DiffSynth-Studio Wan2.2-TI2V-5B Minimal Runtime

This repository has been reduced to the files needed to use the Wan2.2-TI2V-5B related DiffSynth code as a local folder import, without installing the full DiffSynth package.

What remains

• diffsynth/pipelines/wan_video.py: Wan video pipeline, including the WanVideoUnit_ImageEmbedderFused path used by Wan-AI/Wan2.2-TI2V-5B.
• diffsynth/models/wan_video_*.py: Wan video model, text encoder, VAE, and related optional Wan components still imported by the Wan pipeline.
• diffsynth/models/model_loader.py and diffsynth/configs/model_configs.py: model loading and Wan model config metadata.
• diffsynth/core, diffsynth/diffusion, and selected diffsynth/utils modules: shared runtime helpers needed by the Wan pipeline.
• examples/wanvideo/model_inference/Wan2.2-TI2V-5B.py: normal inference example.
• examples/wanvideo/model_inference_low_vram/Wan2.2-TI2V-5B.py: low-VRAM inference example.

Removed content includes unrelated image/audio pipelines, non-Wan model implementations, metrics, documentation, and non-TI2V examples/training scripts.

Local import usage

Put this folder on PYTHONPATH or run your script from the repository root, then import directly:

import torch
from diffsynth import WanVideoPipeline, ModelConfig

pipe = WanVideoPipeline.from_pretrained(
    torch_dtype=torch.bfloat16,
    device="cuda",
    model_configs=[
        ModelConfig(model_id="Wan-AI/Wan2.2-TI2V-5B", origin_file_pattern="models_t5_umt5-xxl-enc-bf16.pth"),
        ModelConfig(model_id="Wan-AI/Wan2.2-TI2V-5B", origin_file_pattern="diffusion_pytorch_model*.safetensors"),
        ModelConfig(model_id="Wan-AI/Wan2.2-TI2V-5B", origin_file_pattern="Wan2.2_VAE.pth"),
    ],
    tokenizer_config=ModelConfig(model_id="Wan-AI/Wan2.1-T2V-1.3B", origin_file_pattern="google/umt5-xxl/"),
)

You can also use:

from diffsynth.pipelines.wan_video import WanVideoPipeline, ModelConfig
from diffsynth.utils.data import save_video