Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-17 16:24:33 +02:00
Code Issues Packages Projects Releases Wiki Activity

fix wan and add latte

Browse Source 
Signed-off-by: Vladimir Mandic <mandic00@live.com>
This commit is contained in:
Vladimir Mandic
2025-03-21 09:39:00 -04:00
parent e5f6250153
commit b2432db88e
7 changed files with 42 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files
CHANGELOG.md
+3
View File
@@ -12,6 +12,8 @@
- Video: HunyuanVideo-I2V incompatible with latest transformers <https://github.com/huggingface/diffusers/issues/11118>
- Video: LTXVideo-095 support for conditioned input
- Video: LTXVideo-095 support for offloading
- Video: FasterCache: https://github.com/huggingface/diffusers/pull/10163
- Video: PyramidAttention: https://github.com/huggingface/diffusers/pull/9562
### Highlights for 2025-03-20
@@ -31,6 +33,7 @@ Plus support for CogView-4, new CLiP models, improvements to remote VAE, additio
- [CogVideoX](https://huggingface.co/THUDM/CogVideoX-5b): *2B, 5B* | *T2V, I2V*
- [Allegro](https://huggingface.co/rhymes-ai/Allegro): *T2V*
- [Mochi1](https://huggingface.co/genmo/mochi-1-preview): *T2V*
- [Latte1](https://huggingface.co/maxin-cn/Latte-1): *T2V
- decoding:
- **Default**: use vae from model
- **Tiny VAE**: support for *Hunyuan, WAN, Mochi*
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 = '56f740051dae2d410677292a5c9e5b66e60f87dc' # diffusers commit hash
sha = '844221ae4e20a8939ee052f75874e284f75d4c5c' # 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/ui_video.py
+24 -20
View File
@@ -26,7 +26,6 @@ def model_change(engine, model):
sd_models.unload_model_weights()
msg = 'Video model unloaded'
return [msg,
gr.update(visible='I2V' in selected.name) if selected else gr.update(visible=False),
video_utils.get_url(selected.url if selected else None),
]
@@ -84,24 +83,29 @@ def create_ui():
seed = gr.Number(label='Initial seed', value=-1, elem_id="video_seed", container=True)
random_seed = ToolButton(ui_symbols.random, elem_id="video_random_seed", label='Random seed')
reuse_seed = ToolButton(ui_symbols.reuse, elem_id="video_reuse_seed", label='Reuse seed')
steps, sampler_index = ui_sections.create_sampler_and_steps_selection(None, "video")
with gr.Row():
sampler_shift = gr.Slider(label='Sampler shift', minimum=0.0, maximum=20.0, step=0.1, value=7.0, elem_id="video_scheduler_shift")
dynamic_shift = gr.Checkbox(label='Dynamic shift', value=False, elem_id="video_dynamic_shift", interactive=False) # TODO video: dynamic shift
with gr.Row():
guidance_scale = gr.Slider(label='Guidance scale', minimum=0.0, maximum=14.0, step=0.1, value=6.0, elem_id="video_guidance_scale")
guidance_true = gr.Slider(label='True guidance', minimum=0.0, maximum=14.0, step=0.1, value=1.0, elem_id="video_guidance_true")
with gr.Row():
vae_type = gr.Dropdown(label='VAE decode', choices=['Default', 'Tiny', 'Remote'], value='Default', elem_id="video_vae_type")
vae_tile_frames = gr.Slider(label='Tile frames', minimum=1, maximum=64, step=1, value=16, elem_id="video_vae_tile_frames")
with gr.Row():
with gr.Group(visible=False, elem_id='video_init_image') as image_group:
gr.HTML("<br>&nbsp Init image")
init_image = gr.Image(elem_id="video_image", show_label=False, type="pil", image_mode="RGB", height=512)
with gr.Row():
save_frames = gr.Checkbox(label='Save image frames', value=False, elem_id="video_save_frames")
with gr.Row():
video_type, video_duration, video_loop, video_pad, video_interpolate = ui_sections.create_video_inputs(tab='video')
with gr.Accordion(open=True, label="Parameters", elem_id='video_parameters_accordion'):
steps, sampler_index = ui_sections.create_sampler_and_steps_selection(None, "video")
with gr.Row():
sampler_shift = gr.Slider(label='Sampler shift', minimum=0.0, maximum=20.0, step=0.1, value=7.0, elem_id="video_scheduler_shift")
dynamic_shift = gr.Checkbox(label='Dynamic shift', value=False, elem_id="video_dynamic_shift", interactive=False) # TODO video: dynamic shift
with gr.Row():
guidance_scale = gr.Slider(label='Guidance scale', minimum=0.0, maximum=14.0, step=0.1, value=6.0, elem_id="video_guidance_scale")
guidance_true = gr.Slider(label='True guidance', minimum=0.0, maximum=14.0, step=0.1, value=1.0, elem_id="video_guidance_true")
with gr.Accordion(open=True, label="Decode", elem_id='video_decode_accordion'):
with gr.Row():
vae_type = gr.Dropdown(label='VAE decode', choices=['Default', 'Tiny', 'Remote'], value='Default', elem_id="video_vae_type")
vae_tile_frames = gr.Slider(label='Tile frames', minimum=1, maximum=64, step=1, value=16, elem_id="video_vae_tile_frames")
with gr.Accordion(open=False, label="Init image", elem_id='video_init_accordion'):
gr.HTML("<br>&nbsp Init image")
init_image = gr.Image(elem_id="video_image", show_label=False, type="pil", image_mode="RGB", height=512)
with gr.Accordion(open=False, label="Accelerate", elem_id='video_accelerate_accordion'):
faster_cache = gr.Checkbox(label='FasterCache', value=False, elem_id="video_faster_cache")
pyramid_attention = gr.Checkbox(label='PyramidAttention', value=False, elem_id="video_pyramid_attention")
with gr.Accordion(open=True, label="Output", elem_id='video_output_accordion'):
with gr.Row():
save_frames = gr.Checkbox(label='Save image frames', value=False, elem_id="video_save_frames")
with gr.Row():
video_type, video_duration, video_loop, video_pad, video_interpolate = ui_sections.create_video_inputs(tab='video')
override_settings = ui_common.create_override_inputs('video')
# output panel with gallery and video tabs
@@ -117,7 +121,7 @@ def create_ui():
random_seed.click(fn=lambda: -1, show_progress=False, inputs=[], outputs=[seed])
# handle engine and model change
engine.change(fn=engine_change, inputs=[engine], outputs=[model])
model.change(fn=model_change, inputs=[engine, model], outputs=[html_log, image_group, url])
model.change(fn=model_change, inputs=[engine, model], outputs=[html_log, url])
# setup extra networks
ui_extra_networks.setup_ui(extra_networks_ui, gallery)
modules/video_models/models_def.py
+9
View File
@@ -139,6 +139,15 @@ models = {
te_cls=transformers.T5EncoderModel,
dit_cls=diffusers.MochiTransformer3DModel),
],
'Latte Video': [
Model(name='None'),
Model(name='Latte 1 T2V',
url='https://huggingface.co/maxin-cn/Latte-1',
repo='maxin-cn/Latte-1',
repo_cls=diffusers.LattePipeline,
te_cls=transformers.T5EncoderModel,
dit_cls=diffusers.LatteTransformer3DModel),
],
'Allegro Video': [
Model(name='None'),
Model(name='Allegro T2V',
modules/video_models/run_wan.py
+2 -2
View File
@@ -13,9 +13,9 @@ def generate(*args, **kwargs):
return video_utils.queue_err('model not selected')
found = [model.name for model in models_def.models.get(engine, [])]
selected: models_def.Model = [m for m in models_def.models[engine] if m.name == model][0] if len(found) > 0 else None
if not shared.sd_loaded or 'WAN' not in shared.sd_model.__class__.__name__:
if not shared.sd_loaded or 'Wan' not in shared.sd_model.__class__.__name__:
video_load.load_model(selected)
if not shared.sd_loaded or 'WAN' not in shared.sd_model.__class__.__name__:
if not shared.sd_loaded or 'Wan' not in shared.sd_model.__class__.__name__:
return video_utils.queue_err('model not loaded')
debug(f'Video generate: task={task_id} args={args} kwargs={kwargs}')
modules/video_models/video_load.py
+2 -1
View File
@@ -73,7 +73,8 @@ def load_model(selected: models_def.Model):
if selected.te_hijack:
shared.sd_model.orig_encode_prompt = shared.sd_model.encode_prompt
shared.sd_model.encode_prompt = video_utils.hijack_encode_prompt
shared.sd_model.vae.enable_slicing()
if hasattr(shared.sd_model.vae, 'enable_slicing'):
shared.sd_model.vae.enable_slicing()
loaded_model = selected.name
msg = f'Video load: cls={shared.sd_model.__class__.__name__} model="{selected.name}" time={t1-t0:.2f}'
shared.log.info(msg)
wiki
+1 -1
Submodule wiki updated: d50882dcb8...3f46b4f742