Commit Graph

17 Commits (794875cadb9ccf6c245e4b0d3fcb517bf2cc4e1e)

Author SHA1 Message Date
Jabasukuriputo Wang bac47592e5
oops (#331) 2023-11-20 20:57:03 -06:00
Jabasukuriputo Wang 4a788668d0
memory efficient attention fallthrough (#330) 2023-11-20 16:38:20 -06:00
Chengsong Zhang 4cd270d556
1.12.1 (#321)
* recovery for i2i-b

* return process result

* support fp8

* simplified fp8

* cast norm

* cn fp8

* improve restore

* Dev add s3 (#315)

* add file output to  s3

* test

* 1

* test

* 11

* add loggger

* 修改描述

* add log

* log formate

* install edit

* 临时修改验证

* remove install

* modify import

---------

Co-authored-by: zhangruicheng <zhangruicheng@migu.cn>

* LCM Sampler (#318)

* lcm, not tested

* lcm appears at UI

* refactor, not working properly

* lcm

* readme

* 1.12.1

---------

Co-authored-by: zhangrc <zrc_java@163.com>
Co-authored-by: zhangruicheng <zhangruicheng@migu.cn>
2023-11-16 20:08:47 -06:00
Chengsong Zhang 7a66b35e28
1.12.0 (#300)
* init

* adxl

* Update README.md

* Update README.md

* Update README.md

* Update README.md
2023-11-10 01:38:01 -06:00
Chengsong Zhang 4a5461ee5a
1.11.0 (#260)
* partially finish hotshotxl, have to deal with another bug

* unknown bug

* remove print

* fix mm

* remove beta, prev

* basic use supported. CN and prompt travel still on the way

* uncomment all control models for sdxl

* prompt travel

* readme

* readme
2023-10-29 06:15:59 -05:00
Chengsong Zhang ea3e6232a5
1.9.3 (#92)
* QOL features and settings, backend cleanup for ease of maintenance, minor bug fixes (#87)

* QOL update

-Set mm_sd_v14 as the default instead of v15, as most end users tend to prefer it
-Load actual motion modules from the model folder instead of hardcoding v14 and v15 in the dropdown
-Add support for webp and apng file formats
-Add quality sliders and lossless options for webp and apng
-Add ability to format the file names
-Add ability to output to a custom folder, along with named subdirectories similar to the feature set for images

* Parity update, UX polish

-Brought injection code up to parity with Kosinkadink's ComfyUI's implementation
-Updated motion_module
-Updated groupnorm hack
-Removed apng option, it just doesn't work well in imageio's implementation, may return in future
-Add option to always save videos
-Add infotext to video generations
-Set UI parameters from infotext when pasting/restoring from params.txt
-Separated settings into separate py file
-Updated info for options
-made some function names more descriptive and moved some code around for readbility
-Make it so the infotext shown underneath the video after a generation reflects the actual video, not just main prompt

* Restore Timestep after video generation

-Fixes an issue that would occur when trying to generate images after generating video. No longer asserts.

* Unet script cleanup

* Groupnorm hack setting polish, fix video duration

-Don't hack groupnorm when not using a v1 module
-Enforce original groupnorm when not requested to be hacked. this avoid getting into a bad state when the groupnorm hack setting is changed mid-generation.
-Fix video duration
-Adjust spacing in unet_injection script

* Update README.md

* init

* don't change anything. Features will be added in future commits

* don't change anything. Features will be added in future commits

* don't change anything. Features will be added in future commits

* readme

* Add WebP output with infotext stored in exif and options for quality and lossless (#233)

Co-authored-by: zappityzap <you@example.com>

* Bug, pass -overwrite_original to exiftool (#234)

* fix (#227)

* pass -overwrite_original to exiftool

---------

Co-authored-by: Chengsong Zhang <continuerevolution@gmail.com>
Co-authored-by: zappityzap <you@example.com>

* Update README.md

* Update README.md

* Update animatediff_ui.py

* date

---------

Co-authored-by: JaredTherriault <noirjt@live.com>
Co-authored-by: zappityzap <128872140+zappityzap@users.noreply.github.com>
Co-authored-by: zappityzap <you@example.com>
2023-10-19 22:42:26 -05:00
Alex Pinilla 552f505997
[FIX] motion_module.py, error: missing encoder_hidden_states (#205)
* [FIX] motion_module.py, error: missing encoder_hidden_states

I fixed this error:
```
TypeError: VanillaTemporalModule.forward() missing 1 required positional argument: 'encoder_hidden_states'
```
With this fix this error won't happen again.

* edit

---------

Co-authored-by: Chengsong Zhang <continuerevolution@gmail.com>
2023-10-11 14:51:57 -05:00
Chengsong Zhang b4a84dbe47
not batch cond uncond (#162)
* start to support prompt travel

* these changes might be aborted

* maybe batch cond uncond has been supported

* forgot to modify one

* readme

* remove unused import

* remove another import
2023-10-01 13:42:14 -05:00
Chengsong Zhang 934e59c4b8
1.5.2 (#111)
* convert params dict to AnimateDiffProcess for API (#110)

* Add GIF palette optimization (#104)

* move gifsicle to settings

* re-format with isort

* xformers

* fix radio

* readme

---------

Co-authored-by: zappityzap <128872140+zappityzap@users.noreply.github.com>
Co-authored-by: rkfg <rkfg@rkfg.me>
2023-09-21 16:48:28 -05:00
Chengsong Zhang 918afd2682 attn opt 2023-09-19 01:17:07 -05:00
Chengsong Zhang 37810ffbbe add mid block removal logic 2023-09-11 19:49:45 -05:00
Chengsong Zhang b88d4ac533 add mid block insertion 2023-09-11 19:41:28 -05:00
Chengsong Zhang 757143c010 add hash for v2 and temporaldiff 2023-09-11 19:10:05 -05:00
Chengsong Zhang 287b30afc3 fix more bugs 2023-07-25 00:07:19 +08:00
Chengsong Zhang 08a40868f3 fix insertion problem 2023-07-24 22:49:35 +08:00
Chengsong Zhang 8cc867f20a finish v1.1.0, remove a bunch of features 2023-07-20 10:54:02 +08:00
Chengsong Zhang e1e37654bf finish 2023-07-18 10:58:59 +08:00