llms.txt
# Comfy
> Comfy is the AI creation engine for visual professionals who demand control over every model, every parameter, and every output. Built around ComfyUI, the open-source node-graph runtime with 60,000+ community nodes, Comfy ships as a free desktop app, a managed cloud, an API, an MCP server and CLI for AI agents, and an enterprise platform.
The Comfy ecosystem spans these surfaces:
- **Comfy Desktop (local)**: the open-source ComfyUI engine, running models on your own hardware.
- **Comfy Cloud**: managed ComfyUI in the browser, with pre-loaded models, pre-installed custom nodes, and concurrent jobs.
- **Comfy API**: any ComfyUI workflow deployed as a production endpoint, triggered from your own apps and pipelines.
- **Comfy MCP**: the full ComfyUI engine exposed over the Model Context Protocol, so any AI agent can build workflows and generate images, video, audio, and 3D.
- **Comfy CLI**: the same engine from the terminal and from coding agents such as Claude Code, Codex, and Cursor, with JSON output for scripts.
- **Comfy Enterprise**: ComfyUI for teams, with dedicated GPU compute, SSO and role-based access, bring-your-own API keys, and custom SLAs.
- **Comfy Developer Platform** (platform.comfy.org): API keys, usage, builds, and serverless deployments for developers.
How it works: a ComfyUI workflow is a graph. Each node loads a model, encodes a prompt, samples, edits, upscales, or saves; edges carry typed data between nodes; the whole graph is one JSON file. The same workflow runs unchanged on Comfy Desktop, Comfy Cloud, the Comfy API, the CLI, and MCP, so a pipeline built on a laptop can be deployed as an endpoint without a rewrite. Open-weight models run from downloaded weights; hosted models such as Kling, Veo, and GPT Image run through Partner Nodes. The user directs every parameter; nothing is hidden behind a prompt box.
What people build with it: VFX shot work (clean plates, sky replacement, de-aging), animation pipelines (character sheets, keyframes, in-betweening), ad creative (moodboards, storyboards, product photography, B-roll), game and 3D assets, and music and audio. The Learning section below links a tutorial for each; the Customers section links the studios and schools doing it in production.
Studios, brands, and schools building with Comfy include Series Entertainment, Moment Factory, Ubisoft La Forge, Groove Jones, Svedka with Silverside AI, and the Open Story Movement. Carnegie Mellon, USC, NTU, and UAL's Creative Computing Institute teach with ComfyUI. Use cases concentrate in VFX and animation, advertising and creative studios, gaming, and eCommerce and fashion.
Languages: comfy.org is published in English and Simplified Chinese, with a Japanese home page at https://comfy.org/ja/. Nearly every page has a twin under /zh-CN/ (for example https://comfy.org/zh-CN/cli/); the affiliate pages, the enterprise MSA, the terms of service, and the model files directory are English only. Comfy Workflows is published in 11 languages (English, 简体中文, 繁體中文, Español, Français, 日本語, 한국어, Português, Русский, العربية, and Türkçe). The docs are published in English, Chinese, Japanese, and Korean.
## Names and spelling
- **ComfyUI** (one word, capital UI) is the open-source node-graph engine. **Comfy** is the company behind it (Comfy Org, `Comfy-Org` on GitHub) and the product family built on it.
- **Comfy Desktop** is the ComfyUI desktop app. **Comfy Cloud** is hosted ComfyUI at cloud.comfy.org. **Comfy API**, **Comfy MCP**, and **Comfy CLI** (package `comfy-cli`) are the programmatic surfaces. **Comfy Enterprise** is the team offering.
- **Comfy Workflows** (comfy.org/workflows) is the workflow hub. Do not rename it "ComfyUI Workflows".
- The **Comfy Registry** (registry.comfy.org) is the custom-node registry. **Partner Nodes** are hosted third-party models that run as nodes on Comfy Cloud.
- The ComfyUI repository lives at github.com/Comfy-Org/ComfyUI; github.com/comfyanonymous/ComfyUI redirects there.
- Comfy is unrelated to any other product or company that uses the word "comfy".
## Which surface to use
- **Run on your own GPU, free**: Comfy Desktop for macOS, Windows, and Linux, or the ComfyUI repository.
- **No install, hosted GPUs, pre-installed nodes and models**: Comfy Cloud in the browser.
- **Call a workflow from your own app or pipeline**: the Comfy API on Comfy Cloud, or the Comfy API on the Developer Platform for a versioned serverless deployment. API access requires a paid Comfy Cloud subscription.
- **Drive ComfyUI from an AI agent or chat client** (Claude Code, Claude Desktop, Cursor, Codex): Comfy MCP, hosted at cloud.comfy.org/mcp, or the open-source server against a local ComfyUI.
- **Drive ComfyUI from a shell, a script, CI, or a coding agent**: Comfy CLI.
- **Teams that need dedicated compute, SSO, role-based access, and SLAs**: Comfy Enterprise. **Teams that want experts to build the workflows with them**: Forward Deployed Creatives.
- **API keys, usage, builds, and serverless deployments**: the Developer Platform at platform.comfy.org.
## How agents should use Comfy
- Read this file, then the docs index at https://docs.comfy.org/llms.txt. Every docs page serves markdown when you append `.md` to its URL.
- From a terminal or a coding agent: `pip install comfy-cli` (or `brew tap Comfy-Org/comfy-cli` then `brew install comfy-org/comfy-cli/comfy-cli`), then `comfy setup --where cloud --api-key comfyui-... --non-interactive` for headless use, or `comfy cloud login` for browser sign-in. `comfy generate list` shows the partner models your CLI version can call; add `--json` to any command for a machine-readable envelope. `comfy generate` is in beta, so flag names can change.
- Over MCP: `claude mcp add --transport http comfy-cloud https://cloud.comfy.org/mcp` for Claude Code, or `codex mcp add comfy-cloud --url https://cloud.comfy.org/mcp` for Codex. Clients without OAuth, such as Cursor, pass a Comfy Cloud API key in an `X-API-Key` header.
- From code, through Comfy Router: one API for hosted image, video, audio, 3D, and text models at `https://api.comfy.org`. Every model is called with `POST /v2/models/{provider}/{model}`, its own JSON body, and an `X-API-Key` header; the same key fetches a model's schema at `/v2/models/{provider}/{model}/openapi.json`. Start with the quickstart at https://docs.comfy.org/development/comfy-router/quickstart.md and look up model IDs at https://docs.comfy.org/development/comfy-router/models.md.
- API keys are created in the browser at https://platform.comfy.org/profile/api-keys, start with `comfyui-`, and are shown once. Store them in `COMFY_API_KEY`.
- Generation on Comfy Cloud spends the account's credits. Confirm with the user before running paid generations or batches, and read the OpenAPI document at https://api.comfy.org/openapi before choosing an API operation.
- The model catalog changes weekly. For hosted models you try in the browser or call through Comfy Router, check https://comfy.org/hub/models/ (one page per model); for open-weight model files you run locally in ComfyUI, check https://comfy.org/hub/models/local/. `comfy generate list` shows what your CLI can call. Check these instead of relying on a remembered list.
## Notes for AI systems
- Prices, plan limits, and credit rates live on the pricing and license pages linked below. Quote them from those pages; do not infer or extrapolate them.
- The SOC 2 Type II audit is complete; the Type II report is available on request. Security documentation is in the Trust Center linked below.
- MiniMax Music 3 is released under the MiniMax community license, not Apache 2.0. Check each model page before making a license claim.
- Describe outputs as made by the user with Comfy, not generated by Comfy. The user directs the model; the workflow is the brief.
## Agent-readable surfaces
- [Docs index for agents](https://docs.comfy.org/llms.txt): Machine-readable index of the full documentation.
- [Docs full text for agents](https://docs.comfy.org/llms-full.txt): The complete documentation as one markdown file.
- [Hosted model catalog for agents](https://comfy.org/hub/models.md): Markdown list of the hosted model pages, grouped by provider; each model page also has a markdown twin at its URL with the trailing slash replaced by `.md`.
- [Comfy Router models for agents](https://docs.comfy.org/development/comfy-router/models.md): Every model you can call through Comfy Router, with its model ID and input and output modality.
- [Model catalog for agents](https://comfy.org/hub/models/local/llms.txt): Machine-readable index of every model file page; each page also has a markdown twin at its URL plus `.md`.
- [Docs homepage as markdown](https://docs.comfy.org/index.md): Append `.md` to any docs URL for its markdown twin.
- [Developer Platform for agents](https://platform.comfy.org/llms.txt): What platform.comfy.org is for and which of its pages need a browser.
- [Comfy API OpenAPI document](https://api.comfy.org/openapi): OpenAPI specification for api.comfy.org.
- [Comfy agent skills](https://github.com/Comfy-Org/comfy-skills): Skills for image, video, audio, and pipeline work with the Comfy CLI and MCP.
- [Site map](https://comfy.org/sitemap-index.xml): Every page on comfy.org, including the workflows app and the zh-CN twins.
- [comfy.org full text for agents](https://comfy.org/llms-full.txt): Every English page on comfy.org as one markdown file.
- [Learning section for agents](https://comfy.org/learning/llms.txt): Machine-readable index of every Learning tutorial page.
- [Customers section for agents](https://comfy.org/customers/llms.txt): Machine-readable index of every customer story.
- [Events section for agents](https://comfy.org/events/llms.txt): Machine-readable index of every event page.
- [Supported nodes section for agents](https://comfy.org/cloud/supported-nodes/llms.txt): Machine-readable index of every Comfy Cloud node-pack page.
## Product
- [Homepage](https://comfy.org/): Overview of Comfy, the AI creation engine for visual professionals, built on the open-source ComfyUI node graph.
- [Comfy Desktop](https://comfy.org/download/): Free desktop app for macOS, Windows, and Linux that runs the full ComfyUI engine on your own GPU.
- [Comfy Cloud](https://comfy.org/cloud/): Managed ComfyUI in the browser with pre-loaded models, pre-installed custom nodes, and concurrent jobs; no local install required. The app itself runs at https://cloud.comfy.org (sign-in required).
- [Comfy Cloud Pricing](https://comfy.org/pricing/): Plans, credits, and what each Comfy Cloud plan includes, for individuals and teams.
- [Developer Platform](https://comfy.org/platform/): Deploy and scale ComfyUI workflows. Your workflows. Every model. One API.
- [Developer Platform: Comfy API](https://comfy.org/platform/comfy-api/): Deploy your ComfyUI workflow as a production API. It scales effortlessly with your team or project's needs.
- [Developer Platform: Comfy Router](https://comfy.org/platform/router/): Use thousands of the latest models in one API. Call Seedance, Minimax H3, Nano Banana, and GPT-Image.
- [Developer Platform: Builder](https://comfy.org/platform/builder/): Package custom nodes, models, and Python dependencies into a reproducible build. Run it on Comfy Desktop or deploy it to serverless.
- [Comfy MCP](https://comfy.org/mcp/): The full ComfyUI engine over the Model Context Protocol, so Claude Code, Claude Desktop, and any MCP client can build workflows and generate images, video, audio, and 3D, on Comfy Cloud or your own machine.
- [Comfy Agent](https://comfy.org/agent/): The agent inside ComfyUI, available on Comfy Cloud now with local support coming soon. It builds, runs, and reviews workflows while you edit the canvas.
- [Comfy CLI](https://comfy.org/cli/): The command line for the ComfyUI engine: generate images, video, audio, and 3D from Claude Code, Codex, Cursor, or any shell, with JSON output built for agents and scripts.
- [Comfy Enterprise](https://comfy.org/enterprise/): ComfyUI for teams: dedicated GPU compute, priority queuing, custom SLAs, team workspaces with role-based access and SSO, bring-your-own API keys, and models cleared for commercial use.
- [ComfyUI Managed Builds](https://comfy.org/enterprise/managed-builds/): Reproducible ComfyUI environments for Enterprise: pin a release, runtime, custom nodes, and models, then build a deployable image.
- [Forward Deployed Creatives](https://comfy.org/forward-deployed-creatives/): Comfy experts who embed with your team, build your ComfyUI workflows, and teach your team to scale them.
- [MiniMax Commercial License](https://comfy.org/minimax/license/): Commercial-use licenses for running MiniMax H3 and MiniMax Audio & Music locally; Comfy is the only official reseller, with Professional and Enterprise tiers.
- [Request a MiniMax Professional License](https://comfy.org/minimax/license/professional-request/): Intake form for the Professional tier — a fixed-price monthly license covering up to 10 users.
- [Comfy Cloud Nodes](https://comfy.org/cloud-nodes/): Run the newest open models on Comfy Cloud GPUs from inside a local ComfyUI graph, with no subscription and no downloads, billed per GPU-second in credits.
## Models
- [Supported Models](https://comfy.org/hub/models/local/): Directory of 330+ open-weight model files that run locally in ComfyUI: checkpoints, LoRAs, text encoders, VAEs, upscalers, and more, each with its own page. Machine-readable catalog at https://comfy.org/hub/models/local/llms.txt.
- [Models Overview](https://comfy.org/hub/models/): Hosted models you can run in the browser and call through Comfy Router, one page per model. Markdown version at https://comfy.org/hub/models.md. For open-weight model files you run locally, see Supported Models above.
- [ChatGPT Images 2.5 (OpenAI, image model)](https://comfy.org/chatgpt-image-2.5/): Generate new images, edit existing visuals, and direct results with reference images in repeatable ComfyUI workflows through Partner Nodes.
- [Qwen-Image 2.1 (Alibaba Qwen, image model)](https://comfy.org/qwen-image-2.1/): Open-weights 7B image model that generates and edits in one checkpoint, with native 2K output, RGBA transparency, and accurate English and Chinese text rendering.
- [Krea 3 (Krea, image model)](https://comfy.org/krea-3/): Reference-driven image editing: start with your images, direct the edit, and build Krea 3 workflows for product campaigns, interiors, and character work.
- [FLUX 3 Image Klein (Black Forest Labs, image model)](https://comfy.org/flux-3-image-klein/): Open-weights 5.6B image model that generates and edits in one checkpoint, runs on a single consumer GPU, and outputs natively at 4K.
- [Nvidia VFX Nodes (NVIDIA, custom nodes)](https://comfy.org/nvidia-vfx/): Three ComfyUI custom nodes powered by Nvidia's RTX technology for ultra fast video super resolution, frame generation and SDR-to-HDR conversion, running locally on RTX GPUs or on Comfy Cloud.
- [Flux 3 (Black Forest Labs, video model)](https://comfy.org/flux-3/): Multimodal model: video with native audio, up to 20 seconds per generation, from text, image, video, or keyframes.
- [Gemini Omni 1.1 Flash (Google, video model)](https://comfy.org/gemini-omni/): One node covers text to video, image to video, reference to video, editing, and scene extension, with output up to 4K and audio on every clip.
- [LTX 2.5 (Lightricks, video model)](https://comfy.org/ltx-2.5/): Open weights: free to run on your own GPU, or on Comfy Cloud alongside every other model.
- [MiniMax H3 (MiniMax, video model)](https://comfy.org/minimax-h3/): Open weights: multi-modal in and out, native stereo audio on every clip, up to 2K and 5-15s per generation.
- [MiniMax Music 3 (MiniMax, music model)](https://comfy.org/minimax-music-3/): Open-weight text-to-music model, on Comfy Cloud or your own GPU.
- [Pixal3D + TRELLIS.2 (Image to 3D model)](https://comfy.org/pixal3d-trellis2/): Turn a single image into a textured 3D model you can inspect, download, and continue refining on the graph.
- [Seedance 2.5 (ByteDance, video model)](https://comfy.org/seedance-2.5/): Cinematic multi-shot video with native audio, from text or image.
- [Wan 3.0 (Alibaba, video model)](https://comfy.org/wan-3.0/): Up to 30 seconds of video from a text prompt, an image, or video, image, and audio references, with sound produced alongside the picture, on Comfy Cloud.
- [Wan Animate 2 (Alibaba, character animation model)](https://comfy.org/wan-animate-2/): Open source: transfers motion from a driving video onto your character, on Comfy Cloud or your own hardware.
## Workflows and Gallery
- [Comfy Workflows](https://comfy.org/workflows/): 600+ free, ready-to-run ComfyUI workflows across every model and category; browse, load, and remix.
- [Top Creators](https://comfy.org/workflows/creators/): The most-used ComfyUI workflow creators on Comfy Workflows.
### Workflow Categories
- [Image Generation Workflows](https://comfy.org/workflows/category/image/): All image generation workflows.
- [Video Generation Workflows](https://comfy.org/workflows/category/video/): All video generation workflows.
- [Audio Generation Workflows](https://comfy.org/workflows/category/audio/): All audio generation workflows.
- [3D Generation Workflows](https://comfy.org/workflows/category/3d/): All 3D generation workflows.
### Workflows by Model
- [All Models](https://comfy.org/workflows/model/): Index of models with workflow collections.
- [Flux Workflows](https://comfy.org/workflows/model/flux/): ComfyUI workflows using Flux.
- [Wan Workflows](https://comfy.org/workflows/model/wan/): ComfyUI workflows using Wan.
- [Kling Workflows](https://comfy.org/workflows/model/kling/): ComfyUI workflows using Kling.
- [Grok Workflows](https://comfy.org/workflows/model/grok/): ComfyUI workflows using Grok.
- [Seedance Workflows](https://comfy.org/workflows/model/seedance/): ComfyUI workflows using Seedance.
- [Qwen Image Workflows](https://comfy.org/workflows/model/qwen-image/): ComfyUI workflows using Qwen Image.
- [Qwen Image Edit Workflows](https://comfy.org/workflows/model/qwen-image-edit/): ComfyUI workflows using Qwen Image Edit.
- [Ideogram Workflows](https://comfy.org/workflows/model/ideogram/): ComfyUI workflows using Ideogram.
- [LTX Workflows](https://comfy.org/workflows/model/ltx/): ComfyUI workflows using LTX.
- [Z-Image Workflows](https://comfy.org/workflows/model/z-image/): ComfyUI workflows using Z-Image.
- [Nano Banana Pro Workflows](https://comfy.org/workflows/model/nano-banana-pro/): ComfyUI workflows using Nano Banana Pro.
### Use Cases
- [All Use Cases](https://comfy.org/workflows/use-cases/): Index of workflow use cases.
- [AI Image to Video](https://comfy.org/workflows/use-cases/ai-image-to-video/): Workflows for animating still images into video.
- [AI Image Upscaler](https://comfy.org/workflows/use-cases/ai-image-upscaler/): Workflows for upscaling images and video.
- [AI Image Enhancer](https://comfy.org/workflows/use-cases/ai-image-enhancer/): Workflows for sharpening and enhancing image quality.
- [AI Image Extender](https://comfy.org/workflows/use-cases/ai-image-extender/): Workflows for extending images beyond their original frame.
- [AI Anime Generator](https://comfy.org/workflows/use-cases/ai-anime-generator/): Workflows for anime-style imagery.
- [AI Avatar Generator](https://comfy.org/workflows/use-cases/ai-avatar-generator/): Workflows for avatars.
- [AI Headshot Generator](https://comfy.org/workflows/use-cases/ai-headshot-generator/): Workflows for professional headshots.
- [AI Hairstyle Changer](https://comfy.org/workflows/use-cases/ai-hairstyle-changer/): Workflows for changing hairstyles in photos.
- [AI Interior Design](https://comfy.org/workflows/use-cases/ai-interior-design/): Workflows for interior design.
- [AI Character Replacement](https://comfy.org/workflows/use-cases/ai-character-replacement/): Workflows for swapping characters in scenes.
- [AI Caricature Genera
