Merge pull request #1006 from lzws/main

add wan2.2-S2V-14B training
2026-03-18 22:08:13 +00:00 · 2025-10-22 09:55:21 +08:00
parent 0a1c172a00 30292d9411
commit 538017177a
12 changed files with 162 additions and 10 deletions
--- a/README.md
+++ b/README.md
@@ -208,7 +208,7 @@ save_video(video, "video1.mp4", fps=15, quality=5)
 | Model ID | Extra Parameters | Inference | Full Training | Validate After Full Training | LoRA Training | Validate After LoRA Training |
 |-|-|-|-|-|-|-|
 |[Wan-AI/Wan2.2-Animate-14B](https://www.modelscope.cn/models/Wan-AI/Wan2.2-Animate-14B)|`input_image`, `animate_pose_video`, `animate_face_video`, `animate_inpaint_video`, `animate_mask_video`|[code](./examples/wanvideo/model_inference/Wan2.2-Animate-14B.py)|[code](./examples/wanvideo/model_training/full/Wan2.2-Animate-14B.sh)|[code](./examples/wanvideo/model_training/validate_full/Wan2.2-Animate-14B.py)|[code](./examples/wanvideo/model_training/lora/Wan2.2-Animate-14B.sh)|[code](./examples/wanvideo/model_training/validate_lora/Wan2.2-Animate-14B.py)|
-|[Wan-AI/Wan2.2-S2V-14B](https://www.modelscope.cn/models/Wan-AI/Wan2.2-S2V-14B)|`input_image`, `input_audio`, `audio_sample_rate`, `s2v_pose_video`|[code](./examples/wanvideo/model_inference/Wan2.2-S2V-14B_multi_clips.py)|-|-|-|-|
+|[Wan-AI/Wan2.2-S2V-14B](https://www.modelscope.cn/models/Wan-AI/Wan2.2-S2V-14B)|`input_image`, `input_audio`, `audio_sample_rate`, `s2v_pose_video`|[code](./examples/wanvideo/model_inference/Wan2.2-S2V-14B_multi_clips.py)|[code](./examples/wanvideo/model_training/full/Wan2.2-S2V-14B.sh)|[code](./examples/wanvideo/model_training/validate_full/Wan2.2-S2V-14B.py)|[code](./examples/wanvideo/model_training/lora/Wan2.2-S2V-14B.sh)|[code](./examples/wanvideo/model_training/validate_lora/Wan2.2-S2V-14B.py)|
 |[Wan-AI/Wan2.2-I2V-A14B](https://modelscope.cn/models/Wan-AI/Wan2.2-I2V-A14B)|`input_image`|[code](./examples/wanvideo/model_inference/Wan2.2-I2V-A14B.py)|[code](./examples/wanvideo/model_training/full/Wan2.2-I2V-A14B.sh)|[code](./examples/wanvideo/model_training/validate_full/Wan2.2-I2V-A14B.py)|[code](./examples/wanvideo/model_training/lora/Wan2.2-I2V-A14B.sh)|[code](./examples/wanvideo/model_training/validate_lora/Wan2.2-I2V-A14B.py)|
 |[Wan-AI/Wan2.2-T2V-A14B](https://modelscope.cn/models/Wan-AI/Wan2.2-T2V-A14B)||[code](./examples/wanvideo/model_inference/Wan2.2-T2V-A14B.py)|[code](./examples/wanvideo/model_training/full/Wan2.2-T2V-A14B.sh)|[code](./examples/wanvideo/model_training/validate_full/Wan2.2-T2V-A14B.py)|[code](./examples/wanvideo/model_training/lora/Wan2.2-T2V-A14B.sh)|[code](./examples/wanvideo/model_training/validate_lora/Wan2.2-T2V-A14B.py)|
 |[Wan-AI/Wan2.2-TI2V-5B](https://modelscope.cn/models/Wan-AI/Wan2.2-TI2V-5B)|`input_image`|[code](./examples/wanvideo/model_inference/Wan2.2-TI2V-5B.py)|[code](./examples/wanvideo/model_training/full/Wan2.2-TI2V-5B.sh)|[code](./examples/wanvideo/model_training/validate_full/Wan2.2-TI2V-5B.py)|[code](./examples/wanvideo/model_training/lora/Wan2.2-TI2V-5B.sh)|[code](./examples/wanvideo/model_training/validate_lora/Wan2.2-TI2V-5B.py)|
--- a/README_zh.md
+++ b/README_zh.md
@@ -208,7 +208,7 @@ save_video(video, "video1.mp4", fps=15, quality=5)
 |模型 ID|额外参数|推理|全量训练|全量训练后验证|LoRA 训练|LoRA 训练后验证|
 |-|-|-|-|-|-|-|
 |[Wan-AI/Wan2.2-Animate-14B](https://www.modelscope.cn/models/Wan-AI/Wan2.2-Animate-14B)|`input_image`, `animate_pose_video`, `animate_face_video`, `animate_inpaint_video`, `animate_mask_video`|[code](./examples/wanvideo/model_inference/Wan2.2-Animate-14B.py)|[code](./examples/wanvideo/model_training/full/Wan2.2-Animate-14B.sh)|[code](./examples/wanvideo/model_training/validate_full/Wan2.2-Animate-14B.py)|[code](./examples/wanvideo/model_training/lora/Wan2.2-Animate-14B.sh)|[code](./examples/wanvideo/model_training/validate_lora/Wan2.2-Animate-14B.py)|
-|[Wan-AI/Wan2.2-S2V-14B](https://www.modelscope.cn/models/Wan-AI/Wan2.2-S2V-14B)|`input_image`, `input_audio`, `audio_sample_rate`, `s2v_pose_video`|[code](./examples/wanvideo/model_inference/Wan2.2-S2V-14B_multi_clips.py)|-|-|-|-|
+|[Wan-AI/Wan2.2-S2V-14B](https://www.modelscope.cn/models/Wan-AI/Wan2.2-S2V-14B)|`input_image`, `input_audio`, `audio_sample_rate`, `s2v_pose_video`|[code](./examples/wanvideo/model_inference/Wan2.2-S2V-14B_multi_clips.py)|[code](./examples/wanvideo/model_training/full/Wan2.2-S2V-14B.sh)|[code](./examples/wanvideo/model_training/validate_full/Wan2.2-S2V-14B.py)|[code](./examples/wanvideo/model_training/lora/Wan2.2-S2V-14B.sh)|[code](./examples/wanvideo/model_training/validate_lora/Wan2.2-S2V-14B.py)|
 |[Wan-AI/Wan2.2-I2V-A14B](https://modelscope.cn/models/Wan-AI/Wan2.2-I2V-A14B)|`input_image`|[code](./examples/wanvideo/model_inference/Wan2.2-I2V-A14B.py)|[code](./examples/wanvideo/model_training/full/Wan2.2-I2V-A14B.sh)|[code](./examples/wanvideo/model_training/validate_full/Wan2.2-I2V-A14B.py)|[code](./examples/wanvideo/model_training/lora/Wan2.2-I2V-A14B.sh)|[code](./examples/wanvideo/model_training/validate_lora/Wan2.2-I2V-A14B.py)|
 |[Wan-AI/Wan2.2-T2V-A14B](https://modelscope.cn/models/Wan-AI/Wan2.2-T2V-A14B)||[code](./examples/wanvideo/model_inference/Wan2.2-T2V-A14B.py)|[code](./examples/wanvideo/model_training/full/Wan2.2-T2V-A14B.sh)|[code](./examples/wanvideo/model_training/validate_full/Wan2.2-T2V-A14B.py)|[code](./examples/wanvideo/model_training/lora/Wan2.2-T2V-A14B.sh)|[code](./examples/wanvideo/model_training/validate_lora/Wan2.2-T2V-A14B.py)|
 |[Wan-AI/Wan2.2-TI2V-5B](https://modelscope.cn/models/Wan-AI/Wan2.2-TI2V-5B)|`input_image`|[code](./examples/wanvideo/model_inference/Wan2.2-TI2V-5B.py)|[code](./examples/wanvideo/model_training/full/Wan2.2-TI2V-5B.sh)|[code](./examples/wanvideo/model_training/validate_full/Wan2.2-TI2V-5B.py)|[code](./examples/wanvideo/model_training/lora/Wan2.2-TI2V-5B.sh)|[code](./examples/wanvideo/model_training/validate_lora/Wan2.2-TI2V-5B.py)|
--- a/diffsynth/pipelines/wan_video_new.py
+++ b/diffsynth/pipelines/wan_video_new.py
@@ -1028,8 +1028,8 @@ class WanVideoUnit_S2V(PipelineUnit):
        if (inputs_shared.get("input_audio") is None and inputs_shared.get("audio_embeds") is None) or pipe.audio_encoder is None or pipe.audio_processor is None:
            return inputs_shared, inputs_posi, inputs_nega
        num_frames, height, width, tiled, tile_size, tile_stride = inputs_shared.get("num_frames"), inputs_shared.get("height"), inputs_shared.get("width"), inputs_shared.get("tiled"), inputs_shared.get("tile_size"), inputs_shared.get("tile_stride")
-        input_audio, audio_embeds, audio_sample_rate = inputs_shared.pop("input_audio"), inputs_shared.pop("audio_embeds"), inputs_shared.get("audio_sample_rate")
-        s2v_pose_video, s2v_pose_latents, motion_video = inputs_shared.pop("s2v_pose_video"), inputs_shared.pop("s2v_pose_latents"), inputs_shared.pop("motion_video")
+        input_audio, audio_embeds, audio_sample_rate = inputs_shared.pop("input_audio", None), inputs_shared.pop("audio_embeds", None), inputs_shared.get("audio_sample_rate", 16000)
+        s2v_pose_video, s2v_pose_latents, motion_video = inputs_shared.pop("s2v_pose_video", None), inputs_shared.pop("s2v_pose_latents", None), inputs_shared.pop("motion_video", None)

        audio_input_positive = self.process_audio(pipe, input_audio, audio_sample_rate, num_frames, audio_embeds=audio_embeds)
        inputs_posi.update(audio_input_positive)
--- a/diffsynth/trainers/unified_dataset.py
+++ b/diffsynth/trainers/unified_dataset.py
@@ -225,6 +225,13 @@ class ToAbsolutePath(DataProcessingOperator):
    def __call__(self, data):
        return os.path.join(self.base_path, data)

+class LoadAudio(DataProcessingOperator):
+    def __init__(self, sr=16000):
+        self.sr = sr
+    def __call__(self, data: str):
+        import librosa
+        input_audio, sample_rate = librosa.load(data, sr=self.sr)
+        return input_audio


 class UnifiedDataset(torch.utils.data.Dataset):
--- a/diffsynth/trainers/utils.py
+++ b/diffsynth/trainers/utils.py
@@ -603,6 +603,7 @@ def wan_parser():
    parser.add_argument("--dataset_repeat", type=int, default=1, help="Number of times to repeat the dataset per epoch.")
    parser.add_argument("--model_paths", type=str, default=None, help="Paths to load models. In JSON format.")
    parser.add_argument("--model_id_with_origin_paths", type=str, default=None, help="Model ID with origin paths, e.g., Wan-AI/Wan2.1-T2V-1.3B:diffusion_pytorch_model*.safetensors. Comma-separated.")
+    parser.add_argument("--audio_processor_config", type=str, default=None, help="Model ID with origin paths to the audio processor config, e.g., Wan-AI/Wan2.2-S2V-14B:wav2vec2-large-xlsr-53-english/")
    parser.add_argument("--learning_rate", type=float, default=1e-4, help="Learning rate.")
    parser.add_argument("--num_epochs", type=int, default=1, help="Number of epochs.")
    parser.add_argument("--output_path", type=str, default="./models", help="Output save path.")
--- a/examples/wanvideo/README.md
+++ b/examples/wanvideo/README.md
@@ -49,7 +49,7 @@ save_video(video, "video1.mp4", fps=15, quality=5)
 | Model ID | Extra Parameters | Inference | Full Training | Full Training Validation | LoRA Training | LoRA Training Validation |
 |-|-|-|-|-|-|-|
 |[Wan-AI/Wan2.2-Animate-14B](https://www.modelscope.cn/models/Wan-AI/Wan2.2-Animate-14B)|`input_image`, `animate_pose_video`, `animate_face_video`, `animate_inpaint_video`, `animate_mask_video`|[code](./model_inference/Wan2.2-Animate-14B.py)|[code](./model_training/full/Wan2.2-Animate-14B.sh)|[code](./model_training/validate_full/Wan2.2-Animate-14B.py)|[code](./model_training/lora/Wan2.2-Animate-14B.sh)|[code](./model_training/validate_lora/Wan2.2-Animate-14B.py)|
-|[Wan-AI/Wan2.2-S2V-14B](https://www.modelscope.cn/models/Wan-AI/Wan2.2-S2V-14B)|`input_image`, `input_audio`, `audio_sample_rate`, `s2v_pose_video`|[code](./model_inference/Wan2.2-S2V-14B_multi_clips.py)|-|-|-|-|
+|[Wan-AI/Wan2.2-S2V-14B](https://www.modelscope.cn/models/Wan-AI/Wan2.2-S2V-14B)|`input_image`, `input_audio`, `audio_sample_rate`, `s2v_pose_video`|[code](./model_inference/Wan2.2-S2V-14B_multi_clips.py)|[code](./model_training/full/Wan2.2-S2V-14B.sh)|[code](./model_training/validate_full/Wan2.2-S2V-14B.py)|[code](./model_training/lora/Wan2.2-S2V-14B.sh)|[code](./model_training/validate_lora/Wan2.2-S2V-14B.py)|
 |[Wan-AI/Wan2.2-I2V-A14B](https://modelscope.cn/models/Wan-AI/Wan2.2-I2V-A14B)|`input_image`|[code](./model_inference/Wan2.2-I2V-A14B.py)|[code](./model_training/full/Wan2.2-I2V-A14B.sh)|[code](./model_training/validate_full/Wan2.2-I2V-A14B.py)|[code](./model_training/lora/Wan2.2-I2V-A14B.sh)|[code](./model_training/validate_lora/Wan2.2-I2V-A14B.py)|
 |[Wan-AI/Wan2.2-T2V-A14B](https://modelscope.cn/models/Wan-AI/Wan2.2-T2V-A14B)||[code](./model_inference/Wan2.2-T2V-A14B.py)|[code](./model_training/full/Wan2.2-T2V-A14B.sh)|[code](./model_training/validate_full/Wan2.2-T2V-A14B.py)|[code](./model_training/lora/Wan2.2-T2V-A14B.sh)|[code](./model_training/validate_lora/Wan2.2-T2V-A14B.py)|
 |[Wan-AI/Wan2.2-TI2V-5B](https://modelscope.cn/models/Wan-AI/Wan2.2-TI2V-5B)|`input_image`|[code](./model_inference/Wan2.2-TI2V-5B.py)|[code](./model_training/full/Wan2.2-TI2V-5B.sh)|[code](./model_training/validate_full/Wan2.2-TI2V-5B.py)|[code](./model_training/lora/Wan2.2-TI2V-5B.sh)|[code](./model_training/validate_lora/Wan2.2-TI2V-5B.py)|
--- a/examples/wanvideo/README_zh.md
+++ b/examples/wanvideo/README_zh.md
@@ -49,7 +49,7 @@ save_video(video, "video1.mp4", fps=15, quality=5)
 |模型 ID|额外参数|推理|全量训练|全量训练后验证|LoRA 训练|LoRA 训练后验证|
 |-|-|-|-|-|-|-|
 |[Wan-AI/Wan2.2-Animate-14B](https://www.modelscope.cn/models/Wan-AI/Wan2.2-Animate-14B)|`input_image`, `animate_pose_video`, `animate_face_video`, `animate_inpaint_video`, `animate_mask_video`|[code](./model_inference/Wan2.2-Animate-14B.py)|[code](./model_training/full/Wan2.2-Animate-14B.sh)|[code](./model_training/validate_full/Wan2.2-Animate-14B.py)|[code](./model_training/lora/Wan2.2-Animate-14B.sh)|[code](./model_training/validate_lora/Wan2.2-Animate-14B.py)|
-|[Wan-AI/Wan2.2-S2V-14B](https://www.modelscope.cn/models/Wan-AI/Wan2.2-S2V-14B)|`input_image`, `input_audio`, `audio_sample_rate`, `s2v_pose_video`|[code](./model_inference/Wan2.2-S2V-14B_multi_clips.py)|-|-|-|-|
+|[Wan-AI/Wan2.2-S2V-14B](https://www.modelscope.cn/models/Wan-AI/Wan2.2-S2V-14B)|`input_image`, `input_audio`, `audio_sample_rate`, `s2v_pose_video`|[code](./model_inference/Wan2.2-S2V-14B_multi_clips.py)|[code](./model_training/full/Wan2.2-S2V-14B.sh)|[code](./model_training/validate_full/Wan2.2-S2V-14B.py)|[code](./model_training/lora/Wan2.2-S2V-14B.sh)|[code](./model_training/validate_lora/Wan2.2-S2V-14B.py)|
 |[Wan-AI/Wan2.2-I2V-A14B](https://modelscope.cn/models/Wan-AI/Wan2.2-I2V-A14B)|`input_image`|[code](./model_inference/Wan2.2-I2V-A14B.py)|[code](./model_training/full/Wan2.2-I2V-A14B.sh)|[code](./model_training/validate_full/Wan2.2-I2V-A14B.py)|[code](./model_training/lora/Wan2.2-I2V-A14B.sh)|[code](./model_training/validate_lora/Wan2.2-I2V-A14B.py)|
 |[Wan-AI/Wan2.2-T2V-A14B](https://modelscope.cn/models/Wan-AI/Wan2.2-T2V-A14B)||[code](./model_inference/Wan2.2-T2V-A14B.py)|[code](./model_training/full/Wan2.2-T2V-A14B.sh)|[code](./model_training/validate_full/Wan2.2-T2V-A14B.py)|[code](./model_training/lora/Wan2.2-T2V-A14B.sh)|[code](./model_training/validate_lora/Wan2.2-T2V-A14B.py)|
 |[Wan-AI/Wan2.2-TI2V-5B](https://modelscope.cn/models/Wan-AI/Wan2.2-TI2V-5B)|`input_image`|[code](./model_inference/Wan2.2-TI2V-5B.py)|[code](./model_training/full/Wan2.2-TI2V-5B.sh)|[code](./model_training/validate_full/Wan2.2-TI2V-5B.py)|[code](./model_training/lora/Wan2.2-TI2V-5B.sh)|[code](./model_training/validate_lora/Wan2.2-TI2V-5B.py)|
--- a/examples/wanvideo/model_training/full/Wan2.2-S2V-14B.sh
+++ b/examples/wanvideo/model_training/full/Wan2.2-S2V-14B.sh
@@ -0,0 +1,17 @@
+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/wans2v \
+  --dataset_metadata_path data/example_video_dataset/wans2v/metadata.csv \
+  --data_file_keys "video,input_audio,s2v_pose_video" \
+  --height 448 \
+  --width 832 \
+  --num_frames 81 \
+  --dataset_repeat 100 \
+  --model_id_with_origin_paths "Wan-AI/Wan2.2-S2V-14B:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.2-S2V-14B:wav2vec2-large-xlsr-53-english/model.safetensors,Wan-AI/Wan2.2-S2V-14B:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.2-S2V-14B:Wan2.1_VAE.pth" \
+  --audio_processor_config "Wan-AI/Wan2.2-S2V-14B:wav2vec2-large-xlsr-53-english/" \
+  --learning_rate 1e-5 \
+  --num_epochs 1 \
+  --trainable_models "dit" \
+  --remove_prefix_in_ckpt "pipe.dit." \
+  --output_path "./models/train/Wan2.2-S2V-14B_full" \
+  --extra_inputs "input_image,input_audio,s2v_pose_video" \
+  --use_gradient_checkpointing_offload
--- a/examples/wanvideo/model_training/lora/Wan2.2-S2V-14B.sh
+++ b/examples/wanvideo/model_training/lora/Wan2.2-S2V-14B.sh
@@ -0,0 +1,19 @@
+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/wans2v \
+  --dataset_metadata_path data/example_video_dataset/wans2v/metadata.csv \
+  --data_file_keys "video,input_audio,s2v_pose_video" \
+  --height 448 \
+  --width 832 \
+  --num_frames 81 \
+  --dataset_repeat 100 \
+  --model_id_with_origin_paths "Wan-AI/Wan2.2-S2V-14B:diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.2-S2V-14B:wav2vec2-large-xlsr-53-english/model.safetensors,Wan-AI/Wan2.2-S2V-14B:models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.2-S2V-14B:Wan2.1_VAE.pth" \
+  --audio_processor_config "Wan-AI/Wan2.2-S2V-14B:wav2vec2-large-xlsr-53-english/" \
+  --learning_rate 1e-4 \
+  --num_epochs 5 \
+  --remove_prefix_in_ckpt "pipe.dit." \
+  --output_path "./models/train/Wan2.2-S2V-14B_lora" \
+  --lora_base_model "dit" \
+  --lora_target_modules "q,k,v,o,ffn.0,ffn.2" \
+  --lora_rank 32 \
+  --extra_inputs "input_image,input_audio,s2v_pose_video" \
+  --use_gradient_checkpointing_offload
--- a/examples/wanvideo/model_training/train.py
+++ b/examples/wanvideo/model_training/train.py
@@ -2,7 +2,7 @@ import torch, os, json
 from diffsynth import load_state_dict
 from diffsynth.pipelines.wan_video_new import WanVideoPipeline, ModelConfig
 from diffsynth.trainers.utils import DiffusionTrainingModule, ModelLogger, launch_training_task, wan_parser
-from diffsynth.trainers.unified_dataset import UnifiedDataset, LoadVideo, ImageCropAndResize, ToAbsolutePath
+from diffsynth.trainers.unified_dataset import UnifiedDataset, LoadVideo, LoadAudio, ImageCropAndResize, ToAbsolutePath
 os.environ["TOKENIZERS_PARALLELISM"] = "false"


@@ -10,7 +10,7 @@ os.environ["TOKENIZERS_PARALLELISM"] = "false"
 class WanTrainingModule(DiffusionTrainingModule):
    def __init__(
        self,
-        model_paths=None, model_id_with_origin_paths=None,
+        model_paths=None, model_id_with_origin_paths=None, audio_processor_config=None,
        trainable_models=None,
        lora_base_model=None, lora_target_modules="q,k,v,o,ffn.0,ffn.2", lora_rank=32, lora_checkpoint=None,
        use_gradient_checkpointing=True,
@@ -22,7 +22,9 @@ class WanTrainingModule(DiffusionTrainingModule):
        super().__init__()
        # Load models
        model_configs = self.parse_model_configs(model_paths, model_id_with_origin_paths, enable_fp8_training=False)
-        self.pipe = WanVideoPipeline.from_pretrained(torch_dtype=torch.bfloat16, device="cpu", model_configs=model_configs)
+        if audio_processor_config is not None:
+            audio_processor_config = ModelConfig(model_id=audio_processor_config.split(":")[0], origin_file_pattern=audio_processor_config.split(":")[1])
+        self.pipe = WanVideoPipeline.from_pretrained(torch_dtype=torch.bfloat16, device="cpu", model_configs=model_configs, audio_processor_config=audio_processor_config)
        
        # Training mode
        self.switch_pipe_to_training_mode(
@@ -109,12 +111,14 @@ if __name__ == "__main__":
            time_division_remainder=1,
        ),
        special_operator_map={
-            "animate_face_video": ToAbsolutePath(args.dataset_base_path) >> LoadVideo(args.num_frames, 4, 1, frame_processor=ImageCropAndResize(512, 512, None, 16, 16))
+            "animate_face_video": ToAbsolutePath(args.dataset_base_path) >> LoadVideo(args.num_frames, 4, 1, frame_processor=ImageCropAndResize(512, 512, None, 16, 16)),
+            "input_audio": ToAbsolutePath(args.dataset_base_path) >> LoadAudio(sr=16000),
        }
    )
    model = WanTrainingModule(
        model_paths=args.model_paths,
        model_id_with_origin_paths=args.model_id_with_origin_paths,
+        audio_processor_config=args.audio_processor_config,
        trainable_models=args.trainable_models,
        lora_base_model=args.lora_base_model,
        lora_target_modules=args.lora_target_modules,
--- a/examples/wanvideo/model_training/validate_full/Wan2.2-S2V-14B.py
+++ b/examples/wanvideo/model_training/validate_full/Wan2.2-S2V-14B.py
@@ -0,0 +1,53 @@
+import torch
+from PIL import Image
+import librosa
+from diffsynth import VideoData, save_video_with_audio, 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="Wan-AI/Wan2.2-S2V-14B", origin_file_pattern="diffusion_pytorch_model*.safetensors"),
+        ModelConfig(model_id="Wan-AI/Wan2.2-S2V-14B", origin_file_pattern="wav2vec2-large-xlsr-53-english/model.safetensors"),
+        ModelConfig(model_id="Wan-AI/Wan2.2-S2V-14B", origin_file_pattern="models_t5_umt5-xxl-enc-bf16.pth"),
+        ModelConfig(model_id="Wan-AI/Wan2.2-S2V-14B", origin_file_pattern="Wan2.1_VAE.pth"),
+    ],
+    audio_processor_config=ModelConfig(model_id="Wan-AI/Wan2.2-S2V-14B", origin_file_pattern="wav2vec2-large-xlsr-53-english/"),
+)
+
+state_dict = load_state_dict("models/train/Wan2.2-S2V-14B_full/epoch-0.safetensors")
+pipe.dit.load_state_dict(state_dict, strict=False)
+pipe.enable_vram_management()
+
+
+num_frames = 81 # 4n+1
+height = 448
+width = 832
+
+prompt = "a person is singing"
+negative_prompt = "画面模糊，最差质量，画面模糊，细节模糊不清，情绪激动剧烈，手快速抖动，字幕，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 = Image.open("data/example_video_dataset/wans2v/pose.png").convert("RGB").resize((width, height))
+# s2v audio input, recommend 16kHz sampling rate
+audio_path = 'data/example_video_dataset/wans2v/sing.MP3'
+input_audio, sample_rate = librosa.load(audio_path, sr=16000)
+# S2V pose video input 
+pose_video_path = 'data/example_video_dataset/wans2v/pose.mp4'
+pose_video = VideoData(pose_video_path, height=height, width=width)
+
+# Speech-to-video with pose
+video = pipe(
+    prompt=prompt,
+    input_image=input_image,
+    negative_prompt=negative_prompt,
+    seed=0,
+    num_frames=num_frames,
+    height=height,
+    width=width,
+    audio_sample_rate=sample_rate,
+    input_audio=input_audio,
+    s2v_pose_video=pose_video,
+    num_inference_steps=40,
+)
+save_video_with_audio(video[1:], "video_pose_with_audio.mp4", audio_path, fps=16, quality=5)
--- a/examples/wanvideo/model_training/validate_lora/Wan2.2-S2V-14B.py
+++ b/examples/wanvideo/model_training/validate_lora/Wan2.2-S2V-14B.py
@@ -0,0 +1,51 @@
+import torch
+from PIL import Image
+import librosa
+from diffsynth import VideoData, save_video_with_audio
+from diffsynth.pipelines.wan_video_new import WanVideoPipeline, ModelConfig
+
+pipe = WanVideoPipeline.from_pretrained(
+    torch_dtype=torch.bfloat16,
+    device="cuda",
+    model_configs=[
+        ModelConfig(model_id="Wan-AI/Wan2.2-S2V-14B", origin_file_pattern="diffusion_pytorch_model*.safetensors"),
+        ModelConfig(model_id="Wan-AI/Wan2.2-S2V-14B", origin_file_pattern="wav2vec2-large-xlsr-53-english/model.safetensors"),
+        ModelConfig(model_id="Wan-AI/Wan2.2-S2V-14B", origin_file_pattern="models_t5_umt5-xxl-enc-bf16.pth"),
+        ModelConfig(model_id="Wan-AI/Wan2.2-S2V-14B", origin_file_pattern="Wan2.1_VAE.pth"),
+    ],
+    audio_processor_config=ModelConfig(model_id="Wan-AI/Wan2.2-S2V-14B", origin_file_pattern="wav2vec2-large-xlsr-53-english/"),
+)
+
+pipe.load_lora(pipe.dit, "models/train/Wan2.2-S2V-14B_lora/epoch-4.safetensors", alpha=1)
+pipe.enable_vram_management()
+
+
+num_frames = 81 # 4n+1
+height = 448
+width = 832
+
+prompt = "a person is singing"
+negative_prompt = "画面模糊，最差质量，画面模糊，细节模糊不清，情绪激动剧烈，手快速抖动，字幕，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 = Image.open("data/example_video_dataset/wans2v/pose.png").convert("RGB").resize((width, height))
+# s2v audio input, recommend 16kHz sampling rate
+audio_path = 'data/example_video_dataset/wans2v/sing.MP3'
+input_audio, sample_rate = librosa.load(audio_path, sr=16000)
+# Pose video input
+pose_video_path = 'data/example_video_dataset/wans2v/pose.mp4'
+pose_video = VideoData(pose_video_path, height=height, width=width)
+
+# Speech-to-video with pose
+video = pipe(
+    prompt=prompt,
+    input_image=input_image,
+    negative_prompt=negative_prompt,
+    seed=0,
+    num_frames=num_frames,
+    height=height,
+    width=width,
+    audio_sample_rate=sample_rate,
+    input_audio=input_audio,
+    s2v_pose_video=pose_video,
+    num_inference_steps=40,
+)
+save_video_with_audio(video[1:], "video_pose_with_audio.mp4", audio_path, fps=16, quality=5)