amanuelcm/Wan2.1-T2V-1.3B-HansCharacter

Instala las dependencias y carga el adaptador LoRA sobre el modelo base de Wan 2.1:

pip install -U diffusers transformers accelerate

import torch
from diffusers.utils import export_to_video
from diffusers import AutoencoderKLWan, WanPipeline
from diffusers.schedulers.scheduling_unipc_multistep import UniPCMultistepScheduler

model_id = "Wan-AI/Wan2.1-T2V-1.3B-Diffusers"

vae = AutoencoderKLWan.from_pretrained(
    model_id,
    subfolder="vae",
    torch_dtype=torch.float32
)

pipe = WanPipeline.from_pretrained(
    model_id,
    vae=vae,
    torch_dtype=torch.bfloat16
)

pipe.scheduler = UniPCMultistepScheduler.from_config(
    pipe.scheduler.config,
    flow_shift=5.0
)
pipe.to("cuda")
pipe.load_lora_weights("amanuelcm/Wan2.1-T2V-1.3B-HansCharacter")
pipe.enable_model_cpu_offload()

prompt = "#hans, sitting by a rainy window at night, reading an old leather book, warm lamp light glowing softly, raindrops streaking down the glass, deep focus cinematic shot"

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=480,
    width=640,
    num_frames=49,
    guidance_scale=5.0,
    num_inference_steps=32
).frames[0]

export_to_video(output, "output.mp4", fps=16)

El prompt debe incluir #hans para activar la representación del personaje. Para ComfyUI puede utilizarse el flujo comfy.json proporcionado por el repositorio junto con los componentes recomendados de Wan 2.1.