SDXS 512 DreamShaper Anime

🚀 SDXS-512-DreamShaper-Anime

SDXS是一个能够基于提示文本实时生成高分辨率图像的模型，它采用分数蒸馏和特征匹配技术进行训练。如需了解更多信息，请参考我们的研究论文：SDXS: Real-Time One-Step Latent Diffusion Models with Image Conditions。我们将该模型作为研究的一部分进行开源。

SDXS-512-DreamShaper-Anime是SDXS-512-DreamShaper的动漫风格LoRA。请关注我们的仓库以获取最新更新。

🚀 快速开始

模型简介

SDXS模型可依据提示文本实时生成高分辨率图像，采用了分数蒸馏和特征匹配技术训练。SDXS-512-DreamShaper-Anime则是基于SDXS-512-DreamShaper的动漫风格LoRA。

依赖库安装

确保你已经安装了以下依赖库：

torch
diffusers
peft

运行示例

运行以下代码即可生成图像：

import torch
from diffusers import StableDiffusionPipeline
import peft

repo = "IDKiro/sdxs-512-dreamshaper"
lora_repo = "IDKiro/sdxs-512-dreamshaper-anime"
weight_type = torch.float16     # or float32

# Load model.
pipe = StableDiffusionPipeline.from_pretrained(repo, torch_dtype=weight_type)
pipe.unet = PeftModel.from_pretrained(pipe.unet, lora_repo)
pipe.to("cuda")

prompt = "Self-portrait oil painting, a beautiful cyborg with golden hair, 8k"

# Ensure using 1 inference step and CFG set to 0.
image = pipe(
    prompt, 
    num_inference_steps=1, 
    guidance_scale=0
).images[0]

image.save("output.png")

运行上述代码后，会生成一张名为output.png的图像，示例图像如下：

💻 使用示例

基础用法

import torch
from diffusers import StableDiffusionPipeline
import peft

repo = "IDKiro/sdxs-512-dreamshaper"
lora_repo = "IDKiro/sdxs-512-dreamshaper-anime"
weight_type = torch.float16     # or float32

# Load model.
pipe = StableDiffusionPipeline.from_pretrained(repo, torch_dtype=weight_type)
pipe.unet = PeftModel.from_pretrained(pipe.unet, lora_repo)
pipe.to("cuda")

prompt = "Self-portrait oil painting, a beautiful cyborg with golden hair, 8k"

# Ensure using 1 inference step and CFG set to 0.
image = pipe(
    prompt, 
    num_inference_steps=1, 
    guidance_scale=0
).images[0]

image.save("output.png")

高级用法

你可以根据需求调整num_inference_steps和guidance_scale等参数，以获得不同风格和质量的图像。例如：

import torch
from diffusers import StableDiffusionPipeline
import peft

repo = "IDKiro/sdxs-512-dreamshaper"
lora_repo = "IDKiro/sdxs-512-dreamshaper-anime"
weight_type = torch.float16     # or float32

# Load model.
pipe = StableDiffusionPipeline.from_pretrained(repo, torch_dtype=weight_type)
pipe.unet = PeftModel.from_pretrained(pipe.unet, lora_repo)
pipe.to("cuda")

prompt = "Self-portrait oil painting, a beautiful cyborg with golden hair, 8k"

# 调整推理步数和引导比例
image = pipe(
    prompt, 
    num_inference_steps=5, 
    guidance_scale=5
).images[0]

image.save("output_advanced.png")

📄 许可证

本模型采用OpenRAIL++许可证进行开源。

📚 引用我们的工作

如果你使用了我们的模型或研究成果，请引用以下论文：

@article{song2024sdxs,
  author    = {Yuda Song, Zehao Sun, Xuanwu Yin},
  title     = {SDXS: Real-Time One-Step Latent Diffusion Models with Image Conditions},
  journal   = {arxiv},
  year      = {2024},
}