Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 09:14:35 +02:00
Code Issues Packages Projects Releases Wiki Activity

add wan21-flv2v-720p

Browse Source 
Signed-off-by: Vladimir Mandic <mandic00@live.com>
This commit is contained in:
Vladimir Mandic
2025-04-18 17:42:36 -04:00
parent d3203253ad
commit 5347b5701d
5 changed files with 23 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
CHANGELOG.md
+2
View File
@@ -8,6 +8,8 @@
see [Nunchaku Wiki](https://github.com/vladmandic/sdnext/wiki/Nunchaku) for installation guide and list of supported models & features
- [LTXVideo 0.9.6](https://github.com/Lightricks/LTX-Video?tab=readme-ov-file) T2V and I2V
in both **Standard** and **Distilled** variants
- [WAN 2.1 FLF2V 14B 720P](https://huggingface.co/Wan-AI/Wan2.1-FLF2V-14B-720P)
new first-to-last image video model from WAN-AI
- [CFG-Zero](https://github.com/WeichenFan/CFG-Zero-star) new guidance method optimized for flow-matching models
implemented for **FLUX.1, HiDream-I1, SD3.x, CogView4, HunyuanVideo, WanAI**
enable and configure in *settings -> pipeline modifiers -> cfg zero*
modules/ui_video.py
+4 -2
View File
@@ -114,9 +114,11 @@ def create_ui():
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'):
init_strength = gr.Slider(label='Init strength', minimum=0.0, maximum=1.0, step=0.01, value=0.5, elem_id="video_denoising_strength")
gr.HTML("<br>&nbsp Init image")
init_image = gr.Image(elem_id="video_image", show_label=False, type="pil", image_mode="RGB", height=512)
init_strength = gr.Slider(label='Init strength', minimum=0.0, maximum=1.0, step=0.01, value=0.5, elem_id="video_denoising_strength")
gr.HTML("<br>&nbsp Last image")
last_image = gr.Image(elem_id="video_last", show_label=False, type="pil", image_mode="RGB", height=512)
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")
@@ -167,7 +169,7 @@ def create_ui():
sampler_shift, dynamic_shift,
seed,
guidance_scale, guidance_true,
init_image, init_strength,
init_image, init_strength, last_image,
vae_type, vae_tile_frames,
save_frames,
video_type, video_duration, video_loop, video_pad, video_interpolate,
modules/video_models/models_def.py
+6
View File
@@ -158,6 +158,12 @@ models = {
repo_cls=diffusers.WanImageToVideoPipeline,
te_cls=transformers.T5EncoderModel,
dit_cls=diffusers.WanTransformer3DModel),
Model(name='WAN 2.1 14B FLF2V 720p',
url='https://huggingface.co/Wan-AI/Wan2.1-FLF2V-14B-720P',
repo='Wan-AI/Wan2.1-FLF2V-14B-720P-diffusers',
repo_cls=diffusers.WanImageToVideoPipeline,
te_cls=transformers.T5EncoderModel,
dit_cls=diffusers.WanTransformer3DModel),
],
'Mochi Video': [
Model(name='None'),
modules/video_models/video_overrides.py
+1 -1
View File
@@ -14,7 +14,7 @@ def load_override(selected: Model):
kwargs['vae'] = diffusers.AutoencoderKLAllegro.from_pretrained(selected.repo, subfolder="vae", torch_dtype=torch.float32, cache_dir=shared.opts.hfcache_dir)
if selected.name == 'LTXVideo 0.9.5 I2V':
kwargs['vae'] = diffusers.AutoencoderKLLTXVideo.from_pretrained(selected.repo, subfolder="vae", torch_dtype=torch.float32, cache_dir=shared.opts.hfcache_dir)
if selected.name == 'WAN 2.1 14B I2V 480p' or selected.name == 'WAN 2.1 14B I2V 720p':
if 'WAN 2.1 14B' in selected.name:
kwargs['vae'] = diffusers.AutoencoderKLWan.from_pretrained(selected.repo, subfolder="vae", torch_dtype=torch.float32, cache_dir=shared.opts.hfcache_dir)
debug(f'Video overrides: model="{selected.name}" kwargs={list(kwargs)}')
return kwargs
modules/video_models/video_run.py
+10 -2
View File
@@ -8,7 +8,7 @@ debug = shared.log.trace if os.environ.get('SD_VIDEO_DEBUG', None) is not None e
def generate(*args, **kwargs):
task_id, ui_state, engine, model, prompt, negative, styles, width, height, frames, steps, sampler_index, sampler_shift, dynamic_shift, seed, guidance_scale, guidance_true, init_image, init_strength, vae_type, vae_tile_frames, save_frames, video_type, video_duration, video_loop, video_pad, video_interpolate, override_settings = args
task_id, ui_state, engine, model, prompt, negative, styles, width, height, frames, steps, sampler_index, sampler_shift, dynamic_shift, seed, guidance_scale, guidance_true, init_image, init_strength, last_image, vae_type, vae_tile_frames, save_frames, video_type, video_duration, video_loop, video_pad, video_interpolate, override_settings = args
if engine is None or model is None or engine == 'None' or model == 'None':
return video_utils.queue_err('model not selected')
found = [model.name for model in models_def.models.get(engine, [])]
@@ -57,9 +57,17 @@ def generate(*args, **kwargs):
return video_utils.queue_err('init image not set')
p.task_args['image'] = images.resize_image(resize_mode=2, im=init_image, width=p.width, height=p.height, upscaler_name=None, output_type='pil')
shared.log.debug(f'Video: op=I2V init={init_image} resized={p.task_args["image"]}')
elif 'FLF2V' in model:
if init_image is None:
return video_utils.queue_err('init image not set')
if last_image is None:
return video_utils.queue_err('last image not set')
p.task_args['image'] = images.resize_image(resize_mode=2, im=init_image, width=p.width, height=p.height, upscaler_name=None, output_type='pil')
p.task_args['last_image'] = images.resize_image(resize_mode=2, im=last_image, width=p.width, height=p.height, upscaler_name=None, output_type='pil')
shared.log.debug(f'Video: op=FLF2V init={init_image} last={last_image} resized={p.task_args["image"]}')
elif 'T2V' in model:
if init_image is not None:
shared.log.debug('Video: op=T2V init image not supported')
shared.log.warning('Video: op=T2V init image not supported')
# cleanup memory
shared.sd_model = sd_models.apply_balanced_offload(shared.sd_model)