deng8470/Wan2.2-T2V-A14B-Lightning-Diffusers-V2.0

Wan2.2-Lightning

Model Weights Source

This model weight is generated by merging the following two checkpoints using the LoRA merge functionality from 🤗 diffusers:

• Base model: Wan-AI/Wan2.2-T2V-A14B-Diffusers
• LoRA adapter: Wan2.2-T2V-A14B-4steps-lora-rank64-Seko-V2.0 from lightx2v/Wan2.2-Lightning

Installation

You can install the required dependencies with:

pip install diffusers ftfy imageio-ffmpeg opencv-python-headless imageio

Usage

Below is a simple example showing how to run the weights:

import torch
import numpy as np
from diffusers import WanPipeline, AutoencoderKLWan
from diffusers.utils import export_to_video, load_image

model_id = "deng8470/Wan2.2-T2V-A14B-Lightning-Diffusers-V2.0"
dtype = torch.bfloat16
vae = AutoencoderKLWan.from_pretrained(model_id, subfolder="vae", torch_dtype=torch.float32)
pipe = WanPipeline.from_pretrained(model_id, vae=vae, torch_dtype=dtype)
pipe.enable_model_cpu_offload()

height = 720
width = 1280

prompt = "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage."
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"
output = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    height=height,
    width=width,
    num_frames=81,
    guidance_scale=1.0,
    guidance_scale_2=1.0,
    num_inference_steps=4,
).frames[0]
export_to_video(output, "t2v_out.mp4", fps=16)

Recommended Inference Configuration

720p resolution + CFG scale = 1.0 + 4 inference steps. This setting is optimized for the 4-step distilled weight (Wan2.2-T2V-A14B-4steps-lora-rank64-Seko-V2.0) trained via Phased DMD, as detailed in the technical report below. More details: Wan2.2-Lightning Official GitHub Repository

License

This project is licensed under the Apache License 2.0.

You may use, modify, and distribute this project in compliance with the terms of the license.