ComfyUI Download: Node-Based AI Image and Video Generator

Most people who get into AI image generation run into ComfyUI sooner or later. Rather than hiding the text-to-image process behind a single button, it shows every step (loading the model, encoding the prompt, sampling, saving the image) as its own box. You connect these boxes with wires to build your own workflow. Its GitHub repository has roughly 144,000 stars as of October 2026.

What can you do with ComfyUI?

• Image generation and editing: Generate images from text and edit existing ones with many open models, from Stable Diffusion 1.5 and SDXL to Flux and Qwen Image.
• Video and audio: Supports video models such as Wan, LTX-Video and HunyuanVideo, as well as models that generate audio.
• Built-in tools: Inpainting (redrawing part of an image), outpainting (extending its edges), upscaling, and LoRA and ControlNet support come built in.
• Workflow sharing: Drag a PNG you generated into the window and the full workflow that produced it, including the seed, is restored.
• Smart execution: When you re-run a workflow, only the parts that changed are recomputed, which speeds up experimenting.
• Extensions: Community-made custom nodes extend what it can do, and ComfyUI-Manager makes installing them from the interface easy.

Who is it for?

It is ideal for designers, illustrators, video creators and AI enthusiasts who want fine-grained control over the generation process. Beginners hoping for one-click results may find it complex at first glance; the ready-made template workflows make getting started easier.

How to install

Desktop app (recommended): The easiest route for new users. The Download button opens comfy.org/download, where you’ll find installers for Windows and macOS.

Windows portable version: The GitHub releases page has separate 7z packages for NVIDIA, AMD and Intel graphics cards. Extract the package with 7-Zip and run the launcher inside. The NVIDIA package is about 1.9 GB.

Manual install (Windows, Linux, macOS): For users comfortable with Python, these are the official steps. First clone the repository:

git clone https://github.com/Comfy-Org/ComfyUI.git
cd ComfyUI

Install PyTorch for an NVIDIA graphics card (the official docs give separate commands for AMD, Intel and Apple Silicon):

pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu130

Then install the dependencies and start the program:

pip install -r requirements.txt
python main.py

First steps

With a manual install, the interface opens in your browser at http://127.0.0.1:8188. To generate images you need a model file: put the checkpoint files you download (.safetensors) in the models/checkpoints subfolder of your ComfyUI folder. Then open one of the ready-made templates, select your file in the model box, describe the image you want in the prompt box and add it to the queue.

Hardware requirements

• Windows desktop app: Windows 10 or newer, x64 or ARM64. An NVIDIA or AMD graphics card is recommended but not required.
• macOS desktop app: macOS 13 Ventura or newer, Apple Silicon (M1 and later).
• Disk: At least 4.85 GB per installation is recommended; models take extra space.
• Memory: According to the developers, smart offloading (loading weights piece by piece) lets even large models run with 4 GB of video memory and 8 GB of RAM, but a more powerful graphics card makes generation noticeably faster.

License

ComfyUI is an open-source project under the GPL-3.0 license and is free to use locally. Comfy Org also offers a paid cloud version (Comfy Cloud) for people without powerful hardware, and paid “partner” nodes that access closed models; neither is required. Also check the license of the model you use for the content you create.