language: zh
license: creativeml-openrail-m
tags:
- diffusion
- zh
- Chinese
Chinese-Style-Stable-Diffusion-2-v0.1
 |
 |
 |
|---|---|---|
 |
 |
 |
 |
 |
 |
大概是Huggingface 🤗社区首个开源的Stable diffusion 2 中文模型。该模型基于stable diffusion V2.1模型，在约500万条的中国风格筛选过的中文数据上进行微调，数据来源于多个开源数据集如LAION-5B, Noah-Wukong, Zero和一些网络数据。
Probably the first open sourced Chinese Stable Diffusion 2 model in Huggingface🤗 community. This model is finetuned based on stable diffusion V2.1 with 5M chinese style filtered data. Dataset is composed of several different chinese open source dataset such as LAION-5B, Noah-Wukong, Zero and some web data.
Model Details
Text Encoder
文本编码器使用冻结参数的lyua1225/clip-huge-zh-75k-steps-bs4096。
Text encoder is frozen lyua1225/clip-huge-zh-75k-steps-bs4096 .
Unet
在筛选过的的500万中文数据集上训练了150K steps，使用指数移动平均值(EMA)做原绘画能力保留，使模型能够在中文风格和原绘画能力之间获得权衡。
Training on 5M chinese style filtered data for 150k steps. Exponential moving average(EMA) is applied to keep the original Stable Diffusion 2 drawing capability and reach a balance between chinese style and original drawing capability.
Usage
因为使用了customized tokenizer, 所以需要优先加载一下tokenizer, 并传入trust_remote_code=True
Customized Tokenizer should be loaded first with 'trust_remote_code=True'.
import torch
from diffusers import StableDiffusionPipeline
from transformers import AutoTokenizer
tokenizer_id = "lyua1225/clip-huge-zh-75k-steps-bs4096"
sd2_id = "Midu/chinese-style-stable-diffusion-2-v0.1"
tokenizer = AutoTokenizer.from_pretrained(tokenizer_id, trust_remote_code=True)
pipe = StableDiffusionPipeline.from_pretrained(sd2_id, torch_dtype=torch.float16, tokenizer=tokenizer)
pipe.to("cuda")
image = pipe("赛博朋克风格的城市街道，8K分辨率，CG渲染", guidance_scale=10, num_inference_steps=20).images[0]
image.save("cyberpunk.jpeg")