关于此LoRA

这是专为Wan2.1 14B视频生成模型训练的LoRA适配器。

兼容Diffusers和ComfyUI框架，可加载至Wan2.1的文本转视频和图像转视频双模型。

训练平台：Replicate
训练工具链：https://replicate.com/ostris/wan-lora-trainer/train

触发词

使用CYB77或"CYB77风格"激活视频生成特性。

使用指南

Replicate平台提供优化后的Wan2.1模型集合，兼顾速度与成本效益，支持本LoRA调用：

• https://replicate.com/collections/wan-video
• https://replicate.com/fofr/wan2.1-with-lora

通过Replicate API调用

import replicate  

input = {  
    "prompt": "CYB77",  
    "image": "https://replicate.delivery/xezq/4BGR8w4ELvJfcqIG4KIF0Kr82JxtfQCM2xVAIUieSeifmTZjC/output_frame.jpg",  
    "lora_url": "https://huggingface.co/fofr/wan-14b-cyberpunk-realistic/resolve/main/wan-14b-i2v-cyberpunk-realistic-lora.safetensors"  
}  

output = replicate.run(  
    "fofr/wan2.1-with-lora:f83b84064136a38415a3aff66c326f94c66859b8ad7a2cb432e2822774f07b08",  
    model="14b",  
    input=input  
)  
for index, item in enumerate(output):  
    with open(f"output_{index}.mp4", "wb") as file:  
        file.write(item.read())  

Diffusers框架集成

pip install git+https://github.com/huggingface/diffusers.git  

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-14B-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)  
flow_shift = 3.0  # 720P设为5.0，480P设为3.0  
pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config, flow_shift=flow_shift)  
pipe.to("cuda")  

pipe.load_lora_weights("fofr/wan-14b-cyberpunk-realistic")  
pipe.enable_model_cpu_offload()  # 低显存环境适用  

prompt = "CYB77"  
negative_prompt = "明亮色调、过曝、静态模糊、细节缺失、字幕、艺术风格、绘画作品、静态图像、整体灰暗、劣质画面、JPEG压缩伪影、丑陋、残缺、多余手指、手部绘制粗糙、面部畸形、肢体变形、手指粘连、静止画面、杂乱背景、三腿生物、背景人群过多、倒行走路"  

output = pipe(  
    prompt=prompt,  
    negative_prompt=negative_prompt,  
    height=480,  
    width=832,  
    num_frames=81,  
    guidance_scale=5.0,  
).frames[0]  
export_to_video(output, "output.mp4", fps=16)  

训练参数

• 训练步数：5000
• 学习率：0.0001
• LoRA秩：32

作品投稿

欢迎在社区讨论区分享您用此LoRA创作的视频作品。