theluyuan/DiffSynth-Studio
1
0
Fork 0
mirror of https://github.com/modelscope/DiffSynth-Studio.git synced 2026-03-23 00:58:11 +00:00
Code Issues Packages Projects Releases Wiki Activity

DiffSynth-Studio 2.0 major update

Browse Source
This commit is contained in:
root
2025-12-04 16:33:07 +08:00
parent afd101f345
commit 72af7122b3
758 changed files with 26462 additions and 2221398 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
examples/wanvideo/model_training/special/direct_distill/Wan2.1-T2V-1.3B.sh Normal file
View File

@@ -0,0 +1,14 @@
accelerate launch --config_file examples/wanvideo/model_training/full/accelerate_config_14B.yaml examples/wanvideo/model_training/train.py \
--dataset_base_path data/example_video_dataset \
--dataset_metadata_path data/example_video_dataset/metadata_distill.csv \
--height 480 \
--width 832 \
--dataset_repeat 160 \
--model_id_with_origin_paths "Wan-AI/Wan2.1-T2V-1.3B:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-1.3B:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-1.3B:Wan2.1_VAE.pth" \
--learning_rate 1e-5 \
--num_epochs 2 \
--remove_prefix_in_ckpt "pipe.dit." \
--output_path "./models/train/Wan2.1-T2V-1.3B_full_distill" \
--trainable_models "dit" \
--task "direct_distill" \
--extra_inputs "seed,rand_device,num_inference_steps,cfg_scale"

23
examples/wanvideo/model_training/special/direct_distill/validate.py Normal file
View File

@@ -0,0 +1,23 @@
import torch
from PIL import Image
from diffsynth.utils.data import save_video, VideoData
from diffsynth.pipelines.wan_video import WanVideoPipeline, ModelConfig
pipe = WanVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig("models/train/Wan2.1-T2V-1.3B_full_distill/epoch-1.safetensors"),
ModelConfig(model_id="Wan-AI/Wan2.1-T2V-1.3B", origin_file_pattern="models_t5_umt5-xxl-enc-bf16.pth"),
ModelConfig(model_id="Wan-AI/Wan2.1-T2V-1.3B", origin_file_pattern="Wan2.1_VAE.pth"),
],
tokenizer_config=ModelConfig(model_id="Wan-AI/Wan2.1-T2V-1.3B", origin_file_pattern="google/umt5-xxl/"),
)
video = pipe(
prompt="纪实摄影风格画面，一只活泼的小狗在绿茵茵的草地上迅速奔跑。小狗毛色棕黄，两只耳朵立起，神情专注而欢快。阳光洒在它身上，使得毛发看上去格外柔软而闪亮。背景是一片开阔的草地，偶尔点缀着几朵野花，远处隐约可见蓝天和几片白云。透视感鲜明，捕捉小狗奔跑时的动感和四周草地的生机。中景侧面移动视角。",
cfg_scale=1, num_inference_steps=4,
seed=0, tiled=True,
)
save_video(video, "video_distill_Wan2.1-T2V-1.3B.mp4", fps=15, quality=5)

16
examples/wanvideo/model_training/special/fp8_training/Wan2.1-I2V-14B-480P.sh Normal file
View File

@@ -0,0 +1,16 @@
accelerate launch examples/wanvideo/model_training/train.py \
--dataset_base_path data/example_video_dataset \
--dataset_metadata_path data/example_video_dataset/metadata.csv \
--height 480 \
--width 832 \
--dataset_repeat 100 \
--model_id_with_origin_paths "Wan-AI/Wan2.1-I2V-14B-480P:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-I2V-14B-480P:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-I2V-14B-480P:Wan2.1_VAE.pth,Wan-AI/Wan2.1-I2V-14B-480P:models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth" \
--learning_rate 1e-4 \
--num_epochs 5 \
--remove_prefix_in_ckpt "pipe.dit." \
--output_path "./models/train/Wan2.1-I2V-14B-480P_lora_fp8" \
--lora_base_model "dit" \
--lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
--lora_rank 32 \
--extra_inputs "input_image" \
--fp8_models "Wan-AI/Wan2.1-I2V-14B-480P:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-I2V-14B-480P:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-I2V-14B-480P:Wan2.1_VAE.pth,Wan-AI/Wan2.1-I2V-14B-480P:models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth"

28
examples/wanvideo/model_training/special/fp8_training/validate.py Normal file
View File

@@ -0,0 +1,28 @@
import torch
from PIL import Image
from diffsynth.utils.data import save_video, VideoData
from diffsynth.pipelines.wan_video import WanVideoPipeline, ModelConfig
from modelscope import dataset_snapshot_download
pipe = WanVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="Wan-AI/Wan2.1-I2V-14B-480P", origin_file_pattern="diffusion_pytorch_model*.safetensors"),
ModelConfig(model_id="Wan-AI/Wan2.1-I2V-14B-480P", origin_file_pattern="models_t5_umt5-xxl-enc-bf16.pth"),
ModelConfig(model_id="Wan-AI/Wan2.1-I2V-14B-480P", origin_file_pattern="Wan2.1_VAE.pth"),
ModelConfig(model_id="Wan-AI/Wan2.1-I2V-14B-480P", origin_file_pattern="models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth"),
],
)
pipe.load_lora(pipe.dit, "models/train/Wan2.1-I2V-14B-480P_lora_fp8/epoch-4.safetensors", alpha=1)
input_image = VideoData("data/example_video_dataset/video1.mp4", height=480, width=832)[0]
video = pipe(
prompt="from sunset to night, a small town, light, house, river",
negative_prompt="色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走",
input_image=input_image,
seed=1, tiled=True
)
save_video(video, "video_Wan2.1-I2V-14B-480P.mp4", fps=15, quality=5)

38
examples/wanvideo/model_training/special/low_vram_training/Wan2.1-I2V-14B-480P.sh Normal file
View File

@@ -0,0 +1,38 @@
accelerate launch examples/wanvideo/model_training/train.py \
--dataset_base_path data/example_video_dataset \
--dataset_metadata_path data/example_video_dataset/metadata.csv \
--height 480 \
--width 832 \
--dataset_repeat 1 \
--model_id_with_origin_paths "Wan-AI/Wan2.1-I2V-14B-480P:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-I2V-14B-480P:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-I2V-14B-480P:Wan2.1_VAE.pth,Wan-AI/Wan2.1-I2V-14B-480P:models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth" \
--learning_rate 1e-4 \
--num_epochs 5 \
--remove_prefix_in_ckpt "pipe.dit." \
--output_path "./models/train/Wan2.1-I2V-14B-480P_lora_lowvram_cache" \
--lora_base_model "dit" \
--lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
--lora_rank 32 \
--extra_inputs "input_image" \
--task "sft:data_process" \
--offload_models "Wan-AI/Wan2.1-I2V-14B-480P:diffusion_pytorch_model*.safetensors" \
--fp8_models "Wan-AI/Wan2.1-I2V-14B-480P:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-I2V-14B-480P:Wan2.1_VAE.pth,Wan-AI/Wan2.1-I2V-14B-480P:models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth" \
--use_gradient_checkpointing_offload
accelerate launch examples/wanvideo/model_training/train.py \
--dataset_base_path "./models/train/Wan2.1-I2V-14B-480P_lora_split_cache" \
--height 480 \
--width 832 \
--dataset_repeat 100 \
--model_id_with_origin_paths "Wan-AI/Wan2.1-I2V-14B-480P:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-I2V-14B-480P:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-I2V-14B-480P:Wan2.1_VAE.pth,Wan-AI/Wan2.1-I2V-14B-480P:models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth" \
--learning_rate 1e-4 \
--num_epochs 5 \
--remove_prefix_in_ckpt "pipe.dit." \
--output_path "./models/train/Wan2.1-I2V-14B-480P_lora_lowvram" \
--lora_base_model "dit" \
--lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
--lora_rank 32 \
--extra_inputs "input_image" \
--task "sft:train" \
--offload_models "Wan-AI/Wan2.1-I2V-14B-480P:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-I2V-14B-480P:Wan2.1_VAE.pth,Wan-AI/Wan2.1-I2V-14B-480P:models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth" \
--fp8_models "Wan-AI/Wan2.1-I2V-14B-480P:diffusion_pytorch_model*.safetensors" \
--use_gradient_checkpointing_offload

28
examples/wanvideo/model_training/special/low_vram_training/validate.py Normal file
View File

@@ -0,0 +1,28 @@
import torch
from PIL import Image
from diffsynth.utils.data import save_video, VideoData
from diffsynth.pipelines.wan_video import WanVideoPipeline, ModelConfig
from modelscope import dataset_snapshot_download
pipe = WanVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="Wan-AI/Wan2.1-I2V-14B-480P", origin_file_pattern="diffusion_pytorch_model*.safetensors"),
ModelConfig(model_id="Wan-AI/Wan2.1-I2V-14B-480P", origin_file_pattern="models_t5_umt5-xxl-enc-bf16.pth"),
ModelConfig(model_id="Wan-AI/Wan2.1-I2V-14B-480P", origin_file_pattern="Wan2.1_VAE.pth"),
ModelConfig(model_id="Wan-AI/Wan2.1-I2V-14B-480P", origin_file_pattern="models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth"),
],
)
pipe.load_lora(pipe.dit, "models/train/Wan2.1-I2V-14B-480P_lora_lowvram/epoch-4.safetensors", alpha=1)
input_image = VideoData("data/example_video_dataset/video1.mp4", height=480, width=832)[0]
video = pipe(
prompt="from sunset to night, a small town, light, house, river",
negative_prompt="色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走",
input_image=input_image,
seed=1, tiled=True
)
save_video(video, "video_Wan2.1-I2V-14B-480P.mp4", fps=15, quality=5)

34
examples/wanvideo/model_training/special/split_training/Wan2.1-I2V-14B-480P.sh Normal file
View File

@@ -0,0 +1,34 @@
accelerate launch examples/wanvideo/model_training/train.py \
--dataset_base_path data/example_video_dataset \
--dataset_metadata_path data/example_video_dataset/metadata.csv \
--height 480 \
--width 832 \
--dataset_repeat 1 \
--model_id_with_origin_paths "Wan-AI/Wan2.1-I2V-14B-480P:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-I2V-14B-480P:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-I2V-14B-480P:Wan2.1_VAE.pth,Wan-AI/Wan2.1-I2V-14B-480P:models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth" \
--learning_rate 1e-4 \
--num_epochs 5 \
--remove_prefix_in_ckpt "pipe.dit." \
--output_path "./models/train/Wan2.1-I2V-14B-480P_lora_split_cache" \
--lora_base_model "dit" \
--lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
--lora_rank 32 \
--extra_inputs "input_image" \
--task "sft:data_process" \
--offload_models "Wan-AI/Wan2.1-I2V-14B-480P:diffusion_pytorch_model*.safetensors"
accelerate launch examples/wanvideo/model_training/train.py \
--dataset_base_path "./models/train/Wan2.1-I2V-14B-480P_lora_split_cache" \
--height 480 \
--width 832 \
--dataset_repeat 100 \
--model_id_with_origin_paths "Wan-AI/Wan2.1-I2V-14B-480P:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-I2V-14B-480P:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-I2V-14B-480P:Wan2.1_VAE.pth,Wan-AI/Wan2.1-I2V-14B-480P:models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth" \
--learning_rate 1e-4 \
--num_epochs 5 \
--remove_prefix_in_ckpt "pipe.dit." \
--output_path "./models/train/Wan2.1-I2V-14B-480P_lora_split" \
--lora_base_model "dit" \
--lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
--lora_rank 32 \
--extra_inputs "input_image" \
--task "sft:train" \
--offload_models "Wan-AI/Wan2.1-I2V-14B-480P:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-I2V-14B-480P:Wan2.1_VAE.pth,Wan-AI/Wan2.1-I2V-14B-480P:models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth"

28
examples/wanvideo/model_training/special/split_training/validate.py Normal file
View File

@@ -0,0 +1,28 @@
import torch
from PIL import Image
from diffsynth.utils.data import save_video, VideoData
from diffsynth.pipelines.wan_video import WanVideoPipeline, ModelConfig
from modelscope import dataset_snapshot_download
pipe = WanVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="Wan-AI/Wan2.1-I2V-14B-480P", origin_file_pattern="diffusion_pytorch_model*.safetensors"),
ModelConfig(model_id="Wan-AI/Wan2.1-I2V-14B-480P", origin_file_pattern="models_t5_umt5-xxl-enc-bf16.pth"),
ModelConfig(model_id="Wan-AI/Wan2.1-I2V-14B-480P", origin_file_pattern="Wan2.1_VAE.pth"),
ModelConfig(model_id="Wan-AI/Wan2.1-I2V-14B-480P", origin_file_pattern="models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth"),
],
)
pipe.load_lora(pipe.dit, "models/train/Wan2.1-I2V-14B-480P_lora_split/epoch-4.safetensors", alpha=1)
input_image = VideoData("data/example_video_dataset/video1.mp4", height=480, width=832)[0]
video = pipe(
prompt="from sunset to night, a small town, light, house, river",
negative_prompt="色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走",
input_image=input_image,
seed=1, tiled=True
)
save_video(video, "video_Wan2.1-I2V-14B-480P.mp4", fps=15, quality=5)