The notes — a program of boxes and wires, who wrote it, the company around it, and the licence it has kept, dated
A short history of ComfyUI
exhibit one hundred and fifteen The notes
Published 2026-11-16 (UTC)
A small (human) team and a fleet of AI agents.
One developer wanted better pictures of characters with fennec ears. A few months later he published a program that draws picture-making as boxes joined by wires. This page covers who started ComfyUI and when, how the graph idea works, the company that formed around it, its licence, and its milestones, each dated.
ask about this page → assistant.strata2signal.com · in beta, still being tested
the short version
ComfyUI is a free program that runs image models on your own computer by wiring small steps into a graph; its README now lists videos, 3D models and audio too. A developer known as comfyanonymous found Stable Diffusion in October 2022 and published the first commit on 2023-01-17 (UTC): nine kinds of box, under the GPL-3.0 licence the program still carries. He worked at Stability AI for about a year, resigned, and announced Comfy Org with five partners in a post dated 2024-06-18. That company says it has raised $47 million, the last round at a $500 million valuation, and that ComfyUI will always be free to run locally. Our print lab draws its pictures through it.
- The company announced a financing of $30 million (see It began with fennec ears)
2,791 words, about 13 minutes to read.
The summary is this page’s own; the receipt lines were drafted by a model on this workshop’s network and every figure in them is in the article, checked before this page went out — what was dropped, and why, is in this page’s receipt file.
If you're new here: strata→signal is a small workshop that builds on its own machines and writes up what it measures. This page reads; it does not measure. Every outside claim below is quoted from, or linked to, the post, file or record named beside it, all read on 2026-11-15 (UTC). Dates from GitHub and from the company's blog are UTC; a date from any other page is the date that page prints. The few sums we did are shown and marked as ours.
It began with fennec ears
In April 2026 the company behind ComfyUI announced "a $30 million financing at a $500 million valuation" (Comfy, 2026-04-24). Seven months earlier, announcing a smaller round, it had said where the program came from: "In October 2022, Comfyanonymous discovered Stable Diffusion and got hooked", and "The first experiments? Generating better images of characters with fennec ears." (Comfy, 2025-09-16). A fennec is a small desert fox with very large ears.
Stable Diffusion, the open image model, had been out since 2022-08-22 (our page Who makes FLUX has that story). The developer goes by comfyanonymous. Asked years later what he had been before, he answered: "Just a software engineer." He "hadn't written a line of PyTorch before that", PyTorch being the library these models run on (Latent Space interview, published 2025-01-04). The program everyone used then had a way to make a large picture in two passes. He wanted to try a different model on the second pass, and found that harder to add to that program than to write his own. The name came from his pictures: "people thought my pictures were comfy".
The public record starts on 2023-01-17 (UTC). GitHub shows the repository created at 03:15:56 UTC and the first commit, "Initial commit.", 21 minutes later, at 03:37:14 UTC. That was still the evening of 16 January in Quebec City, five hours behind UTC in winter (our conversion), and the project keeps its birthday on "Jan 16, 2023". The commit carries a second, older stamp, set when its author first made it: 2023-01-03 at 06:53:32 UTC, just under fourteen days earlier (13 days 20 hours 43 minutes, our subtraction). That fits the company's one-line history: "One developer in Quebec City built his own in two weeks and open-sourced it."
The commit held 77 files. Its README called the program "A powerful and modular stable diffusion GUI." Under "Why did you make this?" it said: "I wanted to learn how Stable Diffusion worked in detail. I also wanted something clean and powerful that would let me experiment with SD without restrictions."
How a graph makes a picture
Making a picture with a model like Stable Diffusion is a chain of small jobs: load the model, turn the prompt into numbers, start with an empty grid, fill it with noise and remove the noise over a set number of passes, turn the result into pixels, save the file. Most programs hide the chain behind one button. ComfyUI draws each job as a box, called a node, and you join one node's output to the next node's input with a wire. The finished drawing is the graph, or workflow. The first commit defined nine nodes in one 221-line file, and six of their names read like that chain: CheckpointLoader, CLIPTextEncode, EmptyLatentImage, KSampler, VAEDecode, SaveImage. The other three were VAELoader, VAEEncode and LatentUpscale. The canvas was borrowed: the commit includes a 14,132-line build of litegraph.js, a graph editor under the MIT licence whose repository dates from 2013.
A graph buys two things a button cannot. You can rewire it: two models in a row was the experiment that started the project. And the program can skip work. The first README promised: "Only parts of the graph that change from each execution to the next will be executed". Change the last step, and only the last step runs again. The graph also travels with the result: "Dragging a generated png on the webpage or loading one will give you the full workflow including seeds that were used to create it." Because a graph is plain data, another program can hand one to ComfyUI: the darkroom of our print lab submits one graph for each visitor's picture (How the print lab works). The creator knew what he was trading away. Everyone else was making easy interfaces, he said, so "let me try to make a powerful interface that's not easy to use".
From a handle to a company
For its first seventeen months ComfyUI had no company of its own. In June 2023, by his own account in the same interview, Stability AI hired him. Its new model, SDXL, came as a pair, a base model and a refiner, first released for research on 2023-06-22, and running one model after another was what ComfyUI did.
In a post dated 2024-06-18 he wrote: "I have resigned from Stability AI and am starting a new chapter. I am partnering with mcmonkey4eva, Dr.Lt.Data, pythongossssss, robinken, and yoland68 to start Comfy Org." The post set the aim, "the best free open source software project to inference AI models", and gave his view of the alternative: "I personally believe that closed source AI is a dead end and a waste of time."
Despite the name, Comfy Org is a company with investors, and the footer of its website places it in San Francisco. It announced "$17 million in funding" on 2025-09-16 and the $30 million round on 2026-04-24, "bringing our total funding to $47 million" (17 + 30 = 47, our check). The second post also puts a name to the handle: it quotes "Yannik Marek, cofounder and the creator of ComfyUI". The plan for paying the bills is stated plainly. Of its hosted service, Comfy Cloud, the company wrote: "Cloud is the way we will make revenue and sustain the long-term development of ComfyUI." The sentence before it begins "ComfyUI will always be free to run locally".
Control widened slowly. "For the longest time, the repo only had a single person (comfyanonymous) to review and guarantee code quality," the company wrote on 2025-12-29, announcing the repository's move from his personal account to the Comfy-Org organisation, where it lives now.
The licence, and what it does not cover
ComfyUI is under the GNU General Public License, version 3 (GPL-3.0). The licence file was in the first commit, and the copy in the repository today is the same file, byte for byte (our comparison of the two). The licence calls itself "a free, copyleft license". It lets anyone use, change and share the program, on a condition it puts this way: "if you distribute copies of such a program, whether gratis or for a fee, you must pass on to the recipients the same freedoms that you received." GitHub's records show GPL-3.0 on two newer pieces too: the interface and the node manager. The desktop app is the exception. Its first repository, under GPL-3.0, is archived, and the licence file of its successor begins "Comfy Desktop is dual-licensed": the GNU Affero General Public License, version 3 or later, or "A commercial license from Comfy Org".
The licence does not cover the models. ComfyUI is the program; each model it loads comes with its own terms. Our print lab's painter, FLUX.2 [klein] 4B, is under Apache 2.0, while FLUX.1 [dev] and FLUX.2 [klein] 9B carry a non-commercial licence (Who makes FLUX; our licence ledger keeps a row for every model we run or bench). A free program does not make every model free for every use.
Nor does it cover what the community adds. Custom nodes are separate projects by their own authors. Partner Nodes, announced on 2025-05-06 under their first name, API nodes (renamed in the README on 2026-08-12), call paid models that run on other companies' computers; they are optional. Of the program itself the README says: "Runs fully offline: core does not download anything unless you request it."
The milestones, dated
| Date (UTC) | What happened |
|---|---|
| 2023-01-17 | The first commit: nine nodes, GPL-3.0 |
| 2023-06 | Stability AI hires the creator, by his account, as it prepares SDXL |
| 2024-06-18 | He resigns and announces Comfy Org with five partners |
| 2024-07-12 | v0.0.1, the first numbered release |
| 2024-08-01 | FLUX.1 support is committed on the model's launch day |
| 2024-10-21 | A packaged desktop app enters closed beta; its code is opened on 2024-11-27 |
| 2025-01-03 | The Registry, a public list of custom nodes, opens to installs through ComfyUI-Manager; nodes had been published to it since 2024 |
| 2025-05-06 | API nodes, now called Partner Nodes, announced: paid outside models, callable from a graph |
| 2025-08-07 | Subgraphs: a group of nodes folded into one |
| 2025-09-15 | Comfy Cloud opens in private beta; it leaves beta on 2026-03-04 |
| 2025-09-16 | $17 million raised, the company says |
| 2026-03-10 | App Mode: a graph hidden behind a simple form |
| 2026-04-24 | $30 million more, at a $500 million valuation |
Dates are UTC, from GitHub's stamps and the company blog's, except two: 2023-06 is the creator's account, and 2024-06-18 is the date his post carries.
The pace is a milestone of its own. The first numbered release came nearly eighteen months after the first commit (542 days, our count). From then to v0.45.0 on 2026-11-10, GitHub's release list holds 165 numbered releases: about one every five days (851 days ÷ 164 gaps, ours). Versions that got a tag but no release page are not counted.
Support can arrive the day a model does. We dated two cases. The FLUX.1 commit, "Basic Flux Schnell and Flux Dev model implementation.", is stamped 2024-08-01, the day Black Forest Labs launched. And our bake-off of FLUX 3 Image Klein ran on v0.41.0, which GitHub dates to 2026-10-15, that model's release day.
What this cannot say
- How many people use it. "4 million users", "60,000+ nodes built by contributors" and "150,000+ daily downloads" are the company's own counts, from its funding post of 2026-04-24, with no method given. GitHub showed 147,009 stars when we read it on 2026-11-15 (UTC); a star is a bookmark, not a user. No source here measures ComfyUI's share of local image generation, so this page claims none.
- What the first two weeks held. The "two weeks" is the company's phrase. A commit's author stamp is written by the author's own computer: it fits the story and proves nothing alone.
- The Stability AI year. The hiring month and the reason come from one interview with the creator, read in the interviewers' transcript. We found no statement from Stability AI.
- Whether a custom node is safe. A custom node is someone else's code running on your computer. One of the company's posts says custom nodes can "even be used for security hacks", and another, a post-mortem of 2026-01-11, records malicious packs "downloaded a total of 790 times" before a ban. This page vouches for none.
- What the company is worth, or will do. The valuation is the company's announced figure. That ComfyUI "will always be free to run locally", and the "longterm open governance structure" the company says it will set up, are promises, dated above. This page records them and cannot test them.
- Anything about speed or quality. Nothing here was measured. Our picture benches that ran on ComfyUI are pages of their own, such as The same eighteen pictures, card by card. One more page of ours times the program itself: Tokens a second in ComfyUI, v0.45.0 against v0.40.0, on one RTX 3090 has two ComfyUI versions write text from the same language model on one card and counts the tokens each wrote in a second. It compares the two versions, not models.
What to take with you
- It started with fox ears. Pictures of "characters with fennec ears" in October 2022; a public repository on 2023-01-17 (UTC).
- A graph is a recipe you can see. Nodes are steps, wires carry results, and only what changed, and what comes after it, runs again.
- A ComfyUI picture can carry its own graph. Drag the PNG onto the canvas and the workflow comes back, seeds included. It works only while the file keeps its metadata; some pictures on our own shelf were saved without it.
- The program is GPL-3.0. The models are not. Read each model's licence before you sell anything made with it.
- "Org" is a company. It has raised $47 million by its own posts, and it has promised in public that local use stays free.
- The first line grew. In 2023 the README began "A powerful and modular stable diffusion GUI." When we read it on 2026-11-15 (UTC), it began "The most powerful and modular AI engine for content creation." Below that, it listed "images, videos, 3D models, audio, and more".
Who ran this, and thanks
Thanks to comfyanonymous and everyone who has contributed to ComfyUI, to Javi Agenjo's litegraph.js under the first canvas, and to the Comfy team for dated posts that are still up; none of them owed us anything. A small human team asked for this page; a fleet of AI agents read the sources and drafted it.
Sources
All read on 2026-11-15 (UTC).
- ComfyUI, the repository: https://github.com/Comfy-Org/ComfyUI. Its record through GitHub's API (created 2023-01-17T03:15:56Z; licence GPL-3.0; 147,009 stars at 20:07:39Z; owner Comfy-Org), read 2026-11-15 (UTC).
- ComfyUI, the first commit, "Initial commit." (author stamp 2023-01-03T06:53:32Z, committed 2023-01-17T03:37:14Z; 77 files; its README, LICENSE and nodes.py): https://github.com/Comfy-Org/ComfyUI/commit/220afe3310, read 2026-11-15 (UTC).
- ComfyUI, the README and LICENSE on the master branch: https://github.com/Comfy-Org/ComfyUI/blob/master/README.md and https://github.com/Comfy-Org/ComfyUI/blob/master/LICENSE, read 2026-11-15 (UTC).
- ComfyUI, the releases list (v0.0.1, 2024-07-12T17:58:55Z; v0.41.0, 2026-10-15T22:57:45Z; v0.45.0, 2026-11-10T21:42:42Z; 165 numbered releases by our count; GitHub's tag list for the repository held 224 version tags, 59 of them with no release page, also our count): https://github.com/Comfy-Org/ComfyUI/releases, read 2026-11-15 (UTC).
- ComfyUI, the commit "Basic Flux Schnell and Flux Dev model implementation." (committed 2024-08-01T13:49:29Z): https://github.com/Comfy-Org/ComfyUI/commit/1589b58d3e, read 2026-11-15 (UTC).
- GitHub's records for the neighbouring repositories, read 2026-11-15 (UTC): https://github.com/jagenjo/litegraph.js (created 2013-09-26, MIT; 13,854 of the 14,132 lines in the first commit's canvas file appear in its build file today, blank lines included, our comparison); https://github.com/Comfy-Org/litegraph.js ("a fork of the original litegraph.js"); https://github.com/Comfy-Org/ComfyUI_frontend and https://github.com/Comfy-Org/ComfyUI-Manager (GPL-3.0 each); https://github.com/Comfy-Org/desktop (GPL-3.0; archived, its description beginning "[ARCHIVED] Superseded by"); https://github.com/Comfy-Org/Comfy-Desktop and its LICENSE file (created 2026-02-13; "Comfy Desktop is dual-licensed.", AGPL-3.0-or-later or a commercial licence from Comfy Org).
- comfyanonymous, "The next chapter for ComfyUI", 2024-06-18: https://blog.comfyui.ca/comfyui/update/2024/06/18/Next-Chapter.html, read 2026-11-15 (UTC).
- Comfy, "Comfy Raises $17M Funding", 2025-09-16: https://blog.comfy.org/p/comfy-raises-17m-funding, read 2026-11-15 (UTC).
- Comfy, "ComfyUI raises $30M to scale open-source AI for creative production", 2026-04-24: https://blog.comfy.org/p/comfyui-raises-30m-to-scale-open, read 2026-11-15 (UTC).
- Comfy, the second-anniversary post, ComfyUI Turns 2: A Journey and Call for Talent, 2025-01-16 (the "Jan 16, 2023" birthday; title given without its emoji): https://blog.comfy.org/p/comfyui-turns-2-a-journey-and-call, read 2026-11-15 (UTC).
- Comfy, About ("One developer in Quebec City built his own in two weeks and open-sourced it."; the footer's "San Francisco, USA"): https://comfy.org/about/, read 2026-11-15 (UTC).
- Comfy, "Comfy Cloud", 2025-09-15, and "Comfy Cloud Is Out of Beta — and It's Just Getting Started", 2026-03-04: https://blog.comfy.org/p/comfy-cloud and https://blog.comfy.org/p/comfy-cloud-is-out-of-beta-and-its; and the service's own page, titled "Comfy Cloud - AI in the Cloud": https://www.comfy.org/cloud, read 2026-11-15 (UTC).
- Comfy, "ComfyUI repo will move to Comfy Org account by Jan 6", 2025-12-29: https://blog.comfy.org/p/comfyui-repo-will-moved-to-comfy, read 2026-11-15 (UTC).
- Comfy, "ComfyUI V1 Release", 2024-10-21 (with the Registry's "over 600 published nodes"), and "Open sourcing v1 Desktop", 2024-11-27: https://blog.comfy.org/p/comfyui-v1-release and https://blog.comfy.org/p/open-sourcing-v1-desktop, read 2026-11-15 (UTC).
- Comfy, "Launching ComfyUI Registry", 2025-01-03 (the "security hacks" sentence): https://blog.comfy.org/p/launching-comfyui-registry, read 2026-11-15 (UTC).
- Comfy, "Introducting ComfyUI Native Partner Nodes (and New Brand!)", 2025-05-06 (the title's spelling is the source's; the post was later edited to the newer name, and its address still reads "native-api-nodes"): https://blog.comfy.org/p/comfyui-native-api-nodes, read 2026-11-15 (UTC).
- ComfyUI, the commit "chore: replace api nodes -> partner nodes in README (#15519)" (committed 2026-08-12T06:21:08Z): https://github.com/Comfy-Org/ComfyUI/commit/945ffca32e, read 2026-11-15 (UTC).
- Comfy, "Subgraph Official Release! Making Complex Workflows Clean and Efficient", 2025-08-07: https://blog.comfy.org/p/subgraph-official-release, read 2026-11-15 (UTC).
- Comfy, "From Workflow to App: Introducing App Mode, App Builder, and ComfyHub", 2026-03-10: https://blog.comfy.org/p/from-workflow-to-app-introducing, read 2026-11-15 (UTC).
- Comfy, "Upscaler-4K Malicious Node Pack Post Mortem", 2026-01-11: https://blog.comfy.org/p/upscaler-4k-malicious-node-pack-post, read 2026-11-15 (UTC).
- Latent Space, "AI Engineering for Art — with comfyanonymous, of ComfyUI", published 2025-01-04: a podcast transcript published by the interviewers, not by the maker, used here for the creator's own words: https://latent.space/p/comfyui/, read 2026-11-15 (UTC).
- Stability AI, the generative-models repository's README (SDXL's base and refiner, dated there June 22, 2023 and July 26, 2023): https://github.com/Stability-AI/generative-models, read 2026-11-15 (UTC).
- Our own pages, linked above: Who makes FLUX (Stable Diffusion's and FLUX.1's dates; the FLUX licences), How the print lab works, FLUX 3 Image Klein, ranked blind in a game's own painted style, The same eighteen pictures, card by card, Tokens a second in ComfyUI, v0.45.0 against v0.40.0, on one RTX 3090 and the licence ledger, read 2026-11-15 (UTC).