0
图像生成GPL-3.0Python

ComfyUI

Node-based interface for diffusion models — images, video and audio.

GitHub官网
Star 数
137.6K
Fork 数
16.4K
未关闭 Issue
5.2K
最近提交
2026-10-15

数据来自 GitHub API · 更新于 2026-10-15

ComfyUI 是什么？

ComfyUI lets you build generation pipelines by wiring nodes together in a graph: load a checkpoint, encode a prompt, sample, decode, upscale. It supports Stable Diffusion 1.5/SDXL/3.5, FLUX, video models such as Wan and HunyuanVideo, and a huge ecosystem of custom nodes. Workflows are saved as JSON (and embedded in output PNGs), so they are easy to share and reproduce.

主要功能

  • Graph/node workflow editor with reusable workflow files
  • Supports SD1.x, SDXL, SD3.5, FLUX, Wan, HunyuanVideo, LTX-Video and more
  • Smart memory management — runs large models on GPUs with little VRAM
  • Only re-executes the parts of a workflow that changed
  • Thousands of community custom nodes via ComfyUI-Manager

系统要求

平台
Windows · macOS · Linux
显卡
NVIDIA (CUDA), AMD (ROCm/DirectML), Intel Arc or Apple Silicon; CPU mode is very slow
显存
4 GB minimum for SD1.5; 12 GB+ recommended for FLUX and video models
内存
16 GB+
磁盘
10–100 GB depending on how many models you download
Python
3.12 or 3.13 recommended for manual installs

如何安装 ComfyUI

根据项目 README 整理。开源项目更新很快，请以 GitHub 页面上的最新说明为准。

Desktop app (Windows / macOS)

  1. 1
    Download ComfyUI Desktop from the official site (comfy.org) and run the installer.
  2. 2
    Pick your GPU type and an install location when prompted; the app sets up Python and PyTorch for you.
  3. 3
    Load a workflow template and let the app download the required models.

Windows portable (NVIDIA)

  1. 1
    Download the portable .7z package from the GitHub Releases page.
  2. 2
    Extract it with 7-Zip.
  3. 3
    Put checkpoint files in ComfyUI\models\checkpoints.
  4. 4
    Run run_nvidia_gpu.bat and open the URL it prints (default http://127.0.0.1:8188).

Manual install (Linux / macOS)

  1. 1
    Clone the repository.
    git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
  2. 2
    Create a virtual environment and install PyTorch for your GPU (see pytorch.org), then the requirements.
    python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
  3. 3
    Start the server and open http://127.0.0.1:8188.
    python main.py

优势

  • Supports new models first, often on release day
  • Most flexible and efficient Stable Diffusion/FLUX frontend
  • Workflows are portable and reproducible

不足

  • Steep learning curve for beginners
  • Custom nodes vary in quality and can break on updates

ComfyUI 替代品对比

软件Star 数许可证平台显存
ComfyUI

Node-based interface for diffusion models — images, video and audio.

137.6KGPL-3.04 GB minimum for SD1.5; 12 GB+ recommended for FLUX and video models
Stable Diffusion WebUI Forge

A faster, leaner fork of the A1111 WebUI with FLUX support.

13.1KAGPL-3.06 GB+ (FLUX NF4/GGUF variants run on 6–8 GB)
SwarmUI

A friendly generate tab on top of ComfyUI, with multi-GPU support.

4.7KMIT6 GB+
InvokeAI

Professional creative engine with a unified canvas for Stable Diffusion and FLUX.

28.6KApache-2.04 GB for SD1.5, 8 GB for SDXL, 10 GB+ for FLUX
Fooocus

Midjourney-like simplicity for SDXL: type a prompt and press Generate.

53.4KGPL-3.04 GB minimum, 6 GB+ recommended
Stable Diffusion WebUI (AUTOMATIC1111)

The classic browser UI for Stable Diffusion with a massive extension ecosystem.

165.1KAGPL-3.04 GB minimum (with --medvram/--lowvram), 8 GB+ recommended

相关模型

FLUX由 Stable Diffusion 原班人马打造的高质量开源权重图像模型。Stable Diffusion生态最庞大的开源图像模型。Wan阿里巴巴的开源权重视频模型，可自行部署。

常见问题

ComfyUI 免费吗？

是的——ComfyUI 是开源软件，采用 GPL-3.0 许可证。商用前请确认软件及模型权重的许可条款。

ComfyUI 支持 Windows、Mac 和 Linux 吗？

ComfyUI 支持：Windows, macOS, Linux。

运行 ComfyUI 需要什么配置？

NVIDIA (CUDA), AMD (ROCm/DirectML), Intel Arc or Apple Silicon; CPU mode is very slow. VRAM: 4 GB minimum for SD1.5; 12 GB+ recommended for FLUX and video models. RAM: 16 GB+

如何安装 ComfyUI？

1. Download ComfyUI Desktop from the official site (comfy.org) and run the installer. 2. Pick your GPU type and an install location when prompted; the app sets up Python and PyTorch for you. 3. Load a workflow template and let the app download the required models.

ComfyUI 有哪些好的替代品？

常见替代品包括 Stable Diffusion WebUI Forge, SwarmUI, InvokeAI, Fooocus，详见本页对比表。

可以不安装、在线使用 ComfyUI 吗？

如果只想要结果，AiLoft 可在浏览器中运行 GPT Image 2.5、Nano Banana Pro 等顶级图像模型，无需显卡和配置。

不想安装？直接在浏览器里生成图片

无需显卡、驱动或 Python。在线使用 GPT Image 2.5 和 Nano Banana Pro，并排对比效果。

AI 图片生成器
开源 AI 图像生成工具