本地部署 Stable Diffusion3

6月13日,Stability AI 正式开源20亿参数版本的Stable Diffusion 3 Medium。本文将在本地部署 SD3,GPU配置如下 GPU 2080TI / 22G

安装依赖

修改 WORKSPACE 位置信息,安装ComfyUI

复制代码
# #@title Environment Setup

from pathlib import Path

OPTIONS = {}
UPDATE_COMFY_UI = True  #@param {type:"boolean"}
INSTALL_COMFYUI_MANAGER = True  #@param {type:"boolean"}
INSTALL_ANIMATEDIFF = True  #@param {type:"boolean"}
INSTALL_CUSTOM_NODES_DEPENDENCIES = True  #@param {type:"boolean"}
OPTIONS['UPDATE_COMFY_UI'] = UPDATE_COMFY_UI
OPTIONS['INSTALL_COMFYUI_MANAGER'] = INSTALL_COMFYUI_MANAGER
OPTIONS['INSTALL_ANIMATEDIFF'] = INSTALL_ANIMATEDIFF
OPTIONS['INSTALL_CUSTOM_NODES_DEPENDENCIES'] = INSTALL_CUSTOM_NODES_DEPENDENCIES

WORKSPACE = "/notebook/SD/ComfyUI"



%cd /notebook/SD/

![ ! -d $WORKSPACE ] && echo -= Initial setup ComfyUI =- && git clone https://github.com/comfyanonymous/ComfyUI

%cd $WORKSPACE

!echo $WORKSPACE

if OPTIONS['UPDATE_COMFY_UI']:
  !echo "-= Updating ComfyUI =-"
  !git pull
 
!pip install -r requirements.txt

下载模型权重

复制代码
#@markdown ###Download standard resources

### SDXL
### I recommend these workflow examples: https://comfyanonymous.github.io/ComfyUI_examples/sdxl/

OPTIONS = {}

#@markdown **Models**

!wget -c "https://modelscope.cn/api/v1/models/AI-ModelScope/stable-diffusion-3-medium/repo?Revision=master&FilePath=sd3_medium.safetensors" -P ./models/checkpoints/

下载 Text Encoder

复制代码
#@markdown **clip**

!wget -c "https://modelscope.cn/api/v1/models/AI-ModelScope/stable-diffusion-3-medium/repo?Revision=master&FilePath=text_encoders%2Fclip_g.safetensors" -P ./models/clip/
!wget -c "https://modelscope.cn/api/v1/models/AI-ModelScope/stable-diffusion-3-medium/repo?Revision=master&FilePath=text_encoders%2Fclip_l.safetensors" -P ./models/clip/
!wget -c "https://modelscope.cn/api/v1/models/AI-ModelScope/stable-diffusion-3-medium/repo?Revision=master&FilePath=text_encoders%2Ft5xxl_fp16.safetensors" -P ./models/clip/
!wget -c "https://modelscope.cn/api/v1/models/AI-ModelScope/stable-diffusion-3-medium/repo?Revision=master&FilePath=text_encoders%2Ft5xxl_fp8_e4m3fn.safetensors" -P ./models/clip/
#通过cloudflared启动 ComfyUI ,
!wget "https://modelscope.oss-cn-beijing.aliyuncs.com/resource/cloudflared-linux-amd64.deb"
!dpkg -i cloudflared-linux-amd64.deb

启动 ComfyUI 服务器

复制代码
%cd /notebook/SD/ComfyUI
import subprocess
import threading
import time
import socket
import urllib.request

def iframe_thread(port):
  while True:
      time.sleep(0.5)
      sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
      result = sock.connect_ex(('127.0.0.1', port))
      if result == 0:
        break
      sock.close()
  print("\nComfyUI finished loading, trying to launch cloudflared (if it gets stuck here cloudflared is having issues)\n")

  p = subprocess.Popen(["cloudflared", "tunnel", "--url", "http://127.0.0.1:{}".format(port)], stdout=subprocess.PIPE, stderr=subprocess.PIPE)
  for line in p.stderr:
    l = line.decode()
    if "trycloudflare.com " in l:
      print("This is the URL to access ComfyUI:", l[l.find("http"):], end='')
    #print(l, end='')


threading.Thread(target=iframe_thread, daemon=True, args=(8891,)).start()
!python main.py --dont-print-server  --listen "0.0.0.0"  --port 8891 

运行

服务器启动之后,导入工作流配置

  1. 导入配置文件,配置文件位置在 https://www.modelscope.cn/models/AI-ModelScope/stable-diffusion-3-medium/files,comfy_example_workflows/sd3_medium_example_workflow_basic.json
  2. 输入提示词,点击 Queue Prompt 生成图片。

总结

根据以下提示词,比较ChatGPT、Google 和 SD 的差距。

复制代码
A transparent glass of very very small slim capsule is being held by fingers, inside the capsule there is a nuclear explosion, there is smoke and fire, the background of a cozy room

ChatGPT

Google Gemini


SD3

ChatGPT最好,比较细腻,其次是Gemini ,SD 稍微差一些,SD的好处是可以私有化部署,可以根据需要进行私有数据训练,SD 生成速度 2.1 it/s。

相关推荐
sigmoidAndRELU2 小时前
读Vista
笔记·stable diffusion·世界模型
修炼室15 天前
Stable Diffusion WebUI 本地部署完整教程
stable diffusion
NetX行者15 天前
Stable Diffusion:开启AI图像生成新纪元
人工智能·stable diffusion
这是一个懒人20 天前
SD和comfyui常用模型介绍和下载
stable diffusion·comfyui·模型下载
有点小帅得平哥哥21 天前
Stable Diffusion WebUI 本地部署指南(Windows 11 + RTX 4060 Ti)
stable diffusion
CoovallyAIHub25 天前
突破异常数据瓶颈!AnomalyAny:一句话+一张图,零样本生成任意异常图像
计算机视觉·stable diffusion
写代码的小阿帆1 个月前
Fractal Generative Models论文阅读笔记与代码分析
论文阅读·stable diffusion·transformer
春末的南方城市1 个月前
港科大&快手提出统一上下文视频编辑 UNIC,各种视频编辑任务一网打尽,还可进行多项任务组合!
人工智能·计算机视觉·stable diffusion·aigc·transformer
多恩Stone1 个月前
【Stable Diffusion 1.5 】在 Unet 中每个 Cross Attention 块中的张量变化过程
stable diffusion
今夕节度使1 个月前
ARM架构推理Stable Diffusiond
stable diffusion