comfy-org/comfyui

ComfyUI is a desktop and cloud application for generating images, videos, 3D models, and audio using AI diffusion models. The README describes it as the most powerful and modular AI engine for content creation, aimed at visual professionals who want fine-grained control over every model and parameter. Instead of typing a prompt into a single box, users build a flowchart on a canvas where each box, called a node, performs one step, loading a model, applying a prompt, sampling an image, upscaling it, masking an area, and so on, and wires connect the outputs of one node into the inputs of the next. This node-graph approach lets people experiment with and reuse complicated pipelines without writing code, and the same graph can be shared by saving it as a JSON file or even by exporting the workflow embedded inside a generated PNG, WebP, or FLAC file. ComfyUI natively supports a long list of open-source models for images, image editing, video, audio, and 3D, listed in the README, and it can also call paid closed-source models through optional API nodes. The README highlights smart memory management that can run large models on GPUs with as little as 1 GB of video memory by offloading, support for running on CPU when no GPU is available, an asynchronous queue, and the ability to expose a finished workflow as a simple app via App Mode. Someone would use ComfyUI when they want repeatable, controllable generative workflows for production work, rather than one-off prompts in a hosted chat product. It runs on Windows, Linux, and macOS, with NVIDIA, AMD, Intel, Apple Silicon, and Ascend GPUs, and is written in Python on PyTorch.