ComfyUI
Node-based workflows for image, video and audio generation, run locally or on Comfy Cloud
ComfyUI is an open-source (GPL-3.0) node editor for building generation pipelines out of models, references and post steps, so each stage of a shot is visible and repeatable. It runs free on your own GPU via Comfy Desktop or a manual install, or on Comfy Cloud, where paid plans start at $20/mo and include partner API models billed in credits. In 2026 it added Comfy Agent (beta) for building workflows from a chat request, and Comfy MCP so outside agents such as Claude can run it.
- 72 creators using
- 5 news items tracked
- 2023 first launched
- Freemium pricing
- Steep learning curve
What ComfyUI can do
- Infinite canvas Partial
- AI agent Partial
- Node / flow builder Yes
- Public API Yes
- MCP server Yes
20 of 26 capabilities confirmed
| Capability | Supported | Details |
|---|---|---|
| Text to image | Yes | Stable Diffusion, SDXL, Flux and partner image models |
| Image editing | Yes | Partner image-edit models such as GPT Image 2; Bria image edit nodes (v0.36.0) |
| Text to video | Yes | Wan 3.0, Seedance 2.5, MiniMax and other video models |
| Image to video | Yes | Wan 3.0 and Seedance 2.5 video nodes (v0.34.0 onward) |
| Video to video | Partial | Video Eraser node (v0.36.0) and trim, crop and concatenate nodes |
| Native audio in video | Not confirmed | |
| Lip-sync | Not confirmed | |
| Voice / TTS | Not confirmed | |
| Music & SFX | Yes | Yue2 music model support (v0.36.0) |
| Upscaling | Partial | Upscaler models via nodes and community packs |
| Character consistency | Not confirmed | |
| Camera control | Not confirmed | |
| Custom training | Partial | Own models on Creator plan and above |
| Third-party models | Yes | Partner Nodes: Claude Opus 5.5 and Sonnet 5.5, GPT-6 Sol and Luna, Seedance 2.5, Wan 3.0, Ideogram 4.5, FLUX 3 and others |
| Capability | Supported | Details |
|---|---|---|
| Infinite canvas | Partial | Infinite node canvas; it is not a moodboard or asset board |
| Node / flow builder | Yes | Node editor, App Mode for simplified views, subgraphs |
| AI agent | Partial | Comfy Agent (beta, Oct 1 2026) builds and runs workflows on the canvas; asks before running unless auto mode is on |
| Chat assistant | Yes | Comfy Agent chat, up to 5 parallel chats |
| Team collaboration | Partial | Team plan with shared credit pool and up to 50 members; workflows can be shared via MCP |
| Capability | Supported | Details |
|---|---|---|
| Public API | Yes | Developer Platform deploys a workflow as an API; Cloud API calls use credits |
| MCP server | Yes | Comfy MCP on Comfy Cloud (June 30 2026) and local comfy-mcp server (Aug 11 2026) |
| Mobile app | Not confirmed | |
| Desktop / local | Yes | Comfy Desktop and manual local install |
| Open source | Yes | GPL-3.0, about 137k GitHub stars |
| Capability | Supported | Details |
|---|---|---|
| Free tier | Yes | Free local app; Comfy Cloud gives 5 free runs to new users |
| Commercial use | Partial | Depends on each model's and custom node's license |
Yes Partial No Not confirmed by the vendor
What makes ComfyUI different
- 01
Open source under GPL-3.0 with a free local app; Comfy Cloud is optional and billed in credits
- 02
Every step of a generation is a visible node, so a workflow can be saved, versioned and rerun for the same shot
- 03
Partner Nodes put hosted models from several vendors (ByteDance, OpenAI, Anthropic, BFL, Ideogram) in one graph, paid from one credit pool
- 04
Comfy MCP and the Developer Platform let a saved workflow be called by an outside agent or deployed as an API
- 05
Community ecosystem of 60,000+ nodes and 900+ pre-installed cloud models, with the compatibility risk that implies
Latest from ComfyUI
- 2w ago $25,000 Bounty Challenges Developers to Rebuild Higgsfield Tools as Open Source Comfy Org launched a $25,000 developer bounty to build open-source video workflow alternatives to Higgsfield using the Comfy Developer Platform.
- 2w ago Krea 3 Joins ComfyUI via Official Partner Nodes ComfyUI integrates Krea 3 as an official Partner Node, bringing multi-reference image generation, rapid iteration, and targeted editing into modular production graphs.
- 2w ago FLUX 3 Image Klein brings 5.6B open weights and multi-reference editing FLUX 3 Image Klein integrates with ComfyUI, offering a 5.6B parameter model with base and distilled weights for bounding box and multi-reference image editing.
- 2w ago Nvidia VFX nodes bring local upscaling and frame interpolation to ComfyUI ComfyUI integrates Nvidia VFX nodes for local video upscaling, frame interpolation, and SDR-to-HDR conversion on RTX GPUs.
- 3w ago ComfyStreamerH3 Runs Live Video Generation on a Single GPU ComfyUI developer builds ComfyStreamerH3, a custom workflow running real-time video-to-video generation using the MiniMax H3 model on a single consumer GPU.
About ComfyUI
ComfyUI gives a filmmaker a visual graph of the whole generation process: load the model and references, run the step, inspect the output, then continue. Its value is that the steps are explicit and saveable, so a shot can be rerun or adjusted without rebuilding it from memory. It is not a guarantee of a visual result. A workflow that works once still needs review for identity, geometry and motion errors before it is used at scale.
In 2026 the product moved well beyond a local Stable Diffusion tool. Partner Nodes call hosted models from OpenAI, Anthropic, ByteDance, Black Forest Labs, Ideogram and others inside the same graph, paid in Comfy Credits. Comfy Agent (beta, launched October 1, 2026) builds and edits the graph from plain-language requests on the open canvas. Comfy MCP, which launched on Comfy Cloud on June 30, 2026 and as a local server on August 11, 2026, lets Claude Code, Cursor and similar tools drive ComfyUI directly. App Mode gives a simplified front end for a finished workflow.
For a film or video team, the practical split is local versus cloud. A local install costs nothing beyond the hardware, electricity and maintenance, but model speed depends entirely on the GPU and the chosen model. Comfy Cloud removes the hardware problem and bills by credits, with plan credits that reset monthly and do not roll over. Partner-model runs also spend credits, so a cost estimate should be made per accepted shot, not per test render. Check the license of each model and custom node before using client material.
What's new in ComfyUI
-
Comfy Agent launches in beta
A chat agent plans, places and runs workflows on the canvas; new accounts get some free chats, and a Desktop version is promised within weeks.
Source: comfyui-wiki.com ↗ -
ComfyUI v0.39.0 adds partner nodes and offline flag
New partner nodes for Anthropic Sonnet 5.5, Ideogram 4.5, HeyGen Video 1.0 and BFL FLUX 3 Image, plus a --offline flag.
Source: github.com ↗ -
ComfyUI v0.38.0 adds Seedance 2.5 Draft mode
Seedance 2.5 Draft mode, Claude Opus 5.5, GPT-6 Sol and Luna, and Seedream 5.0 Flash partner nodes, and Qwen-Image 2.1 work.
Source: github.com ↗ -
ComfyUI v0.35.0 adds Comfy Compiler and 3D import
Introduces the Comfy Compiler, WAN3-Prime support, VideoTrim and VideoCrop nodes, and a File3DToMesh node for GLB, GLTF, OBJ and STL files.
Source: github.com ↗ -
Comfy MCP runs against local ComfyUI
The open-source comfy-mcp server (v0.10.0 on PyPI) lets agents such as Claude Code and Cursor run workflows on a local install at 127.0.0.1:8188.
Source: comfyui-wiki.com ↗ -
Comfy MCP launches on Comfy Cloud
A Model Context Protocol server at cloud.comfy.org/mcp exposes model and template search, partner-API generation and job control; generations need an active subscription.
Source: docs.comfy.org ↗
Models available in ComfyUI
-
Seedance 2.5 Latest $$$ ByteDance Draft mode added in v0.38.0; video extension in v0.34.0
-
MiniMax H3 $ MiniMax Available as a partner node
-
Wan 3.0 Latest $ Alibaba Video nodes added in v0.34.0; WAN3-Prime in v0.35.0
-
GPT Image 2 — OpenAI Transparent background output added in v0.37.0
-
FLUX 3 Image Latest $ Black Forest Labs Partner node added in v0.39.0
-
Ideogram 4.5 Latest — Ideogram Partner node added in v0.39.0
-
Recraft V4 Styles $$ Recraft Added in v0.35.0
-
-
Meshy 7.1 Latest — Meshy Added in v0.37.0
-
Claude Opus 5.5 Latest — Anthropic Partner node in v0.38.0; text and prompt-side use
$ · $$ · $$$ = relative cost against other models of the same type, on fal.ai list prices. Cost inside ComfyUI depends on its own plans and credits.
Before you build a production workflow
Choose one representative shot, check its model requirements, and review the output before committing a whole sequence. This is a documentation-based guide, not a hands-on performance review.
Technical copy checked on 15 September 2026 by ReelStack using official documentation. Earlier blanket hardware and reproducibility claims have been corrected.
Plan the shots before building a workflow → · Explore creator lessons about ComfyUI →
Key Features
- Visual node editor with inspectable intermediate outputs
- App Mode: simplified front end for a finished workflow, switchable to the node graph
- Comfy Agent (beta): builds and runs workflows on the canvas from a chat request
- Comfy MCP: cloud and local servers for Claude Code, Claude Desktop, Cursor and Codex
- Partner Nodes for hosted models, paid in Comfy Credits
- Developer Platform: deploy a workflow as an API
- Subgraphs and reusable workflow blueprints
- Local install via Comfy Desktop or manual setup, GPL-3.0
- Comfy Cloud with 900+ pre-installed models and own-model upload on paid plans
When to reach for it — and when to skip
Reach for it when…
- Makes a multi-step generation process inspectable and repeatable
- Free local app with no per-generation fee
- Fast release cadence: partner models and video nodes arrive within weeks
- Agent and MCP access for teams already working in Claude or Cursor
Skip it when…
- Node graphs take time to learn; App Mode hides the detail until you need it
- Local speed depends entirely on your GPU and model choice
- Custom nodes and frequent updates can break saved workflows
- Credits are spent on partner models and cloud GPU time; per-shot cost is hard to forecast
- Comfy Agent and Comfy MCP are beta or early release
Best For
✓ Ideal for
- Filmmakers who need a repeatable, documented process for a series of shots
- Motion and VFX teams combining several models in one pass
- Studios that want local GPU generation with a free app and no per-render fee
- Teams building a template for a client's recurring deliverable
✗ Not built for
- Creators who want a one-prompt tool with no graph to manage
- Productions that need guaranteed output fidelity or identity lock without review
Working Tips from Filmmakers Using ComfyUI
- 01 Start with one representative shot and record the model, workflow version and settings before scaling.
- 02 Price each workflow per accepted shot, not per test render; partner models spend credits on every run.
- 03 Pin custom node versions and keep a known-working workflow before updating, since updates can break graphs.
- 04 Use the free local app to build and debug graphs, then move only the final runs to Cloud to save credits.
ComfyUI pricing in 2026
| Plan | Price | What you get |
|---|---|---|
| Local | Free none |
|
| Standard | $20/mo or $16/mo yearly monthly or yearly |
|
| Creator | $35/mo or $28/mo yearly monthly or yearly |
|
| Pro | $100/mo or $80/mo yearly monthly or yearly |
|
| Team | $700/mo or $630/mo yearly monthly or yearly |
|
| Enterprise | Custom custom |
|
Prices checked against ComfyUI's pricing page on 9 Oct 2026. Vendors change plans often, so confirm before you buy.
ComfyUI vs Replicate
| ComfyUI | Replicate | |
|---|---|---|
| Price | Local app free (GPL-3.0); Comfy Cloud Standard $20/mo ($16/mo billed yearly) | Pay per second of hardware: $0.000225/s (T4) to $0.001525/s (H100). No monthly plan on the pricing page; enterprise is custom. |
| Infinite canvas | Partial | No |
| Node / flow builder | Yes | Not confirmed |
| AI agent | Partial | No |
| Chat assistant | Yes | No |
| Team collaboration | Partial | Not confirmed |
| Text to image | Yes | Yes |
| Image editing | Yes | Not confirmed |
| Text to video | Yes | Yes |
| Image to video | Yes | Yes |
| Video to video | Partial | Not confirmed |
| Native audio in video | Not confirmed | Yes |
| Voice / TTS | Not confirmed | Yes |
| Music & SFX | Yes | Yes |
| Upscaling | Partial | Not confirmed |
| Custom training | Partial | Yes |
| Third-party models | Yes | Yes |
| Public API | Yes | Yes |
| MCP server | Yes | Yes |
| Desktop / local | Yes | Partial |
| Open source | Yes | Partial |
| Free tier | Yes | Not confirmed |
| Commercial use | Partial | Not confirmed |
| Starting price | Free local; Cloud from $20/mo | Pay per use |
| Pricing model | Local free; Cloud credits in monthly plans | Per run time or per output (image or second of video) |
| Video price example | Credits per run; RTX 6000 Pro at 957.94 credits/hr | Wan 2.1 i2v 480p at $0.09 per output second |
| Local install | Yes, GPL-3.0 | No; custom models can be packaged with Cog |
| Hosted third-party models | Yes, via Partner Nodes paid in credits | Yes, e.g. FLUX 1.1 Pro at $0.04 per image |
Each column comes from that tool's own researched page. "Not confirmed" means the vendor doesn't say.
Open the full ComfyUI vs Replicate comparison →Use Cases
Integrations
Tags
Sources
Facts on this page were checked against ComfyUI's own pricing, product and changelog pages on , then cross-checked with independent coverage where vendor pages were silent. Anything we couldn't confirm is marked "Not confirmed" rather than guessed.
- Comfy Cloud pricing page comfy.org
- ComfyUI GitHub releases github.com
- Comfy homepage comfy.org
- Comfy MCP documentation docs.comfy.org
- Comfy Agent launch coverage comfyui-wiki.com
- Comfy MCP local server coverage comfyui-wiki.com
- ComfyUI repository (license, stars) github.com
- Replicate pricing page replicate.com
Spotted something out of date? Tell us and we'll re-check it.
Discussion
No comments yet — be the first.