【支持中文】stable-diffusion-3安装部署教程-SD3_stable diffusion3部署

运行界面

一、下载模型

注意是下载的是stable-diffusion-3-medium-diffusers

git clone https://www.modelscope.cn/AI-ModelScope/stable-diffusion-3-medium-diffusers.git

二、搭建基础环境

conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia
 
# 或者pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
 
pip install gradio
 
pip install spaces

三、安装SD3的环境

由于pip环境的diffusers和transformers还不支持SD3，所以需要编译安装

git clone https://github.com/huggingface/diffusers.git
git clone https://github.com/huggingface/transformers
 
cd diffusers/
python setup.py develop
pip install -e .
 
cd ../transformers
python setup.py develop
pip install -e .

三、运行SD3 DEMO

将下面的代码保存为demo.py，再直接python demo.py即可

翻译用到了百度的接口，可以换成自己的AK/SK，不喜欢可以去掉

import gradio as gr
import numpy as np
import random
import torch
from diffusers import StableDiffusion3Pipeline, SD3Transformer2DModel, FlowMatchEulerDiscreteScheduler
import spaces
import requests
import hashlib
import random
import json
 
device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.float16
 
repo = "/你的模型路径/stable-diffusion-3-medium-diffusers"
pipe = StableDiffusion3Pipeline.from_pretrained(repo, torch_dtype=torch.float16).to(device)
 
MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1344
 
def translate_baidu(query, from_lang='auto', to_lang='en'):
    base_url = "https://fanyi-api.baidu.com/api/trans/vip/translate"
    appid = '你的APPID'
    secret_key = '你的SK'
    salt = str(random.randint(32768, 65536))
    sign = appid + query + salt + secret_key
    sign = hashlib.md5(sign.encode()).hexdigest()
 
    params = {
        'q': query,
        'from': from_lang,
        'to': to_lang,
        'appid': appid,
        'salt': salt,
        'sign': sign
    }
 
    response = requests.get(base_url, params=params)
    if response.status_code == 200:
        result = response.json()
        if "trans_result" in result:
            return result["trans_result"][0]["dst"]
        else:
            return result
    else:
        return f"Error: {response.status_code}"
 
@spaces.GPU
def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, progress=gr.Progress(track_tqdm=True)):
 
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
        
    generator = torch.Generator().manual_seed(seed)
 
    print(prompt)
    prompt = translate_baidu(prompt, 'zh', 'en')
    print(prompt)
 
    image = pipe(
        prompt = prompt, 
        negative_prompt = negative_prompt,
        guidance_scale = guidance_scale, 
        num_inference_steps = num_inference_steps, 
        width = width, 
        height = height,
        generator = generator
    ).images[0] 
    
    return image, seed
 
examples = [
 
]
 
css="""
#col-container {
    margin: 0 auto;
    max-width: 800px;
}
"""
 
with gr.Blocks(css=css) as demo:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown(f"""
        ### Stable Diffusion 3 测试
        #### By 你的名字
        """)
        
        with gr.Row():
            
            prompt = gr.Text(
                label="提示词",
                show_label=False,
                max_lines=1,
                placeholder="请输入提示词",
                container=False,
            )
            
            run_button = gr.Button("生成", scale=0)
        
        result = gr.Image(label="Result", show_label=False)
 
        with gr.Accordion("更多参数", open=False):
            
            negative_prompt = gr.Text(
                label="负面提示词",
                max_lines=1,
                placeholder="请输入负面提示词",
            )
            
            seed = gr.Slider(
                label="Seed",
                minimum=0,
                maximum=MAX_SEED,
                step=1,
                value=0,
            )
            
            randomize_seed = gr.Checkbox(label="随机种子", value=True)
            
            with gr.Row():
                
                width = gr.Slider(
                    label="宽",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=64,
                    value=1024,
                )
                
                height = gr.Slider(
                    label="高",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=64,
                    value=1024,
                )
            
            with gr.Row():
                
                guidance_scale = gr.Slider(
                    label="Guidance scale",
                    minimum=0.0,
                    maximum=10.0,
                    step=0.1,
                    value=5.0,
                )
                
                num_inference_steps = gr.Slider(
                    label="迭代步数",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=28,
                )
        
 
    gr.on(
        triggers=[run_button.click, prompt.submit, negative_prompt.submit],
        fn = infer,
        inputs = [prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
        outputs = [result, seed]
    )
 
demo.launch(server_name="0.0.0.0", server_port=7862, show_error=True)