Assemble image, video, and 3D workflows locally using nodes without coding
Control models, parameters, and outputs directly to create visual results by assembling the process on screen. By connecting nodes and lines to design complex generation processes, this tool allows users without coding knowledge to build complex workflows. ComfyUI is an open-source AI generation engine with this node-based interface.
Designing the generation process with node graphs
You can separate and connect each element, such as models, VAEs, text encoders, LoRA, and ControlNet, as individual nodes, going beyond just entering text prompts. The resulting graph can be saved as a JSON file for reuse or sharing, and the generated media files include workflow information, allowing you to restore seeds and settings. App Mode exposes complex workflows as a simple UI, making them easy for non-developers to use. Additionally, you can integrate workflows into applications or production pipelines via API endpoints.
Running large models on limited hardware
By applying asynchronous weight streaming technology, it processes large open-source models relatively quickly with only 4GB of VRAM and 8GB of RAM. It maximizes GPU computing resources and automatically optimizes VRAM and RAM management. Support for quantized models and model offloading enables stable inference even with limited resources. These optimizations allow high-quality generation locally without cloud costs.
Supported models and extensions
It natively supports the latest open-source image, video, audio, and 3D models, including Stable Diffusion, Flux, Wan, and HunyuanVideo.
- Various generation models It supports a wide range of open-source models for image, video, audio, 3D, and text generation.
- Access to closed-source models You can access commercial models through partner nodes.
- Custom node extensions You can extend and manage functionality through custom nodes and managers.
- Production integration You can integrate workflows into applications via API and App Mode.
Installation and execution environment
Desktop application installation is recommended for Windows and macOS, as it is the easiest way to start. Linux and other GPU environments (NVIDIA, AMD, Intel, Apple Silicon, Ascend) support manual installation. Python 3.13 is best supported, and it is recommended to use the latest major version of PyTorch. NVIDIA 20 series or higher GPUs require PyTorch cu130 or higher. You can simply start execution with the python main.py command.
Stability and compatibility checks
ComfyUI follows a weekly release cycle, but the schedule may change depending on model releases or codebase changes. The core, desktop, and frontend are managed in separate repositories, with differences between stable and development versions. Custom nodes may cause compatibility issues with core updates, so commits outside stable release tags may be unstable. You should check your local hardware specifications, especially GPU VRAM capacity, and ensure that required model files are placed in the correct folders. If you need fully offline execution, you can disable paid API nodes using the --disable-api-nodes flag.
