ComfyUI
The node-based studio for open image, video, audio and 3D models: build visual workflows and run them on your own GPU.
ComfyUI is an open-source application for generating images, video, audio and 3D with open models. Instead of a single prompt box, you build a graph of nodes, for example one for the model, one for the text prompt, one for the sampler, and connect them. Workflows can be saved, shared and reused, and a local API lets other software run them.
It supports a long list of open models, including Qwen-Image, FLUX.2 Klein, Wan 2.1 and 2.2 and LTX-Video, plus separate VAEs, text encoders, LoRAs and ControlNets. It manages video memory and system memory aggressively and can offload or stream weights, which is how it runs large models on modest cards. The project says even the biggest open models can run on as little as 4 GB of VRAM with 8 GB of RAM, although more memory is much faster.
You can use the desktop app on Windows and macOS, a portable build, or a manual install on Windows, Linux and macOS with NVIDIA, AMD, Intel, Apple silicon and Ascend hardware. The core runs fully offline; optional partner nodes call paid closed models through an API, and you can disable them with the --offline flag. ComfyUI is licensed under GPL-3.0.
Good for
- Running open image models such as FLUX and Qwen-Image on a home GPU
- Repeatable, shareable image and video workflows without writing code
- Production pipelines that call workflows through an API
- Trying new open image and video models soon after release
Watch out for
- The node graph is powerful but has a real learning curve for beginners
- Custom nodes are community code, so install them with the same care as any software
- GPL-3.0 requires you to share changes if you distribute a modified ComfyUI
- Partner nodes use paid external APIs, which are not local or free