From c0b589d934e4abf7959e3731bbbe69f41c4175df Mon Sep 17 00:00:00 2001 From: lzws <2538048363@qq.com> Date: Mon, 22 Sep 2025 01:57:05 +0800 Subject: [PATCH] add wan2.2-VACE-Fun infereance and trining --- .../model_inference/Wan2.2-VACE-Fun-A14B.py | 54 +++++++++++++++++++ .../full/Wan2.2-VACE-Fun-A14B.sh | 40 ++++++++++++++ .../lora/Wan2.2-VACE-Fun-A14B.sh | 43 +++++++++++++++ .../validate_full/Wan2.2-VACE-Fun-A14B.py | 33 ++++++++++++ .../Wan2.2-VACE-Fun-A14B-lora.py | 31 +++++++++++ 5 files changed, 201 insertions(+) create mode 100644 examples/wanvideo/model_inference/Wan2.2-VACE-Fun-A14B.py create mode 100644 examples/wanvideo/model_training/full/Wan2.2-VACE-Fun-A14B.sh create mode 100644 examples/wanvideo/model_training/lora/Wan2.2-VACE-Fun-A14B.sh create mode 100644 examples/wanvideo/model_training/validate_full/Wan2.2-VACE-Fun-A14B.py create mode 100644 examples/wanvideo/model_training/validate_lora/Wan2.2-VACE-Fun-A14B-lora.py diff --git a/examples/wanvideo/model_inference/Wan2.2-VACE-Fun-A14B.py b/examples/wanvideo/model_inference/Wan2.2-VACE-Fun-A14B.py new file mode 100644 index 0000000..a768192 --- /dev/null +++ b/examples/wanvideo/model_inference/Wan2.2-VACE-Fun-A14B.py @@ -0,0 +1,54 @@ +import torch +from PIL import Image +from diffsynth import save_video, VideoData +from diffsynth.pipelines.wan_video_new import WanVideoPipeline, ModelConfig +from modelscope import dataset_snapshot_download + + +pipe = WanVideoPipeline.from_pretrained( + torch_dtype=torch.bfloat16, + device="cuda", + model_configs=[ + ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="high_noise_model/diffusion_pytorch_model*.safetensors", offload_device="cpu"), + ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="low_noise_model/diffusion_pytorch_model*.safetensors", offload_device="cpu"), + ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="models_t5_umt5-xxl-enc-bf16.pth", offload_device="cpu"), + ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="Wan2.1_VAE.pth", offload_device="cpu"), + ], +) +pipe.enable_vram_management() + + +dataset_snapshot_download( + dataset_id="DiffSynth-Studio/examples_in_diffsynth", + local_dir="./", + allow_file_pattern=["data/examples/wan/depth_video.mp4", "data/examples/wan/cat_fightning.jpg"] +) + +# Depth video -> Video +control_video = VideoData("data/examples/wan/depth_video.mp4", height=480, width=832) +video = pipe( + prompt="两只可爱的橘猫戴上拳击手套，站在一个拳击台上搏斗。", + 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", + vace_video=control_video, + seed=1, tiled=True +) +save_video(video, "video1_14b.mp4", fps=15, quality=5) + +# Reference image -> Video +video = pipe( + prompt="两只可爱的橘猫戴上拳击手套，站在一个拳击台上搏斗。", + 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", + vace_reference_image=Image.open("data/examples/wan/cat_fightning.jpg").resize((832, 480)), + seed=1, tiled=True +) +save_video(video, "video2_14b.mp4", fps=15, quality=5) + +# Depth video + Reference image -> Video +video = pipe( + prompt="两只可爱的橘猫戴上拳击手套，站在一个拳击台上搏斗。", + 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", + vace_video=control_video, + vace_reference_image=Image.open("data/examples/wan/cat_fightning.jpg").resize((832, 480)), + seed=1, tiled=True +) +save_video(video, "video3_14b.mp4", fps=15, quality=5) diff --git a/examples/wanvideo/model_training/full/Wan2.2-VACE-Fun-A14B.sh b/examples/wanvideo/model_training/full/Wan2.2-VACE-Fun-A14B.sh new file mode 100644 index 0000000..0ee97da --- /dev/null +++ b/examples/wanvideo/model_training/full/Wan2.2-VACE-Fun-A14B.sh @@ -0,0 +1,40 @@ +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_vace.csv \ + --data_file_keys "video,vace_video,vace_reference_image" \ + --height 480 \ + --width 832 \ + --num_frames 17 \ + --dataset_repeat 100 \ + --model_id_with_origin_paths "PAI/Wan2.2-VACE-Fun-A14B:high_noise_model/diffusion_pytorch_model*.safetensors,PAI/Wan2.2-VACE-Fun-A14B:models_t5_umt5-xxl-enc-bf16.pth,PAI/Wan2.2-VACE-Fun-A14B:Wan2.1_VAE.pth" \ + --learning_rate 1e-4 \ + --num_epochs 2 \ + --remove_prefix_in_ckpt "pipe.vace." \ + --output_path "./models/train/Wan2.2-VACE-Fun-A14B_high_noise_full" \ + --trainable_models "vace" \ + --extra_inputs "vace_video,vace_reference_image" \ + --use_gradient_checkpointing_offload \ + --max_timestep_boundary 0.358 \ + --min_timestep_boundary 0 +# boundary corresponds to timesteps [900, 1000] + + +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_vace.csv \ + --data_file_keys "video,vace_video,vace_reference_image" \ + --height 480 \ + --width 832 \ + --num_frames 17 \ + --dataset_repeat 100 \ + --model_id_with_origin_paths "PAI/Wan2.2-VACE-Fun-A14B:low_noise_model/diffusion_pytorch_model*.safetensors,PAI/Wan2.2-VACE-Fun-A14B:models_t5_umt5-xxl-enc-bf16.pth,PAI/Wan2.2-VACE-Fun-A14B:Wan2.1_VAE.pth" \ + --learning_rate 1e-4 \ + --num_epochs 2 \ + --remove_prefix_in_ckpt "pipe.vace." \ + --output_path "./models/train/Wan2.2-VACE-Fun-A14B_low_noise_full" \ + --trainable_models "vace" \ + --extra_inputs "vace_video,vace_reference_image" \ + --use_gradient_checkpointing_offload \ + --max_timestep_boundary 1 \ + --min_timestep_boundary 0.358 +# boundary corresponds to timesteps [0, 900] \ No newline at end of file diff --git a/examples/wanvideo/model_training/lora/Wan2.2-VACE-Fun-A14B.sh b/examples/wanvideo/model_training/lora/Wan2.2-VACE-Fun-A14B.sh new file mode 100644 index 0000000..93b38cf --- /dev/null +++ b/examples/wanvideo/model_training/lora/Wan2.2-VACE-Fun-A14B.sh @@ -0,0 +1,43 @@ +accelerate launch examples/wanvideo/model_training/train.py \ + --dataset_base_path data/example_video_dataset \ + --dataset_metadata_path data/example_video_dataset/metadata_vace.csv \ + --data_file_keys "video,vace_video,vace_reference_image" \ + --height 480 \ + --width 832 \ + --num_frames 17 \ + --dataset_repeat 100 \ + --model_id_with_origin_paths "PAI/Wan2.2-VACE-Fun-A14B:high_noise_model/diffusion_pytorch_model*.safetensors,PAI/Wan2.2-VACE-Fun-A14B:models_t5_umt5-xxl-enc-bf16.pth,PAI/Wan2.2-VACE-Fun-A14B:Wan2.1_VAE.pth" \ + --learning_rate 1e-4 \ + --num_epochs 5 \ + --remove_prefix_in_ckpt "pipe.vace." \ + --output_path "./models/train/Wan2.2-VACE-Fun-A14B_high_noise_lora" \ + --lora_base_model "vace" \ + --lora_target_modules "q,k,v,o,ffn.0,ffn.2" \ + --lora_rank 32 \ + --extra_inputs "vace_video,vace_reference_image" \ + --use_gradient_checkpointing_offload \ + --max_timestep_boundary 0.358 \ + --min_timestep_boundary 0 +# boundary corresponds to timesteps [900, 1000] + +accelerate launch examples/wanvideo/model_training/train.py \ + --dataset_base_path data/example_video_dataset \ + --dataset_metadata_path data/example_video_dataset/metadata_vace.csv \ + --data_file_keys "video,vace_video,vace_reference_image" \ + --height 480 \ + --width 832 \ + --num_frames 17 \ + --dataset_repeat 100 \ + --model_id_with_origin_paths "PAI/Wan2.2-VACE-Fun-A14B:low_noise_model/diffusion_pytorch_model*.safetensors,PAI/Wan2.2-VACE-Fun-A14B:models_t5_umt5-xxl-enc-bf16.pth,PAI/Wan2.2-VACE-Fun-A14B:Wan2.1_VAE.pth" \ + --learning_rate 1e-4 \ + --num_epochs 5 \ + --remove_prefix_in_ckpt "pipe.vace." \ + --output_path "./models/train/Wan2.2-VACE-Fun-A14B_low_noise_lora" \ + --lora_base_model "vace" \ + --lora_target_modules "q,k,v,o,ffn.0,ffn.2" \ + --lora_rank 32 \ + --extra_inputs "vace_video,vace_reference_image" \ + --use_gradient_checkpointing_offload \ + --max_timestep_boundary 1 \ + --min_timestep_boundary 0.358 +# boundary corresponds to timesteps [0, 900] \ No newline at end of file diff --git a/examples/wanvideo/model_training/validate_full/Wan2.2-VACE-Fun-A14B.py b/examples/wanvideo/model_training/validate_full/Wan2.2-VACE-Fun-A14B.py new file mode 100644 index 0000000..e566dba --- /dev/null +++ b/examples/wanvideo/model_training/validate_full/Wan2.2-VACE-Fun-A14B.py @@ -0,0 +1,33 @@ +import torch +from PIL import Image +from diffsynth import save_video, VideoData, load_state_dict +from diffsynth.pipelines.wan_video_new import WanVideoPipeline, ModelConfig + + +pipe = WanVideoPipeline.from_pretrained( + torch_dtype=torch.bfloat16, + device="cuda", + model_configs=[ + ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="high_noise_model/diffusion_pytorch_model*.safetensors", offload_device="cpu"), + ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="low_noise_model/diffusion_pytorch_model*.safetensors", offload_device="cpu"), + ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="models_t5_umt5-xxl-enc-bf16.pth", offload_device="cpu"), + ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="Wan2.1_VAE.pth", offload_device="cpu"), + ], +) +state_dict = load_state_dict("models/train/Wan2.2-VACE-Fun-A14B_high_noise_full/epoch-1.safetensors") +pipe.vace.load_state_dict(state_dict) +state_dict = load_state_dict("models/train/Wan2.2-VACE-Fun-A14B_low_noise_full/epoch-1.safetensors") +pipe.vace2.load_state_dict(state_dict) +pipe.enable_vram_management() + +video = VideoData("data/example_video_dataset/video1_softedge.mp4", height=480, width=832) +video = [video[i] for i in range(17)] +reference_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", + vace_video=video, vace_reference_image=reference_image, num_frames=17, + seed=1, tiled=True +) +save_video(video, "video_Wan2.2-VACE-A14B.mp4", fps=15, quality=5) diff --git a/examples/wanvideo/model_training/validate_lora/Wan2.2-VACE-Fun-A14B-lora.py b/examples/wanvideo/model_training/validate_lora/Wan2.2-VACE-Fun-A14B-lora.py new file mode 100644 index 0000000..b6e6aff --- /dev/null +++ b/examples/wanvideo/model_training/validate_lora/Wan2.2-VACE-Fun-A14B-lora.py @@ -0,0 +1,31 @@ +import torch +from PIL import Image +from diffsynth import save_video, VideoData +from diffsynth.pipelines.wan_video_new import WanVideoPipeline, ModelConfig + + +pipe = WanVideoPipeline.from_pretrained( + torch_dtype=torch.bfloat16, + device="cuda", + model_configs=[ + ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="high_noise_model/diffusion_pytorch_model*.safetensors", offload_device="cpu"), + ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="low_noise_model/diffusion_pytorch_model*.safetensors", offload_device="cpu"), + ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="models_t5_umt5-xxl-enc-bf16.pth", offload_device="cpu"), + ModelConfig(model_id="PAI/Wan2.2-VACE-Fun-A14B", origin_file_pattern="Wan2.1_VAE.pth", offload_device="cpu"), + ], +) +pipe.load_lora(pipe.vace, "models/train/Wan2.2-VACE-Fun-A14B_high_noise_lora/epoch-4.safetensors", alpha=1) +pipe.load_lora(pipe.vace2, "models/train/Wan2.2-VACE-Fun-A14B_low_noise_lora/epoch-4.safetensors", alpha=1) +pipe.enable_vram_management() + +video = VideoData("data/example_video_dataset/video1_softedge.mp4", height=480, width=832) +video = [video[i] for i in range(17)] +reference_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", + vace_video=video, vace_reference_image=reference_image, num_frames=17, + seed=1, tiled=True +) +save_video(video, "video_Wan2.2-VACE-Fun-A14B.mp4", fps=15, quality=5)