Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 01:04:32 +02:00
Code Issues Packages Projects Releases Wiki Activity

update video notes and todo

Browse Source 
Signed-off-by: Vladimir Mandic <mandic00@live.com>
This commit is contained in:
Vladimir Mandic
2025-03-22 09:26:03 -04:00
parent 8fdcc5711c
commit 85bda171f4
5 changed files with 30 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files
CHANGELOG.md
+3 -19
View File
@@ -1,33 +1,17 @@
# Change Log for SD.Next
## Update for 2025-03-21
## Update for 2025-03-22
### ToDo/Limitations
- VLM Gemma3: requires `transformers==git+https://github.com/huggingface/transformers@v4.49.0-Gemma-3`
- VAE Remote encode: SD15 and Flux.1 issues: <https://github.com/huggingface/diffusers/issues/11069>
- Video: API support is TBD
- Video: Wiki page is TBD
- Video: LTXVideo-095 params
- Video: LTXVideo-095 support for conditioned input
- Video: LTXVideo-095 support for offloading
- Video: HunyuanVideo-I2V incompatible with latest transformers <https://github.com/huggingface/diffusers/issues/11118>
- Video: FasterCache and PyramidAttentionBroadcast granular config
- Video: FasterCache and PyramidAttentionBroadcast for LTX and WAN <https://github.com/huggingface/diffusers/issues/11134>
- Video: HunyuanVideo-I2V-16/33ch <https://github.com/huggingface/diffusers/pull/11066>
- Video: CogVideo-15 support
### Highlights for 2025-03-21
### Highlights for 2025-03-22
Brand new Video processing module with support for all latest models: **WAN21, Hunyuan, LTX, Cog, Allegro, Mochi1, Latte1**
And combined with *on-the-fly quantization*, support for *Local/Tiny/Remote* VAE, acceleration modules such as *FasterCache or PAB* and more!
Plus support for CogView-4, new CLiP models, improvements to remote VAE, additional docs/guides
### Details for 2025-03-21
### Details for 2025-03-22
- **Video tab**
- initial release so consider this as alpha version
- new top-level tab, replaces previous *video* script in text/image tabs
old scripts are still present, but will be removed in the future
- support for all latest models:
TODO.md
+18
View File
@@ -4,6 +4,24 @@ Main ToDo list can be found at [GitHub projects](https://github.com/users/vladma
## Current
### Issues/Limitations
- VLM Gemma3: requires `transformers==git+https://github.com/huggingface/transformers@v4.49.0-Gemma-3`
- VAE Remote encode: SD15 and Flux.1 issues: <https://github.com/huggingface/diffusers/issues/11069>
- Video: API support is TBD
- Video: Hunyuan Video I2V: transformers incompatibility <https://github.com/huggingface/diffusers/issues/11118>
- Video: Hunyuan Video I2V: 16ch vs 33ch processing <https://github.com/huggingface/diffusers/pull/11066>
- Video: WAN 2.1 14B I2V 480p: broken offload
- Video: WAN 2.1 14B I2V 720p: broken offload
- Video: CogVideoX 1.5 5B T2V/I2V: requires pipeline update
- Video: CogVideoX 1.5 5B I2V: requires pipeline update
- Video: LTXVideo 0.9.5 T2V/I2V: broken offload, new pipeline
- Video: LTXVideo 0.9.5 T2V/I2V: set preset params
- Video: LTXVideo 0.9.5 T2V/I2V: support for conditioned input
- Video: LTXVideo 0.9.1 I2V: generator list mismatch
- Video: FasterCache and PyramidAttentionBroadcast granular config
- Video: FasterCache and PyramidAttentionBroadcast for LTX and WAN <https://github.com/huggingface/diffusers/issues/11134>
## Future Candidates
- Redesign postprocessing
installer.py
+1 -1
View File
@@ -538,7 +538,7 @@ def check_diffusers():
t_start = time.time()
if args.skip_all or args.skip_git:
return
sha = '844221ae4e20a8939ee052f75874e284f75d4c5c' # diffusers commit hash
sha = 'a7d53a59394d5d8367826663601b69828e9f74fc' # diffusers commit hash
pkg = pkg_resources.working_set.by_key.get('diffusers', None)
minor = int(pkg.version.split('.')[1] if pkg is not None else 0)
cur = opts.get('diffusers_version', '') if minor > 0 else ''
modules/video_models/models_def.py
+7 -7
View File
@@ -12,23 +12,23 @@ import transformers
- SkyReels Hunyuan I2V: pass/pass/pass
- Fast Hunyuan T2V: pass/pass/pass
- LTXVideo 0.9.5 T2V: pass/pass/fail, v095 pipeline is tbd
- LTXVideo 0.9.5 I2V: pass/pass/fail, v095 pipeline is tbd
- LTXVideo 0.9.5 T2V: pass/pass/fail, completely broken offload, new pipeline
- LTXVideo 0.9.5 I2V: pass/pass/fail, completely broken offload, new pipeline
- LTXVideo 0.9.1 T2V: pass/pass/pass
- LTXVideo 0.9.1 I2V: pass/pass/pass
- LTXVideo 0.9.1 I2V: pass/pass/fail, generator list mismatch
- LTXVideo 0.9.0 T2V: pass/pass/pass
- LTXVideo 0.9.0 I2V: pass/pass/pass
- WAN 2.1 1.3B T2V: pass/pass/pass
- WAN 2.1 14B T2V: pass/pass/pass
- WAN 2.1 14B I2V 480p: pass/pass/tbd
- WAN 2.1 14B I2V 720p: pass/pass/tbd
- WAN 2.1 14B I2V 480p: pass/pass/fail, offloading cpu vs cuda
- WAN 2.1 14B I2V 720p: pass/pass/fail, offloading cpu vs cuda
- CogVideoX 1.0 2B T2V: pass/pass/pass
- CogVideoX 1.0 5B T2V: pass/pass/pass
- CogVideoX 1.0 5B I2V: pass/pass/pass
- CogVideoX 1.5 5B T2V: download/load/fail, v15 pipeline is tbd
- CogVideoX 1.5 5B I2V: download/load/fail, v15 pipeline is tbd
- CogVideoX 1.5 5B T2V: download/load/fail, pipeline is tbd
- CogVideoX 1.5 5B I2V: download/load/fail, pipeline is tbd
- Mochi 1 T2V: pass/pass/pass
- Latte 1 T2V: pass/pass/fail, float vs bfloat during generate
wiki
+1 -1
Submodule wiki updated: 3f46b4f742...00145b30c5