ComfyUI

Overview

ComfyUI is an AI creation engine for visual professionals, centered on a modular node-graph interface for building and reusing image, video, audio, 3D, and text workflows. It offers local and cloud options, an API, App Mode, and Comfy Agent, which can plan, build, and run workflows from natural-language descriptions inside ComfyUI.

Key Features

• Visual node graph for building and reusing image, video, audio, 3D, and text workflows without code, with visible and adjustable models, parameters, and processing steps.
• Comfy Agent lets users describe what they want inside ComfyUI; it plans, builds, and runs workflows while editing the canvas alongside the user.
• Efficient local execution with asynchronous queueing, partial graph re-execution, smart VRAM and RAM management, model offloading, and quantized model support.
• Broad native model support spanning image generation and editing, video generation, audio and music, 3D and vision, and text generation.
• Reusable subgraphs, workflow templates, App Mode, a local API, and JSON workflow save/load with workflow and seed recovery from supported generated media.
• Runs fully offline for built-in functionality; the core does not download anything unless requested, and optional API nodes can be disabled.

Use Cases

• Creating product scenes and video ads from base compositions, product references, keyframes, and motion references.
• Generating and editing images with models such as Stable Diffusion, SDXL, SD3.5, Flux.1, and Flux.2.
• Generating video with models such as Wan, LTX-Video, HunyuanVideo, CogVideoX, and Mochi.
• Running 3D, vision, and media processing tasks such as depth estimation, segmentation, upscaling, frame interpolation, and inpainting or outpainting.

Getting Started

• Download the desktop application or try Comfy Cloud for free.
• Load a workflow from a community template or build one from scratch.
• Generate and iterate by running and adjusting workflows, then scale when ready.

Deployment & Requirements

• Available locally on Windows, Linux, and macOS through the desktop application, portable install, or manual install, and available in the cloud via Comfy Cloud.
• Manual install supports all operating systems and GPU types, including NVIDIA, AMD, Intel, Apple Silicon, and Ascend.
• With asynchronous weight streaming, the biggest open-source models can run on as low as 4GB VRAM plus 8GB RAM relatively quickly.

Before You Adopt

• License: GPL-3.0. Review its terms before using, modifying, or distributing the project.
• Commits outside stable release tags may be very unstable and can break many custom nodes.