ComfyUI
Open-source node-graph engine for local Stable Diffusion, Flux and video AI
Overview
ComfyUI is the open-source (GPL-3.0) node-based engine for generative media: every stage of a diffusion pipeline is a visual node, workflows run locally on your own GPU and serialise into shareable, reproducible graphs. Built by pseudonymous developer comfyanonymous in January 2023, it is now stewarded by San Francisco-based Comfy Org alongside a Desktop app, hosted Comfy Cloud and a production API.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 22 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- Full node-graph engine, no feature gates
- Runs SD, SDXL, SD3.5, Flux, video, audio and 3D model families
- ComfyUI Manager + Comfy Registry custom nodes
- NVIDIA, AMD, Intel Arc, Apple Silicon support
- 4,200 monthly credits
- 30-minute max runtime per workflow
- 1 concurrent workflow via API
- Top-up credits anytime (valid 1 year)
- 7,400 monthly credits
- Custom LoRA importing
- 3 concurrent workflows via API
- 21,100 monthly credits
- 1-hour max runtime per workflow
- 5 concurrent workflows via API
- Enterprise-grade infrastructure
- Custom deployment and support
What you can produce with ComfyUI
- Build a text-to-image pipeline as an explicit node graph — checkpoint loader, prompt encoders, sampler, VAE decode — and re-run it deterministically with every parameter exposed.
- Drag any ComfyUI-generated PNG back onto the canvas to rebuild the exact workflow that produced it; the graph is embedded in the image metadata.
- Run current open-weight model families locally — SD1.x through SD3.5, SDXL, Flux, Stable Cascade, video (Wan, Hunyuan Video, LTX-Video, Mochi), audio (Stable Audio, ACE Step) and Hunyuan3D — typically with day-zero support.
- Extend the graph with community nodes installed through ComfyUI Manager and the Comfy Registry, which now scans submissions for eval/exec calls, runtime pip installs and obfuscated code.
- Run on the hardware you have: NVIDIA CUDA, AMD ROCm, Intel Arc, Apple Silicon, Ascend NPU — or --cpu as a fallback the docs themselves flag as slower.
- Skip the Python setup entirely with the free Desktop app for Windows and macOS, or script installs with the official comfy-cli.
- Lift a finished graph unchanged into Comfy Cloud (hosted RTX 6000 Pro, 96GB VRAM) or expose it as a production endpoint through the Comfy API.
ASEAN Perspective
ComfyUI in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
ComfyUI is for building generative-media pipelines as visual node graphs, giving you full control over the process. It turns every stage of a diffusion workflow into an editable node, runs the open-weight model families named in its README (SD1.x–SD3.5, SDXL, Flux, Stable Cascade, Wan/Hunyuan/LTX video, Stable Audio, Hunyuan3D — checkpoints downloaded separately, none bundled), and serialises whole workflows into JSON that ships inside the output images for exact reproduction.
It's aimed at professionals and power users who need parameter-level control, want reproducible workflows, and demand first-day access to new open models — it is the closest thing local generative AI has to a de facto studio standard. Anyone who just wants a prompt box should look elsewhere, as the node graph has a genuine learning curve, and a heavily extended install needs active tending as custom-node dependencies drift. GPU strongly recommended; CPU mode exists but the docs themselves call it slower.
Availability: free and open source under GPL-3.0 (copyleft — redistributing modified builds means sharing source; generated images are unaffected), as a Desktop app (Windows/macOS), Windows portable build, manual install or comfy-cli. Comfy Cloud is hosted on RTX 6000 Pro 96GB GPUs at US$20/month (Standard, 4,200 credits), US$35 (Creator, custom LoRAs), US$100 (Pro, 1-hour runtimes) and custom-priced Enterprise, billed only for active GPU time; partner-model API nodes draw on the same credits.
What people say
A familiar arc plays out all over r/StableDiffusion. Someone will open ComfyUI, recoil at the spaghetti of nodes and wires, and retreat to Automatic1111. They almost always come back within months, though, because the thing they wanted to build was only possible in ComfyUI.
The praise it receives is very specific. Every stage of a diffusion pipeline is an editable node, so nothing hides behind a form, and switchers consistently report faster generation on identical hardware. Workflows serialise to JSON and travel inside the PNGs they produce; drag a finished image onto the canvas and the exact graph that made it rebuilds itself. That reproducibility, plus day-zero support when new open-weight models drop — Flux, SD3.5, Wan and Hunyuan video, audio, even 3D — is why studio pipelines standardised on it. Comfy.org's own customer wall lists Netflix, Ubisoft, Amazon Studios and Apple.
Its drawbacks are just as concrete. A1111 gets a beginner generating in half an hour; ComfyUI demands you understand the pipeline before your first image. The community-node ecosystem, 60,000 strong by Comfy Org's count, is both the moat and the liability: packs pin conflicting torch versions, a workflow that ran Friday breaks Monday because one author pushed an update, and rolling back to fix one node breaks two others. Security has bitten for real — the ComfyUI_LLMVISION custom node shipped a credential stealer in June 2024, and Comfy Org's January 2025 security update owned that incident and the Ultralytics supply-chain compromise while rolling out registry scanning, eval/exec bans and obfuscation checks.
The project's journey from solo-dev passion to professional operation has been swift. Comfyanonymous shipped it solo in January 2023; Comfy Org has since raised $17M (May 2025) and $30M at a $500M valuation (April 2026, led by Craft), and now maintains a near-weekly release cadence (v0.28.0, July 2026), a Desktop app and a hosted Cloud. GPL-3.0 keeps forks honest without touching your generated images. The de facto professional tool, with homework attached.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including ComfyUI's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with ComfyUI unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to ComfyUI directly →
Spotted something out of date? Suggest an update →
More in Image Generation
