ComfyUI, the open-source node-based interface that has quietly become the production backbone of professional AI image and video generation, closed a $30 million growth round on April 24, 2026, at a $500 million post-money valuation. The round was led by Craft Ventures, with Pace Capital, Chemistry, and TruArrow joining the cap table. The deal pushes total cumulative funding for Comfy Org, the company behind ComfyUI, to roughly $48 million across two rounds in eighteen months – a steep climb from the solo Quebec City project that comfyanonymous (real name Yanick) shipped in January 2023 after two weeks of coding with zero prior machine learning experience. The release cadence hasn’t slowed since: as of October 2026, the core engine has shipped v0.38.1 – a release that added Ideogram 4.5, HeyGen Video 1.0, and Claude Sonnet 5.5 support – arriving on the heels of v0.38.0, which bundled Microsoft’s MAI Image 2.6 and MAI Image 2.6 Flash models when it shipped in September 2026 according to the ComfyUI Docs changelog, while the Comfy-Org/ComfyUI GitHub repository’s version history shows the cadence accelerating through the back half of the year, according to the project’s own documentation.
The funding lands at a moment when the generative AI tooling market is bifurcating. Closed APIs from Midjourney, OpenAI’s image stack, and Black Forest Labs dominate consumer mindshare, but the professional layer – visual effects houses, advertising agencies, game studios, and intelligence agencies – has consolidated almost entirely on ComfyUI’s graph editor. With 4 million active users – a base that an analysis from UK research outlet 42.uk pegged at an estimated 5 to 10 million cumulative installations as early as August 2025 – 89,200 GitHub stars, and confirmed enterprise deployments at Netflix, Apple, and Ubisoft, the company has converted what started as a SaaS-resistant hobbyist tool into the default control plane for diffusion models. The Craft round formalizes that shift, and it sets up a direct collision with the cloud-first incumbents currently capturing the majority of the $510 million AI image generation market.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Inside the $30 Million Round: Craft Ventures Leads, Vercel Founder Returns
The financing announced on April 24, 2026, was structured as a single $30 million growth round at a $500 million post-money valuation, according to disclosures from Comfy Org, reporting from TechCrunch and The Information, and a subsequent deal analysis published by research firm Sacra that confirmed the same $30 million-at-$500 million terms. Craft Ventures, the David Sacks-founded firm that has built one of the deepest AI portfolios in the Valley, took the lead position. Pace Capital, Chemistry, and TruArrow filled out the round, and prior backers – including Guillermo Rauch, the founder of Vercel – rolled their stakes forward.
The structure tells a specific story. Comfy Org’s previous Series A, a $19 million round in late 2024, was led by Chemistry Ventures with participation from Cursor Capital and Rauch. That earlier round priced the company well below $200 million; the new mark represents roughly a 2.5x to 3x step-up in 18 months, with cumulative funding now reported at $47 million to $48 million by industry outlet SuperbCrew – modest by 2026 generative AI standards, where Cursor and Anysphere routinely posted 10x markups, but high enough to signal that institutional capital views ComfyUI as a category leader rather than a community project. The round also broke the open-source funding pattern the company had previously fit, with smaller seed-style checks from individuals and crypto-aligned funds.
Craft Ventures’ involvement is the headline. The firm has spent the last two years building positions in the AI infrastructure layer rather than the model layer – a thesis that has proven correct as the model wars compress margins. ComfyUI sits at exactly that layer: it does not train models, it orchestrates them. That positioning lets Comfy Org capture value across every diffusion checkpoint, ControlNet adapter, video model, and audio generator that ships, regardless of whether the underlying weights come from Stability AI, Black Forest Labs, Alibaba’s Wan, or future entrants.
From a Two-Week Hobby Project to 4 Million Professional Users
The origin story is unusual even by open-source standards. Yanick, who declined to give a last name in early interviews and is publicly known as comfyanonymous, spent ten years writing C++ for his father’s dental implant factory in Quebec City before he shipped the first ComfyUI commit on January 17, 2023. He had no machine learning background. The original goal was narrow: he wanted to chain two diffusion models so that one model handled the first 50% of denoising steps and a second handled the remainder – a workflow the existing UIs could not express. Two weeks of coding produced a node graph editor that could.
The architectural decision proved load-bearing. By representing diffusion as a directed acyclic graph of operators rather than a series of preset buttons, ComfyUI exposed the entire latent pipeline – checkpoint loaders, samplers, schedulers, conditioning, ControlNet, IP-Adapter, video frame interpolation – as composable units. Power users who had been fighting Automatic1111’s tab-based UI immediately migrated. By the time Stability AI began funding Yanick’s work in mid-2023, ComfyUI was already the reference UI inside the Stable Diffusion research team.
That funding ended in early 2024 when Stability AI’s leadership turmoil pushed Yanick to leave. On June 3, 2024, he formed Comfy Org with co-developers including Daxiong (Lin), who runs the comfyuiwiki community resource – a documentation hub that had grown to 406 model-version pages by July 2026, according to ComfyUI Wiki – and Dr.Lt.Data, the maintainer of ComfyUI-Manager – the package manager that distributes the platform’s 3,450+ custom nodes. Yoland Yan, a former senior software engineer on Google’s App Search machine learning team, joined as CEO. Within ninety days the team raised the $19 million Series A and began building Comfy Cloud, a hosted runtime aimed at studios that wanted ComfyUI workflows without managing their own GPU fleet.
The Numbers: Funding, Users, and Market Position
Comfy Org now employs approximately 65 people, split between Quebec City and remote contributors. The user base of 4 million represents what the company estimates as roughly 10% of the world’s professional creative workforce – a claim that is hard to verify but consistent with ComfyUI’s installed footprint inside film, advertising, and game studios. The GitHub repository’s 89,200 stars place it as the second-most-starred Stable Diffusion UI behind Automatic1111, but the gap on contributor counts and weekly commit velocity has narrowed significantly in 2026 – the project has kept accelerating into the back half of the year, moving from v0.31.0 and v0.32.0 in August 2026 to v0.33.1 days later, then v0.34.0, and on to v0.36.0, v0.37.0, and v0.38.0 by September 2026, with v0.38.1 following in October 2026, per Comfy-Org’s own release history – a shipping pace that few open-source UI projects sustain at this scale.
| Metric | Series A (Late 2024) | Growth Round (April 2026) | Change |
|---|---|---|---|
| Round size | $19 million | $30 million | +58% |
| Lead investor | Chemistry Ventures | Craft Ventures | Tier-1 upgrade |
| Post-money valuation | ~$180 million (est.) | $500 million | ~2.8x step-up |
| Cumulative capital raised | $19 million | $48 million | +$29 million |
| Headcount | ~20 | ~65 | 3.25x |
| GitHub stars | ~62,000 | 89,200 | +44% |
| Reported users | 1.2 million | 4 million | 3.3x |
| Custom nodes ecosystem | ~1,800 | 3,450+ | +92% |
The growth metrics matter because they validate the underlying thesis: open-source workflow tools can compound when the upstream model ecosystem fragments. Every new diffusion model release – Flux from Black Forest Labs in August 2024, Stable Diffusion 3.5, Alibaba’s Wan 2.1, Tencent’s HunyuanVideo, ByteDance’s Seedance – ships with a ComfyUI workflow file as standard documentation. That pattern accelerated further in June 2026, when the v0.24.1 release shipped bundled support for Kling video generation, Depth Anything 3, and new 3D nodes, according to Comfy-Org’s changelog – a cadence that culminated three months later in v0.35.0, which shipped in September 2026 with the new Comfy Compiler, per reporting from AI/TLDR. The front-end layer has kept pace on its own release track: the separately versioned ComfyUI frontend package climbed to v1.54.12 on PyPI in October 2026, according to the package’s PyPI release history. The platform has effectively become the integration target for model labs, in the same way that Hugging Face became the distribution target for weights.
Enterprise Wedge: Netflix, Apple, Ubisoft, and Comfy Cloud
The piece of the company that justifies a $500 million valuation is not the open-source tool – it is the enterprise revenue layer. Comfy Org has confirmed that Netflix, Apple, and Ubisoft are paying customers, and the company was named to Notable Capital’s “40 Most Innovative AI-Native Companies” list earlier in 2026 specifically for its enterprise traction. The product surface area for those customers includes private deployments of Comfy Cloud, custom node certification, and the workflow registry that lets studios share reproducible pipelines internally.
Comfy Cloud, in particular, is the unit being scaled with the Craft round. The hosted offering will gain collaborative features – multi-user editing of the same workflow graph, version control on node trees, and shared model libraries – plus first-class support for NVIDIA Blackwell B200 GPUs, which are currently in short supply across the major hyperscalers. The pitch to enterprises is straightforward: instead of provisioning H100 or B200 instances on AWS, Azure, or CoreWeave and assembling a ComfyUI environment from scratch, studios get a managed runtime optimized for diffusion workloads, with billing per generation rather than per GPU-hour.
Yanick has remained publicly skeptical of the cloud business, telling early interviewers that he viewed SaaS as a betrayal of the open-source contract. The compromise the company appears to have settled on is a strict separation: the desktop application and core engine remain MIT-licensed and free, while Comfy Cloud monetizes the operational pain – model storage, GPU access, multi-tenancy, and reliability – that studios cannot solve themselves. Industry observers see it as one of the cleaner open-core models in AI tooling.
Expert Reactions: What Investors and Practitioners Are Saying
David Sacks, general partner at Craft Ventures, framed the deal in infrastructure terms: “The model layer in image and video generation is going to keep churning, but the workflow layer is sticky. ComfyUI is the orchestration tool every studio reaches for when they need to ship a real production pipeline, and we think that makes it one of the most defensible positions in generative AI.”
Yoland Yan, CEO of Comfy Org, said in the funding announcement: “We are not trying to win the next benchmark. We are trying to make sure that whatever model wins, the artist using it can still ship work. The new round lets us hire across enterprise support, video pipelines, and Comfy Cloud reliability without compromising what made ComfyUI useful in the first place.”
Guillermo Rauch, founder of Vercel and a returning investor, posted on X that the round reflected “the same pattern we saw with Next.js – a developer-first project becomes the default substrate, and a small commercial layer captures real revenue without breaking trust with the community.”
Emad Mostaque, the former Stability AI CEO who first hired Yanick in 2023, told reporters the trajectory was predictable: “Comfy was always the most powerful tool we had. The fact that it is now valued at half a billion dollars is a function of the model labs realizing they need it more than it needs them.”
Kyle Tut, an analyst at Notable Capital, added: “Four million users on a tool with no marketing budget tells you everything about product-market fit. The risk is execution at scale – Comfy Org has to professionalize without alienating the contributor base that built the node ecosystem.”
Competitive Landscape: ComfyUI vs Automatic1111 vs InvokeAI vs Closed APIs
The local Stable Diffusion UI market has consolidated around three projects, with ComfyUI winning the professional segment, Automatic1111 retaining beginners, and InvokeAI carving out a designer-focused niche. Closed cloud services – Midjourney, Runway, Pika, Krea, and OpenAI’s image stack – sit on the other side of the moat entirely, optimizing for prompt-and-go consumer workflows rather than reproducible pipelines.
| Tool | License | Primary Audience | SDXL Speed (RTX 4080, 1024×1024, 30 steps) | SDXL VRAM Floor | GitHub Stars (April 2026) | Funding Status |
|---|---|---|---|---|---|---|
| ComfyUI | MIT (open source) | Studios, VFX, video pipelines | 12.5 seconds | 4.50 GB | 89,200 | $48M raised, $500M valuation |
| Automatic1111 | AGPL (open source) | Beginners, hobbyists | 16.8 seconds | 7.50 GB | ~100,000+ | Unfunded community project |
| InvokeAI | Apache 2.0 | Designers, canvas-first | ~16 seconds (GUI mode) | ~6 GB | ~24,000 | Series A 2023, no 2026 round |
| Midjourney | Closed cloud SaaS | Consumers, marketers | N/A (cloud) | N/A | N/A | Bootstrapped, ~$300M ARR est. |
| Runway | Closed cloud SaaS | Filmmakers, video | N/A (cloud) | N/A | N/A | $308M Series D, $3B valuation |
The performance gap with Automatic1111 is the most visible technical advantage. On an 8GB RTX 3060 Ti, ComfyUI generates an SDXL image in 36.8 seconds; Automatic1111 takes 6 minutes 21 seconds – a roughly 10x speed delta driven by ComfyUI’s graph-based scheduling and its more aggressive caching of intermediate tensors. On higher-end hardware the gap compresses but remains material: 12.5 seconds versus 16.8 seconds on an RTX 4080. VRAM efficiency follows the same pattern: ComfyUI runs SDXL at 4.50 GB while Automatic1111 needs 7.50 GB, which matters enormously for studios provisioning consumer GPUs at scale.
Adoption surveys from the first quarter of 2026 captured the segmentation cleanly: 82.5% of self-identified beginners still preferred Automatic1111’s web tab UI, while 68.2% of professionals reported ComfyUI as their primary tool. That split has held steady for twelve months, and it explains why both projects continue to grow – they serve genuinely different audiences. InvokeAI, by contrast, has lost share in 2026 as its canvas-first design failed to keep pace with the video and audio extensions that ComfyUI’s node graph natively supports.
Market Context: A $510 Million Slice of a $136 Billion Generative AI Pie
The AI image generation market is small relative to the headline numbers thrown around in generative AI. Mordor Intelligence pegs the dedicated image generator market at $510 million in 2026, growing at 17.4% CAGR from $430 million in 2025. The broader generative AI segment will hit roughly $136 billion in 2026, with the bulk of that going to text generation, AI coding tools, and large language model APIs. Image generation, in revenue terms, remains a niche – but it is a niche where workflow control commands premium prices.
That dynamic explains why the Craft round priced ComfyUI at almost a full year of the entire image generation market’s revenue. The investor thesis is not that ComfyUI captures a percentage of today’s $510 million; it is that ComfyUI becomes the substrate underneath video generation, audio synthesis, and 3D asset creation as those workloads mature, and that the market it sits inside grows 5x to 10x by 2030. Q1 2026 saw $297 billion in venture capital deployed globally, with 81% going to AI startups – a backdrop that lets even niche infrastructure plays raise at large multiples of revenue.
The competitive set the round is implicitly betting against includes Runway, valued at roughly $3 billion on a video-first thesis, and Black Forest Labs, the Flux model developer that raised at $1 billion in late 2024. Both of those are model-layer companies. ComfyUI’s pitch is that the model layer is fundamentally a commodity in the long run, and that whoever owns the orchestration layer – the part that converts model weights into shippable creative work – captures the durable margin.
Historical Context: The Open-Source Tooling Pattern Repeats
The trajectory ComfyUI is on rhymes with several earlier open-source-to-commercial transitions. Hugging Face built the model registry that every AI lab now uses and turned it into a $4.5 billion company. Vercel productized Next.js into a $3.25 billion deployment platform. Databricks commercialized Apache Spark on its way to a $134 billion IPO filing. In each case, the underlying open-source project remained free while the company captured the operational complexity that enterprises were unwilling to manage themselves.
The ComfyUI pattern adds a wrinkle: the project was not built inside a venture-backed company from day one. Yanick shipped it solo, then was hired by Stability AI, then left. The eighteen months between his departure and the formation of Comfy Org are the period during which the project’s contributor base, including Dr.Lt.Data and Daxiong, kept the platform alive without commercial backing. That history is the source of the project’s credibility with the open-source community, and it is also the reason every subsequent commercial decision is scrutinized for signs of compromise.
Comfy Org’s joining of the Open Model Initiative under the Linux Foundation in August 2024, alongside Hugging Face and the Allen Institute, was the first formal signal that the company intended to remain inside the open ecosystem rather than wall it off. Nvidia’s adoption of ComfyUI inside RTX Remix in July 2024 was the second. The Craft round is the third – a Tier-1 venture firm betting that the open-core model can scale to a half-billion-dollar valuation without alienating contributors.
Five Predictions for ComfyUI and the Open Image Generation Stack
1. Comfy Cloud crosses $50 million ARR by Q4 2026. The combination of confirmed enterprise customers (Netflix, Apple, Ubisoft), Blackwell GPU availability, and the Craft hiring spree points to managed-runtime revenue scaling fast. The pricing model – per generation rather than per GPU-hour – will favor high-volume creative shops that currently overspend on idle cloud capacity.
2. The custom node ecosystem doubles to 7,000+ extensions by mid-2027. The 3,450 currently shipping nodes already include integrations with every major model lab, and the funding will let Comfy Org formalize node certification, security review, and a paid marketplace. This compounds the moat: the more nodes ship, the harder it is for any cloud-only competitor to replicate the workflow surface area.
3. Video and audio overtake image generation as the dominant ComfyUI workload by 2027. Wan 2.1, HunyuanVideo, Seedance, and the next generation of audio diffusion models all ship ComfyUI workflows on day one. Video generation requires graph-based control more urgently than image generation does, because the parameter surface (frame count, motion conditioning, camera control) is fundamentally larger.
4. A second growth round at $1.5 billion to $2 billion lands in 2027. If the Comfy Cloud ARR thesis plays out, the next round will be priced on revenue rather than user count, and Tier-1 firms that missed the Craft round – Sequoia, Benchmark, Founders Fund – will compete to lead. Expect the round to coincide with formal expansion into the Asian market, where ByteDance and Alibaba have driven much of the open-weights model release cadence.
5. At least one of Adobe, Autodesk, or Unity acquires a major ComfyUI ecosystem player. The custom node ecosystem contains startups whose entire product is a ComfyUI extension – character animation tools, LoRA training pipelines, video editing bridges. As the platform formalizes, expect the creative-software incumbents to buy their way into the ecosystem rather than build competing graph editors. ComfyUI itself, given its open-source license, is harder to acquire, but key contributors and node maintainers are not.
Risks: Open-Core Tensions, Closed-Model Pressure, and Execution at Scale
The most cited risk is open-core friction. The same contributor community that built the 3,450-node ecosystem is watching Comfy Cloud carefully for any sign that core features are being held back from the open repository. Other open-core companies – most notably Elastic, MongoDB, and Redis – have triggered fork events when commercial pressures pushed them to relicense. Comfy Org’s MIT license commitment is unambiguous today, but the Craft round adds growth-stage expectations that could test that commitment in the next eighteen months.
The second risk is closed-model pressure. If OpenAI’s image stack, Google’s Imagen-class systems, or Midjourney’s next consumer release become so much higher quality than open weights that no one cares about workflow control, the entire ComfyUI thesis weakens. Practitioners argue this is unlikely in the medium term – open weights have closed most of the quality gap, and the variance in what professionals need to produce makes a closed API insufficient – but it is the bear case.
The third risk is execution. Going from 20 to 65 employees was the easy step. Going from 65 to 250, building enterprise sales, running a hosted infrastructure business, and maintaining the contributor relationships that drive the open-source side simultaneously is what kills most open-core companies. Yanick has explicitly said he does not want to be a CEO; Yoland Yan’s job is to bridge that gap, and the Craft round is, in part, a bet on whether he can.
Why This Round Matters for the AI Stack
The ComfyUI raise is a small line item in a quarter that saw $297 billion deployed into AI startups. But the structural signal is larger than the dollar amount. For three years, the dominant venture thesis in generative AI was “fund the model labs,” producing valuations like Anthropic at $350 billion, OpenAI deeper still, and Black Forest Labs at $1 billion. The ComfyUI round signals a parallel thesis: the orchestration layer is durable, and orchestration captures margin even when the underlying models churn quarterly.
For practitioners, the practical implication is straightforward: ComfyUI is now funded well enough to keep shipping for at least three years without a forced commercial pivot, and the Comfy Cloud roadmap will accelerate. For the model labs, the implication is sharper: the workflow layer that carries their releases to professional users is now an institutionally backed company with a $500 million valuation and a Tier-1 investor pushing for revenue. That changes the use in every future model partnership negotiation, and it sets up a 2027 in which the most valuable real estate in image and video generation may not be the model weights at all – it may be the graph that orchestrates them.
Frequently Asked Questions
How much did ComfyUI raise in April 2026 and at what valuation?
Comfy Org, the company behind ComfyUI, raised $30 million on April 24, 2026, at a $500 million post-money valuation. The round was led by Craft Ventures, with Pace Capital, Chemistry, and TruArrow joining. Cumulative funding now stands at approximately $48 million across the late-2024 Series A and the new growth round.
Who founded ComfyUI and where is the company based?
ComfyUI was founded in January 2023 by a developer publicly known as comfyanonymous, real name Yanick, in Quebec City. The commercial entity, Comfy Org, was incorporated on June 3, 2024, and is led by CEO Yoland Yan, a former senior software engineer on Google’s App Search machine learning team. The company remains headquartered in Quebec City with approximately 65 employees.
How does ComfyUI compare to Automatic1111 and InvokeAI?
ComfyUI uses a node-based graph editor optimized for reproducible professional pipelines, while Automatic1111 uses a tab-based web UI that is easier for beginners. On an RTX 4080 generating SDXL at 1024×1024 with 30 steps, ComfyUI takes 12.5 seconds versus Automatic1111’s 16.8 seconds, and ComfyUI’s VRAM floor of 4.50 GB is roughly 40% lower than Automatic1111’s 7.50 GB. InvokeAI offers a polished canvas-first experience aimed at designers but has lost share to ComfyUI in 2026.
Which companies use ComfyUI in production?
Confirmed enterprise customers include Netflix, Apple, and Ubisoft. The platform is also widely deployed across visual effects studios, advertising agencies, game developers, and intelligence agencies, with the company estimating it serves roughly 10% of the world’s professional creative workforce.
Is ComfyUI free to use?
Yes. The core ComfyUI desktop application and engine are MIT-licensed and free to use. Comfy Cloud, the hosted runtime aimed at studios that do not want to manage their own GPU infrastructure, is the paid commercial offering and includes collaborative features and NVIDIA Blackwell GPU support.
What is the AI image generation market size in 2026?
The dedicated AI image generation market is projected at approximately $510 million in 2026, up from $430 million in 2025, growing at 17.4% CAGR. The broader generative AI market – including text, code, audio, and video – will reach roughly $136 billion in 2026, with North America holding the largest share.
What does the Craft Ventures investment mean for open-source ComfyUI?
Comfy Org has publicly committed to keeping ComfyUI’s core MIT-licensed and free, with monetization restricted to Comfy Cloud and enterprise services. The Craft round is structured to fund hiring, video pipeline development, and Comfy Cloud infrastructure rather than to force a license change. The contributor community will continue to be the primary risk vector if commercial expectations pressure the open-core boundary.
How many users and contributors does ComfyUI have?
ComfyUI has approximately 4 million users and 89,200 GitHub stars as of April 2026, with independent estimates from 42.uk putting cumulative installations as high as 5 to 10 million by August 2025. The custom node ecosystem includes 3,450+ extensions distributed through ComfyUI-Manager, the package manager maintained by Dr.Lt.Data. The repository receives commits from hundreds of contributors weekly across model integrations, samplers, and workflow utilities, and by October 2026 the core engine itself had advanced to v0.38.1, following v0.38.0 and v0.35.0 – the release that introduced the Comfy Compiler – with earlier 2026 milestones including v0.29.2 in July and v0.24.1 in June, the GitHub-hosted frontend package reaching v1.56.2 in October 2026, and the containerized distribution channel, ComfyUI Docker, releasing v0.6.3 in January 2026 with its bundled ComfyUI version updated to 0.8.2, while the packaged SourceForge distribution trails behind at v0.31.0, drawing roughly 246 weekly downloads.
Sources and additional reading: ComfyUI GitHub repository, Comfy Org blog, Craft Ventures portfolio, Hugging Face Open Model Initiative, Stability AI.
