, provided_image_path, max_dimension): 106 | if not os.path.exists(output_folder): 107 | os.makedirs(output_folder) 108 | if not os.path.exists(flow_output_folder): 109 | os.makedirs(flow_output_folder) 110 | 111 | image_files = sorted(glob.glob(os.path.join(input_folder, '*.png'))) 112 | num_images = len(image_files) 113 | 114 | if num_images < 1: 115 | raise ValueError("At least one image is required to compute optical flow.") 116 | 117 | provided_image = read_image(provided_image_path) 118 | provided_image = resize_image(provided_image, max_dimension) 119 | 120 | for i in range(num_images - 1): 121 | image1 = read_image(image_files[i]) 122 | image2 = read_image(image_files[i + 1]) 123 | 124 | provided_image = process_image(image1, image2, provided_image, output_folder, flow_output_folder, max_dimension, i) 125 | 126 | def main(): 127 | input_folder = "Input_Images" 128 | output_folder = "output_images" 129 | flow_output_folder = "flow_output_images" 130 | provided_image_path = "init.png" 131 | max_dimension = 320 # Change this value to your preferred maximum dimension 132 | #process_images(input_folder, output_folder, flow_output_folder, provided_image_path, max_dimension) 133 | 134 | 135 | if __name__ == "__main__": 136 | main() 137 | 138 | def combine_images(img1, img2, img3, img4, output_file, output_folder="debug"): 139 | img1_pil = Image.fromarray(np.uint8(img1)) 140 | img2_pil = Image.fromarray(np.uint8(img2)) 141 | img3_pil = Image.fromarray(np.uint8(img3)) 142 | img4_pil = Image.fromarray(np.uint8(img4)) 143 | 144 | # Create output folder if it doesn't exist 145 | if not os.path.exists(output_folder): 146 | os.makedirs(output_folder) 147 | 148 | # Combine images horizontally 149 | combined_horizontal1 = Image.new('RGB', (img1_pil.width + img2_pil.width, img1_pil.height)) 150 | combined_horizontal1.paste(img1_pil, (0, 0)) 151 | combined_horizontal1.paste(img2_pil, (img1_pil.width, 0)) 152 | 153 | combined_horizontal2 = Image.new('RGB', (img3_pil.width + img4_pil.width, img3_pil.height)) 154 | combined_horizontal2.paste(img3_pil, (0, 0)) 155 | combined_horizontal2.paste(img4_pil, (img3_pil.width, 0)) 156 | 157 | # Combine images vertically 158 | combined_image = Image.new('RGB', (combined_horizontal1.width, combined_horizontal1.height + combined_horizontal2.height)) 159 | combined_image.paste(combined_horizontal1, (0, 0)) 160 | combined_image.paste(combined_horizontal2, (0, combined_horizontal1.height)) 161 | 162 | # Save combined image to output folder 163 | output_path = os.path.join(output_folder, output_file) 164 | combined_image.save(output_path) 165 | 166 | -------------------------------------------------------------------------------- /scripts/sd-TemporalKit-UI.py: -------------------------------------------------------------------------------- 1 | from __future__ import annotations 2 | import math 3 | import random 4 | import sys 5 | from argparse import ArgumentParser 6 | from collections import namedtuple, deque 7 | import einops 8 | import gradio as gr 9 | import numpy as np 10 | import torch 11 | import torch.nn as nn 12 | from einops import rearrange 13 | from omegaconf import OmegaConf 14 | from PIL import Image, ImageOps 15 | from torch import autocast 16 | import os 17 | import shutil 18 | import time 19 | import stat 20 | import gradio as gr 21 | import modules.extras 22 | from modules.ui_components import FormRow, FormGroup, ToolButton, FormHTML 23 | from modules.ui import create_sampler_and_steps_selection 24 | import json 25 | from modules.sd_samplers import samplers, samplers_for_img2img 26 | import re 27 | import modules.images as images 28 | from modules.call_queue import wrap_gradio_gpu_call, wrap_queued_call, wrap_gradio_call 29 | from modules import ui_extra_networks, devices, shared, scripts, script_callbacks, sd_hijack_unet, sd_hijack_utils 30 | from modules.shared import opts, cmd_opts, OptionInfo 31 | from pathlib import Path 32 | from typing import List, Tuple 33 | from PIL.ExifTags import TAGS 34 | from PIL.PngImagePlugin import PngImageFile, PngInfo 35 | from datetime import datetime 36 | from modules.generation_parameters_copypaste import quote 37 | from copy import deepcopy 38 | import platform 39 | import modules.generation_parameters_copypaste as parameters_copypaste 40 | import scripts.Berry_Method as General_SD 41 | import glob 42 | import base64 43 | import io 44 | import scripts.Ebsynth_Processing as ebsynth 45 | import scripts.berry_utility as sd_utility 46 | 47 | 48 | diffuseimg = None 49 | SamplerData = namedtuple('SamplerData', ['name', 'constructor', 'aliases', 'options']) 50 | lastmadefilename = "" 51 | def upload_file(files): 52 | file_paths = [file.name for file in files] 53 | return file_paths 54 | 55 | 56 | def preprocess_video(video,fps,batch_size,per_side,resolution,batch_run,max_frames,output_path,border_frames,ebsynth_mode,split_video,split_based_on_cuts): 57 | input_folder_loc = os.path.join(output_path, "input") 58 | output_folder_loc = os.path.join(output_path, "output") 59 | if not os.path.exists(input_folder_loc): 60 | os.makedirs(input_folder_loc) 61 | if not os.path.exists(output_folder_loc): 62 | os.makedirs(output_folder_loc) 63 | 64 | max_keys = max_frames 65 | if max_keys < 0: 66 | max_keys = 100000 67 | max_frames = (max_frames * (batch_size)) 68 | if max_frames < 1: 69 | max_frames = 100000 70 | # would use mathf.inf in c#, dunno what that is in python 71 | # potential bug later, low priority 72 | if ebsynth_mode == True: 73 | if split_video == False: 74 | border_frames = 0 75 | if batch_run == False: 76 | max_frames = per_side * per_side * (batch_size + 1) 77 | 78 | 79 | if split_video == True: 80 | 81 | #max frames only applies in batch mode 82 | #otherwise it limmits the number of *total frames* 83 | border_frames = border_frames * batch_size 84 | 85 | max_frames = (20 * batch_size) - border_frames 86 | max_total_frames = int((max_keys / 20) * max_frames) 87 | existing_frames = [] 88 | 89 | if split_based_on_cuts == True: 90 | existing_frames = sd_utility.split_video_into_numpy_arrays(video,fps,False) 91 | else: 92 | data = General_SD.convert_video_to_bytes(video) 93 | existing_frames = [sd_utility.extract_frames_movpie(data, fps,max_frames=max_total_frames,perform_interpolation=False)] 94 | 95 | 96 | split_video_paths,transition_data = General_SD.split_videos_into_smaller_videos(max_keys,existing_frames,fps,max_frames,output_path,border_frames,split_based_on_cuts) 97 | for index,individual_video in enumerate(split_video_paths): 98 | 99 | generated_textures = General_SD.generate_squares_to_folder(individual_video,fps=fps,batch_size=batch_size, resolution=resolution,size_size=per_side,max_frames=None, output_folder=os.path.dirname(individual_video),border=0, ebsynth_mode=ebsynth_mode,max_frames_to_save=max_frames) 100 | input_location = os.path.join(os.path.dirname(os.path.dirname(individual_video)),"input") 101 | for tex_index,texture in enumerate(generated_textures): 102 | individual_file_name = os.path.join(input_location,f"{index}and{tex_index}.png") 103 | General_SD.save_square_texture(texture,individual_file_name) 104 | transitiondatapath = os.path.join(output_path,"transition_data.txt") 105 | with open(transitiondatapath, "w") as f: 106 | f.write(str(transition_data) + "\n") 107 | f.write(str(border_frames) + "\n") 108 | main_video_path = os.path.join(output_path,"main_video.mp4") 109 | sd_utility.copy_video(video,main_video_path) 110 | return main_video_path 111 | 112 | new_video_loc = os.path.join(output_path, f"input_video.mp4") 113 | shutil.copyfile(video,new_video_loc) 114 | if ebsynth_mode == True: 115 | border = 0 116 | 117 | image = General_SD.generate_squares_to_folder(video,fps=fps,batch_size=batch_size, resolution=resolution,size_size=per_side,max_frames=max_frames, output_folder=output_path,border=border_frames, ebsynth_mode=True,max_frames_to_save=max_frames) 118 | return image[0] 119 | 120 | if batch_run == False: 121 | image = General_SD.generate_square_from_video(video,fps=fps,batch_size=batch_size, resolution=resolution,size_size=per_side ) 122 | processed = numpy_array_to_temp_url(image) 123 | else: 124 | image = General_SD.generate_squares_to_folder(video,fps=fps,batch_size=batch_size, resolution=resolution,size_size=per_side,max_frames=max_frames, output_folder=output_path,border=border_frames,ebsynth_mode=False,max_frames_to_save=max_frames) 125 | processed = image[0] 126 | return processed 127 | 128 | 129 | 130 | def apply_image_to_video(image,video,fps,per_side,output_resolution,batch_size): 131 | return General_SD.process_video_single(video_path=video,fps=fps,per_side=per_side,batch_size=batch_size,fillindenoise=0,edgedenoise=0,_smol_resolution=output_resolution,square_texture=image) 132 | 133 | def apply_image_to_vide_batch(input_folder,video,fps,per_side,output_resolution,batch_size,max_frames,border_frames): 134 | input_images_folder = os.path.join (input_folder,"output") 135 | images = read_images_folder(input_images_folder) 136 | print(len(images)) 137 | return General_SD.process_video_batch(video_path_old=video,fps=fps,per_side=per_side,batch_size=batch_size,fillindenoise=0,edgedenoise=0,_smol_resolution=output_resolution,square_textures=images,max_frames=max_frames,output_folder=input_folder,border=border_frames) 138 | 139 | def post_process_ebsynth(input_folder,video,fps,per_side,output_resolution,batch_size,max_frames,border_frames): 140 | input_images_folder = os.path.join (input_folder,"output") 141 | images = read_images_folder(input_images_folder) 142 | print(len(images)) 143 | split_mode = os.path.join(input_folder, "keys") 144 | if os.path.exists(split_mode): 145 | return ebsynth.sort_into_folders(video_path=video,fps=fps,per_side=per_side,batch_size=batch_size,_smol_resolution=output_resolution,square_textures=images,max_frames=max_frames,output_folder=input_folder,border=border_frames) 146 | else: 147 | img_folder = os.path.join(input_folder, "output") 148 | # define a regular expression pattern to match directory names with one or more digits 149 | pattern = r'^\d+$' 150 | 151 | # get a list of all directories in the specified path 152 | all_dirs = os.listdir(input_folder) 153 | 154 | # use a list comprehension to filter the directories based on the pattern 155 | numeric_dirs = sorted([d for d in all_dirs if re.match(pattern, d)], key=lambda x: int(x)) 156 | max_frames = max_frames + border_frames 157 | for d in numeric_dirs: 158 | # create a list to store the filenames of the images that match the directory name 159 | img_names = [] 160 | folder_video = os.path.join(input_folder, d, "input_video.mp4") 161 | # loop through each image file in the image folder 162 | for img_file in os.listdir(img_folder): 163 | # check if the image filename starts with the directory name followed by the word "and" and a sequence of one or more digits, then ends with '.png' 164 | if re.match(f"^{d}and\d+.*\.png$", img_file): 165 | img_names.append(img_file) 166 | print(f"post processing = {os.path.dirname(folder_video)}") 167 | square_textures = [] 168 | # loop through each image file name 169 | for img_name in sorted(img_names, key=lambda x: int(re.search(r'and(\d+)', x).group(1))): 170 | img = Image.open(os.path.join(input_images_folder, img_name)) 171 | # Convert image to NumPy array and append to images list 172 | print(f"saving {os.path.join(input_images_folder, img_name)}") 173 | square_textures.append(np.array(img)) 174 | 175 | ebsynth.sort_into_folders(video_path=folder_video, fps=fps, per_side=per_side, batch_size=batch_size, 176 | _smol_resolution=output_resolution, square_textures=square_textures, 177 | max_frames=max_frames, output_folder=os.path.dirname(folder_video), 178 | border=border_frames) 179 | 180 | def recombine_ebsynth(input_folder,fps,border_frames,batch): 181 | if os.path.exists(os.path.join(input_folder, "keys")): 182 | return ebsynth.crossfade_folder_of_folders(input_folder,fps=fps,return_generated_video_path=True) 183 | else: 184 | generated_videos = [] 185 | pattern = r'^\d+$' 186 | 187 | # get a list of all directories in the specified path 188 | all_dirs = os.listdir(input_folder) 189 | 190 | # use a list comprehension to filter the directories based on the pattern 191 | numeric_dirs = sorted([d for d in all_dirs if re.match(pattern, d)], key=lambda x: int(x)) 192 | 193 | for d in numeric_dirs: 194 | folder_loc = os.path.join(input_folder,d) 195 | # loop through each image file in the image folder 196 | new_video = ebsynth.crossfade_folder_of_folders(folder_loc,fps=fps) 197 | #print(f"generated new video at location {new_video}") 198 | generated_videos.append(new_video) 199 | 200 | overlap_data_path = os.path.join(input_folder,"transition_data.txt") 201 | with open(overlap_data_path, "r") as f: 202 | merge = str(f.readline().strip()) 203 | 204 | overlap_indicies = [] 205 | int_list = eval(merge) 206 | for num in int_list: 207 | overlap_indicies.append(int(num)) 208 | 209 | 210 | 211 | output_video = sd_utility.crossfade_videos(video_paths=generated_videos,fps=fps,overlap_indexes= overlap_indicies,num_overlap_frames= border_frames,output_path=os.path.join(input_folder,"output.mp4")) 212 | return output_video 213 | return None 214 | 215 | 216 | def atoi(text): 217 | return int(text) if text.isdigit() else text 218 | 219 | def natural_keys(text): 220 | return [atoi(c) for c in re.split(r'(\d+)', text)] 221 | 222 | def read_images_folder(folder_path): 223 | images = [] 224 | filenames = os.listdir(folder_path) 225 | 226 | # Sort filenames based on the order of the numbers in their names 227 | filenames.sort(key=natural_keys) 228 | 229 | for filename in filenames: 230 | # Check if file is an image (assumes only image files are in the folder) 231 | if (filename.endswith('.jpg') or filename.endswith('.png') or filename.endswith('.jpeg')) and (not re.search(r'-\d', filename)): 232 | if re.match(r".*(input).*", filename): 233 | # Open image using Pillow library 234 | 235 | img = Image.open(os.path.join(folder_path, filename)) 236 | # Convert image to NumPy array and append to images list 237 | images.append(np.array(img)) 238 | else: 239 | print(f"[${filename}] File name must contain \"input\". Skip processing.") 240 | return images 241 | 242 | 243 | 244 | def numpy_array_to_data_uri(img_array): 245 | # convert the array to an image using PIL 246 | img = Image.fromarray(img_array) 247 | 248 | # create a BytesIO object to hold the image data 249 | buffer = io.BytesIO() 250 | 251 | # save the image to the BytesIO object as PNG 252 | img.save(buffer, format='PNG') 253 | 254 | # get the PNG data from the BytesIO object 255 | png_data = buffer.getvalue() 256 | 257 | # convert the PNG data to base64-encoded string 258 | base64_str = base64.b64encode(png_data).decode() 259 | 260 | # combine the base64-encoded string with the image format prefix 261 | data_uri = 'data:image/png;base64,' + base64_str 262 | 263 | return data_uri 264 | 265 | def numpy_array_to_temp_url(img_array): 266 | # create a filename for the temporary file 267 | filename = 'generatedsquare.png' 268 | extension_path = os.path.abspath(__file__) 269 | extension_dir = os.path.dirname(os.path.dirname(extension_path)) 270 | extension_folder = os.path.join(extension_dir,"squares") 271 | if not os.path.exists(extension_folder): 272 | os.makedirs(extension_folder) 273 | # create a path for the temporary file 274 | file_path = os.path.join(extension_folder, filename) 275 | 276 | # convert the array to an image using PIL 277 | img = Image.fromarray(img_array) 278 | 279 | # save the image to the temporary file as PNG 280 | img.save(file_path, format='PNG') 281 | 282 | # create a URL for the temporary file 283 | #url = 'file://' + file_path 284 | 285 | return file_path 286 | 287 | def display_interface(interface): 288 | return interface.display() 289 | 290 | 291 | def get_most_recent_file(provided_directory): 292 | if not os.path.exists(provided_directory) or not os.path.isdir(provided_directory): 293 | raise ValueError("Invalid directory provided") 294 | 295 | # Get all files in the provided directory 296 | files = glob.glob(os.path.join(provided_directory, '*')) 297 | 298 | if not files: 299 | return None 300 | 301 | # Sort the files based on modification time and get the most recent one 302 | most_recent_file = max(files, key=os.path.getmtime) 303 | 304 | return most_recent_file 305 | 306 | def update_image(): 307 | global diffuseimg 308 | extension_path = os.path.abspath(__file__) 309 | # get the directory name of the extension 310 | extension_dir = os.path.dirname(os.path.dirname(extension_path)) 311 | extension_folder = os.path.join(extension_dir,"squares") 312 | most_recent_image = get_most_recent_file(extension_folder) 313 | print(most_recent_image) 314 | pilImage = Image.open(most_recent_image) 315 | print("running") 316 | return most_recent_image 317 | 318 | def update_settings(): 319 | extension_path = os.path.abspath(__file__) 320 | extension_dir = os.path.dirname(os.path.dirname(extension_path)) 321 | tempfile = os.path.join(extension_dir,"temp_file.txt") 322 | with open(tempfile, "r") as f: 323 | fps = int(f.readline().strip()) 324 | sides = int(f.readline().strip()) 325 | batch_size = int(f.readline().strip()) 326 | video_path = f.readline().strip() 327 | return fps,sides,batch_size,video_path 328 | 329 | def update_settings_from_file(folderpath): 330 | read_path = os.path.join(folderpath,"batch_settings.txt") 331 | border = None 332 | print (f"batch settings exists = {os.path.exists(read_path)}") 333 | if os.path.exists(read_path) == False: 334 | read_path = os.path.join(folderpath,"0/batch_settings.txt") 335 | video_path = os.path.join(folderpath,"main_video.mp4") 336 | transition_data_path = os.path.join(folderpath,"transition_data.txt") 337 | if os.path.exists(transition_data_path): 338 | with open(transition_data_path, "r") as b: 339 | merge = str(b.readline().strip()) 340 | border = int(b.readline().strip()) 341 | print (f"reading path at {read_path}") 342 | with open(read_path, "r") as f: 343 | fps = int(f.readline().strip()) 344 | sides = int(f.readline().strip()) 345 | batch_size = int(f.readline().strip()) 346 | video_path = f.readline().strip() 347 | max_frames = int(f.readline().strip()) 348 | if border == None: 349 | border = int(f.readline().strip()) 350 | 351 | 352 | return fps,sides,batch_size,video_path,max_frames,border 353 | 354 | 355 | def save_settings(fps,sides,batch_size,video): 356 | extension_path = os.path.abspath(__file__) 357 | extension_dir = os.path.dirname(os.path.dirname(extension_path)) 358 | tempfile = os.path.join(extension_dir,"temp_file.txt") 359 | with open(tempfile, "w") as f: 360 | f.write(str(fps) + "\n") 361 | f.write(str(sides) + "\n") 362 | f.write(str(batch_size) + "\n") 363 | f.write(str(video) + "\n") 364 | 365 | def create_video_Processing_Tab(): 366 | with gr.Column(visible=True, elem_id="Temporal_Kit") as main_panel: 367 | dummy_component = gr.Label(visible=False) 368 | with gr.Row(): 369 | with gr.Tabs(elem_id="mode_TemporalKit"): 370 | with gr.Row(): 371 | with gr.Tab(elem_id="input_TemporalKit", label="Input"): 372 | with gr.Row(): 373 | with gr.Column(): 374 | video = gr.Video(label="Input Video", elem_id="input_video",type="filepath") 375 | with gr.Row(): 376 | sides = gr.Number(value=2,label="Sides", precision=0, interactive=True) 377 | resolution = gr.Number(value=1024,label="Height Resolution", precision=1, interactive=True) 378 | with gr.Row(): 379 | batch_size = gr.Number(value=5, label="frames per keyframe", precision=1, interactive=True) 380 | fps = gr.Number(value=30, precision=1, label="fps", interactive=True) 381 | ebsynth_mode = gr.Checkbox(label="EBSynth Mode", value=False) 382 | with gr.Row(): 383 | savesettings = gr.Button("Save Settings") 384 | with gr.Row(): 385 | batch_folder = gr.Textbox(label="Target Folder",placeholder="This is ignored if neither batch run or ebsynth are checked") 386 | 387 | with gr.Row(): 388 | with gr.Accordion("Batch Settings",open=False): 389 | with gr.Row(): 390 | batch_checkbox = gr.Checkbox(label="Batch Run", value=False) 391 | max_keyframes = gr.Number(value=-1, label="Max key frames", precision=1, interactive=True,placeholder="for all frames") 392 | border_frames = gr.Number(value=2, label="Border Key Frames", precision=1, interactive=True,placeholder="border frames") 393 | with gr.Row(): 394 | with gr.Accordion("EBSynth Settings",open=False): 395 | with gr.Row(): 396 | split_video = gr.Checkbox(label="Split Video", value=False) 397 | split_based_on_cuts = gr.Checkbox(label="Split based on cuts (as well)", value=False) 398 | #interpolate = gr.Checkbox(label="Interpolate(high memory)", value=False) 399 | 400 | 401 | savesettings.click( 402 | fn=save_settings, 403 | inputs=[fps,sides,batch_size,video] 404 | ) 405 | with gr.Tabs(elemn_id="TemporalKit_gallery_container"): 406 | with gr.TabItem(elem_id="output_TemporalKit", label="Output"): 407 | with gr.Row(): 408 | result_image = gr.outputs.Image(type='pil') 409 | with gr.Row(): 410 | runbutton = gr.Button("Run") 411 | with gr.Row(): 412 | send_to_buttons = parameters_copypaste.create_buttons(["img2img"]) 413 | 414 | try: 415 | parameters_copypaste.bind_buttons(send_to_buttons, result_image, [""]) 416 | except: 417 | print("failed") 418 | pass 419 | parameters_copypaste.add_paste_fields("TemporalKit", result_image,None) 420 | runbutton.click(preprocess_video, [video,fps,batch_size,sides,resolution,batch_checkbox,max_keyframes,batch_folder,border_frames,ebsynth_mode,split_video,split_based_on_cuts], result_image) 421 | 422 | 423 | def show_textbox(option): 424 | if option == True: 425 | return gr.inputs.Textbox(lines=2, placeholder="Enter your text here") 426 | else: 427 | return False 428 | 429 | def create_diffusing_tab (): 430 | global diffuseimg 431 | with gr.Column(visible=True, elem_id="Processid") as second_panel: 432 | dummy_component = gr.Label(visible=False) 433 | with gr.Row(): 434 | with gr.Tabs(elem_id="mode_TemporalKit"): 435 | with gr.Row(): 436 | with gr.Tab(elem_id="input_diffuse", label="Generate"): 437 | with gr.Column(): 438 | with gr.Row(): 439 | input_image = gr.Image(label="Input_Image", elem_id="input_page2") 440 | input_video = gr.Video(label="Input Video", elem_id="input_videopage2") 441 | with gr.Row(): 442 | read_last_settings = gr.Button("read_last_settings", elem_id="read_last_settings") 443 | read_last_image = gr.Button("read_last_image", elem_id="read_last_image") 444 | with gr.Row(): 445 | fps = gr.Number(label="FPS",value=10,precision=1) 446 | per_side = gr.Number(label="per side",value=3,precision=1) 447 | output_resolution_single = gr.Number(label="output height resolution",value=1024,precision=1) 448 | batch_size_diffuse = gr.Number(label="batch size",value=10,precision=1) 449 | with gr.Row(): 450 | runButton = gr.Button("run", elem_id="run_button") 451 | 452 | 453 | with gr.Tabs(elem_id="mode_TemporalKit"): 454 | with gr.Row(): 455 | with gr.Tab(elem_id="input_diffuse", label="Output"): 456 | with gr.Column(): 457 | #newbutton = gr.Button("update", elem_id="update_button") 458 | outputfile = gr.Video() 459 | 460 | read_last_image.click( 461 | fn=update_image, 462 | outputs=input_image 463 | ) 464 | read_last_settings.click( 465 | fn=update_settings, 466 | outputs=[fps,per_side,batch_size_diffuse,input_video] 467 | ) 468 | runButton.click( 469 | fn=apply_image_to_video, 470 | inputs=[input_image, input_video,fps,per_side,output_resolution_single,batch_size_diffuse], 471 | outputs=outputfile 472 | ) 473 | 474 | 475 | def create_batch_tab (): 476 | with gr.Column(visible=True, elem_id="batch_process") as second_panel: 477 | with gr.Row(): 478 | with gr.Tabs(elem_id="mode_TemporalKit"): 479 | with gr.Row(): 480 | with gr.Tab(elem_id="input_diffuse", label="Generate Batch"): 481 | with gr.Column(): 482 | with gr.Row(): 483 | input_folder = gr.Textbox(label="Input Folder",placeholder="the whole folder, generated before, not just the output folder") 484 | input_video = gr.Video(label="Input Video", elem_id="input_videopage2") 485 | with gr.Row(): 486 | read_last_settings = gr.Button("read_last_settings", elem_id="read_last_settings") 487 | with gr.Row(): 488 | fps = gr.Number(label="FPS",value=10,precision=1) 489 | per_side = gr.Number(label="per side",value=3,precision=1) 490 | output_resolution_batch = gr.Number(label="output resolution",value=1024,precision=1) 491 | batch_size = gr.Number(label="batch size",value=5,precision=1) 492 | max_frames = gr.Number(label="max frames",value=100,precision=1) 493 | border_frames = gr.Number(label="border frames",value=1,precision=1) 494 | with gr.Row(): 495 | runButton = gr.Button("run", elem_id="run_button") 496 | 497 | 498 | 499 | with gr.Tabs(elem_id="mode_TemporalKit"): 500 | with gr.Row(): 501 | with gr.Tab(elem_id="input_diffuse", label="Output"): 502 | with gr.Column(): 503 | #newbutton = gr.Button("update", elem_id="update_button") 504 | outputfile = gr.Video() 505 | 506 | read_last_settings.click( 507 | fn=update_settings_from_file, 508 | inputs=[input_folder], 509 | outputs=[fps,per_side,batch_size,input_video,max_frames,border_frames] 510 | ) 511 | runButton.click( 512 | fn=apply_image_to_vide_batch, 513 | inputs=[input_folder,input_video,fps,per_side,output_resolution_batch,batch_size,max_frames,border_frames], 514 | outputs=outputfile 515 | ) 516 | 517 | 518 | def create_ebsynth_tab(): 519 | with gr.Column(visible=True, elem_id="batch_process") as second_panel: 520 | with gr.Row(): 521 | with gr.Tabs(elem_id="mode_TemporalKit"): 522 | with gr.Row(): 523 | with gr.Tab(elem_id="input_diffuse", label="Generate Batch"): 524 | with gr.Column(): 525 | with gr.Row(): 526 | input_folder = gr.Textbox(label="Input Folder",placeholder="the whole folder, generated before, not just the output folder") 527 | input_video = gr.Video(label="Input Video", elem_id="input_videopage2") 528 | with gr.Row(): 529 | read_last_settings_synth = gr.Button("read_last_settings", elem_id="read_last_settings") 530 | with gr.Row(): 531 | fps = gr.Number(label="FPS",value=10,precision=1) 532 | per_side = gr.Number(label="per side",value=3,precision=1) 533 | output_resolution_batch = gr.Number(label="output resolution",value=1024,precision=1) 534 | batch_size = gr.Number(label="batch size",value=5,precision=1) 535 | max_frames = gr.Number(label="max frames",value=100,precision=1) 536 | border_frames = gr.Number(value=1, label="Border Frames", precision=1, interactive=True,placeholder="border frames") 537 | with gr.Row(): 538 | runButton = gr.Button("prepare ebsynth", elem_id="run_button") 539 | recombineButton = gr.Button("recombine ebsynth", elem_id="recombine_button") 540 | with gr.Tabs(elem_id="mode_TemporalKit"): 541 | with gr.Row(): 542 | with gr.Tab(elem_id="input_diffuse", label="Output"): 543 | with gr.Column(): 544 | #newbutton = gr.Button("update", elem_id="update_button") 545 | outputvideo = gr.File() 546 | read_last_settings_synth.click( 547 | fn=update_settings_from_file, 548 | inputs=[input_folder], 549 | outputs=[fps,per_side,batch_size,input_video,max_frames,border_frames] 550 | ) 551 | runButton.click( 552 | fn=post_process_ebsynth, 553 | inputs=[input_folder,input_video,fps,per_side,output_resolution_batch,batch_size,max_frames,border_frames], 554 | outputs=outputvideo 555 | ) 556 | recombineButton.click( 557 | fn=recombine_ebsynth, 558 | inputs=[input_folder,fps,border_frames,batch_size], 559 | outputs=outputvideo 560 | ) 561 | tabs_list = ["TemporalKit"] 562 | 563 | def on_ui_tabs(): 564 | 565 | with gr.Blocks(analytics_enabled=False) as temporalkit: 566 | with gr.Tabs(elem_id="TemporalKit-Tab") as tabs: 567 | with gr.Tab(label="Pre-Processing"): 568 | with gr.Blocks(analytics_enabled=False): 569 | create_video_Processing_Tab() 570 | with gr.Tab(label="Temporal-Warp",elem_id="processbutton"): 571 | with gr.Blocks(analytics_enabled=False): 572 | create_diffusing_tab() 573 | with gr.Tab(label="Batch-Warp",elem_id="batch-button"): 574 | with gr.Blocks(analytics_enabled=False): 575 | create_batch_tab() 576 | with gr.Tab(label="Ebsynth-Process",elem_id="Ebsynth-Process"): 577 | with gr.Blocks(analytics_enabled=False): 578 | create_ebsynth_tab() 579 | return (temporalkit, "Temporal-Kit", "TemporalKit"), 580 | 581 | 582 | 583 | def generate( 584 | input_image: Image.Image, 585 | instruction: str, 586 | steps: int, 587 | randomize_seed: bool, 588 | seed: int, 589 | randomize_cfg: bool, 590 | text_cfg_scale: float, 591 | image_cfg_scale: float, 592 | negative_prompt: str, 593 | batch_number: int, 594 | scale: int, 595 | batch_in_check, 596 | batch_in_dir, 597 | sampler 598 | ): 599 | 600 | model = shared.sd_model 601 | model.eval().to(shared.device) 602 | 603 | animated_gifs = [] 604 | 605 | 606 | def on_ui_settings(): 607 | section = ('TemporalKit', "Temporal-Kit") 608 | shared.opts.add_option("def_img_cfg", shared.OptionInfo("1.5", "Default Image CFG", section=('ip2p', "Instruct-pix2pix"))) 609 | 610 | 611 | 612 | from fastapi import FastAPI, Body 613 | from base64 import b64decode, b64encode 614 | from io import BytesIO 615 | 616 | def img_to_b64(image: Image.Image): 617 | buf = BytesIO() 618 | image.save(buf, format="png") 619 | return b64encode(buf.getvalue()).decode("utf-8") 620 | 621 | def b64_to_img(enc: str): 622 | if enc.startswith('data:image'): 623 | enc = enc[enc.find(',')+1:] 624 | return Image.open(BytesIO(b64decode(enc))) 625 | 626 | 627 | 628 | 629 | script_callbacks.on_ui_settings(on_ui_settings) 630 | script_callbacks.on_ui_tabs(on_ui_tabs) 631 | -------------------------------------------------------------------------------- /scripts/stable_diffusion_processing.py: -------------------------------------------------------------------------------- 1 | import os 2 | import glob 3 | import requests 4 | import json 5 | from pprint import pprint 6 | import base64 7 | import numpy as np 8 | from io import BytesIO 9 | import scripts.optical_flow_simple as opflow 10 | from PIL import Image, ImageOps,ImageFilter 11 | import io 12 | from collections import deque 13 | import scripts.berry_utility as utilityb 14 | import cv2 15 | import scripts.optical_flow_raft as raft 16 | 17 | # Replace with the actual path to your image file and folder 18 | x_path = "./init.png" 19 | y_folder = "./Input_Images" 20 | temp_folder = "./temp" 21 | frame_count = 0 22 | 23 | img2imgurl = "http://localhost:7860/sdapi/v1/img2img" 24 | 25 | output_folder = "output" 26 | os.makedirs(output_folder, exist_ok=True) 27 | 28 | if not os.path.exists('intensitymaps'): 29 | os.makedirs('intensitymaps') 30 | 31 | if not os.path.exists('temp'): 32 | os.makedirs('temp') 33 | 34 | y_paths = utilityb.get_image_paths(y_folder) 35 | 36 | 37 | # get the initial image 38 | def square_Image_request (image_path,prompt,denoise_strength,resolution,seed): 39 | print(len(image_path),prompt,denoise_strength,resolution,seed) 40 | data = { 41 | "init_images": [image_path], 42 | "resize_mode": 0, 43 | "denoising_strength": denoise_strength, 44 | "prompt": prompt, 45 | "negative_prompt": "", 46 | #"control_net_enabled": "true", 47 | "alwayson_scripts": { 48 | "ControlNet":{ 49 | "args": [ 50 | { 51 | "input_image": image_path, 52 | "module": "hed", 53 | # "model": "control_canny-fp16 [e3fe7712]", 54 | "model": "control_hed-fp16 [13fee50b]", 55 | "processor_res": 1024, 56 | "weight": 1 57 | } 58 | ] 59 | } 60 | }, 61 | "seed": seed, 62 | "sampler_index": "Euler a", 63 | "batch_size": 1, 64 | "n_iter": 1, 65 | "steps": 20, 66 | "cfg_scale": 6, 67 | "width": resolution, 68 | "height": resolution, 69 | "restore_faces": True, 70 | "include_init_images": False, 71 | "override_settings": {}, 72 | "override_settings_restore_afterwards": True 73 | } 74 | response = requests.post(img2imgurl, json=data) 75 | #print (response.content) 76 | print(len(json.loads(response.content)["images"])) 77 | if response.status_code == 200: 78 | return json.loads(response.content)["images"][0] 79 | else: 80 | try: 81 | error_data = response.json() 82 | print("Error:") 83 | print(str(error_data)) 84 | 85 | except json.JSONDecodeError: 86 | error_data = response.content 87 | print(f"Error: Unable to parse JSON error data. {error_data}") 88 | return None 89 | 90 | 91 | def prepare_request(allpaths,index,last_stylized,resolution,seed,last_mask,last_last_mask,fillindenoise,edgedenoise,target,diffuse,forwards): 92 | 93 | warped_path,flow,unused_mask,whitepixels,flow_img = raft.apply_flow_based_on_images(allpaths[index - 1],allpaths[index],last_stylized,resolution,index,temp_folder) 94 | 95 | #warped_path,flow,unused_mask,whitepixels,flow_img = raft.apply_flow_based_on_images(allpaths[index - 1],allpaths[index],allpaths[index],resolution,index,temp_folder) 96 | if diffuse: 97 | hed = gethedfromsd(warped_path,resolution) 98 | hed = utilityb.mask_to_grayscale( utilityb.scale_mask_intensity(utilityb.base64_to_texture(hed),edgedenoise)) 99 | #hardened_hed = utilityb.harden_mask(hed,True) 100 | # flow_adjusted_mask = utilityb.modify_intensity_based_on_flow(hardened_hed,flow) 101 | if last_mask is None: 102 | last_mask = np.zeros((resolution,resolution)) 103 | if last_last_mask is None: 104 | last_last_mask = np.zeros((resolution,resolution)) 105 | if diffuse: 106 | combined_mask = utilityb.combine_masks([unused_mask,utilityb.scale_mask_intensity(last_mask,0.6),utilityb.scale_mask_intensity(last_last_mask,0.3),hed]) 107 | #combine_mask_no_hed = utilityb.combine_masks([unused_mask,utilityb.scale_mask_intensity(last_mask,0.6),utilityb.scale_mask_intensity(last_last_mask,0.3)]) 108 | #if whitepixels > 0: 109 | #replaced = utilityb.replace_masked_area(flow,index,warped_path,unused_mask,warped_path) 110 | if target is not None and index > 1: 111 | replaced = utilityb.replaced_mask_from_other_direction_debug(index,Image.open(warped_path).convert("RGBA"),unused_mask,flow,Image.fromarray(cv2.cvtColor(utilityb.base64_to_texture( target), cv2.COLOR_BGR2RGB)).convert("RGBA"),forwards) 112 | else: 113 | replaced = utilityb.replaced_mask_from_other_direction_debug(index,Image.open(warped_path).convert("RGBA"),unused_mask,flow,None) 114 | #else: 115 | # replaced = warped_path 116 | #replaced = replace_masked_area(flow,index,last_stylized,hed,allpaths[index]) 117 | 118 | 119 | # Save the mask as a PNG file in the temporary folder 120 | file_path = os.path.join("debug2", f'mask{index}.png') 121 | cv2.imwrite(file_path, unused_mask) 122 | 123 | 124 | 125 | if diffuse: 126 | return send_request_in_chain(last_stylized,allpaths[index],replaced,utilityb.texture_to_base64(combined_mask),index,seed,fillindenoise,resolution),unused_mask,flow_img 127 | else: 128 | return replaced, unused_mask,flow_img 129 | # return send_request(last_stylized,y_folder,allpaths[index],replaced,hed,index) 130 | 131 | 132 | # Send the request for hed map from the server based on a filepath 133 | def gethedfromsd(image_path,resolution): 134 | print(resolution) 135 | #with open(image_path, "rb") as f: 136 | # image = base64.b64encode(f.read()).decode("utf-8") 137 | image_smol = utilityb.resize_base64_image(image_path,resolution,resolution) 138 | url = "http://127.0.0.1:7860/controlnet/detect" 139 | data2 = { 140 | "controlnet_module": "hed", 141 | "controlnet_input_images": [image_smol], 142 | "controlnet_processor_res": resolution, 143 | } 144 | 145 | response = requests.post(url, json=data2) 146 | if response.status_code == 200: 147 | data = response.content 148 | loaded = json.loads(data) 149 | #print(response.content) 150 | encoded_image = loaded["images"][0] 151 | #print(f"encoded: {encoded_image}") 152 | 153 | return encoded_image 154 | elif response.status_code == 422: 155 | error = response.json() 156 | print("Validation error:", error) 157 | else: 158 | print("Unexpected error from hed:", response.status_code, response.text) 159 | 160 | 161 | 162 | 163 | 164 | #send based on current situation 165 | def send_request_in_chain(last_image_path,current_image_path,last_warped_path,mask,index,seed,fillindenoise,resolution): 166 | 167 | # with open(last_image_path, "rb") as f: 168 | # last_image = base64.b64encode(f.read()).decode("utf-8") 169 | # last_image = last_image_path 170 | #with open(current_image_path, "rb") as b: 171 | # current_image = base64.b64encode(b.read()).decode("utf-8") 172 | current_image = current_image_path 173 | 174 | if not last_warped_path == "": 175 | if os.path.isfile(last_warped_path): 176 | with open(last_warped_path, "rb") as c: 177 | last_warped = base64.b64encode(c.read()).decode("utf-8") 178 | else: 179 | last_warped = last_warped_path 180 | else: 181 | last_warped = current_image 182 | 183 | 184 | 185 | 186 | data = { 187 | "init_images": [last_warped], 188 | "inpainting_fill": 1, 189 | "inpaint_full_res": False, 190 | "inpaint_full_res_padding": 1, 191 | "inpainting_mask_invert": 0, 192 | "resize_mode": 0, 193 | "denoising_strength": fillindenoise, 194 | "prompt":"", 195 | "negative_prompt": "(ugly:1.3), (fused fingers), (too many fingers), (bad anatomy:1.5), (watermark:1.5), (words), letters, untracked eyes, asymmetric eyes, floating head, (logo:1.5), (bad hands:1.3), (mangled hands:1.2), (missing hands), (missing arms), backward hands, floating jewelry, unattached jewelry, floating head, doubled head, unattached head, doubled head, head in body, (misshapen body:1.1), (badly fitted headwear:1.2), floating arms, (too many arms:1.5), limbs fused with body, (facial blemish:1.5), badly fitted clothes, imperfect eyes, untracked eyes, crossed eyes, hair growing from clothes, partial faces, hair not attached to head", 196 | "alwayson_scripts": { 197 | "ControlNet":{ 198 | "args": [ 199 | { 200 | "input_image": current_image, 201 | "module": "hed", 202 | "model": "control_hed-fp16 [13fee50b]", 203 | "weight": 1, 204 | "guidance": 1, 205 | }, 206 | { 207 | "input_image": last_warped, 208 | "model": "diff_control_sd15_temporalnet_fp16 [adc6bd97]", 209 | "module": "none", 210 | "weight": 1, 211 | "guidance": 1, 212 | } 213 | 214 | ] 215 | } 216 | }, 217 | "seed": seed, 218 | "subseed": -1, 219 | "subseed_strength": -1, 220 | "sampler_index": "Euler a", 221 | "batch_size": 1, 222 | "n_iter": 1, 223 | "steps": 20, 224 | "cfg_scale": 6, 225 | "width": resolution, 226 | "height": resolution, 227 | "restore_faces": True, 228 | "include_init_images": True, 229 | "override_settings": {}, 230 | "override_settings_restore_afterwards": True 231 | } 232 | 233 | if not mask == "": 234 | data["mask"] = mask 235 | if not os.path.exists("./debug2/"): 236 | os.makedirs("./debug2/") 237 | with open(f"./debug2/{index}.png", "wb") as e: 238 | e.write(base64.b64decode(mask)) 239 | print(f"debug mask saved at ./debug2/{index}.png") 240 | else: 241 | data['denoising_strength'] = 0 242 | response = requests.post(img2imgurl, json=data) 243 | # print (response.content) 244 | print(response.status_code) 245 | if response.status_code == 200: 246 | return json.loads(response.content)["images"][0] 247 | else: 248 | try: 249 | error_data = response.json() 250 | print("Error:") 251 | print(str(error_data)) 252 | 253 | except json.JSONDecodeError: 254 | error_data = response.content 255 | print(f"Error: Unable to parse JSON error data. {error_data}") 256 | return None 257 | 258 | 259 | def batch_sd_run (y_paths, initial,count,seed,skip_first,fillindenoise,edgedenoise,smol_resolution,Forwards,target,diffuse): 260 | output_images = [] 261 | all_flow = [] 262 | output_images.append(initial) 263 | last_mask = None 264 | last_last_mask = None 265 | allpaths = y_paths 266 | for i in range(1, len(y_paths)): 267 | current_frame = count + i 268 | result,mask,flow = prepare_request(allpaths,i,output_images[i-1],smol_resolution,seed,last_mask,last_last_mask,fillindenoise,edgedenoise,target,diffuse,Forwards) 269 | all_flow.append(flow) 270 | output_images.append(result) 271 | print(f"Written data for frame {current_frame}:") 272 | last_last_mask = last_mask 273 | last_mask = mask 274 | if (skip_first == True): 275 | output_images.pop(0) 276 | 277 | 278 | 279 | return output_images,all_flow 280 | 281 | 282 | #output_images = [] 283 | #datanew = send_request_in_chain(y_paths[0], x_path,"","",0) 284 | #output_images.append(datanew) 285 | #output_paths = [] 286 | 287 | #for i in range(1, len(y_paths)): 288 | # frame_count = frame_count + 1 289 | # result_image = output_images[i-1] 290 | # temp_image_path = os.path.join(output_folder, f"temp_image_{i}.png") 291 | # data = json.loads(result_image) 292 | # encoded_image = data["images"][0] 293 | # with open(temp_image_path, "wb") as f: 294 | # f.write(base64.b64decode(encoded_image)) 295 | # output_paths.append(temp_image_path) 296 | # #result = send_request(temp_image_path, y_folder, y_paths[i]) 297 | # result = prepare_request(y_paths,i,temp_image_path,512) 298 | # output_images.append(result) 299 | # print(f"Written data for frame {i}:") 300 | 301 | 302 | -------------------------------------------------------------------------------- /style.css: -------------------------------------------------------------------------------- 1 | video.svelte-1vnmhm4 { 2 | max-height: 500px; 3 | } -------------------------------------------------------------------------------- /temp_file.txt: -------------------------------------------------------------------------------- 1 | 10 2 | 3 3 | 5 4 | C:\Users\crowl\AppData\Local\Temp\f9a5b76a71d3fb93d978662b3488fbd4a4fd00fc\006f4310-402eea6f.mp4 5 | --------------------------------------------------------------------------------