ComfyUI raises $30M at a $500M valuation as demand grows for controllable AI content

ComfyUI has raised $30M at a $500M valuation, a good signal of where the generative media market is heading: investors are starting to value not only the models themselves but also the tools that give creators control over the result. While mass-market services sell the idea of "type a prompt and get a finished picture," ComfyUI bets on a different scenario: finely tunable pipelines for images, video, 3D and audio that can be repeated, edited and built into real production.

The ComfyUI team grew out of an open-source project that appeared in 2023, on the wave of the first diffusion models. Back then, generation often got even basic things wrong, and reaching the desired result was hard: a small change to a prompt could completely wreck an already good frame. Instead of yet another "magic button," the creators built a node-based interface where each stage can be controlled separately: models, LoRA, ControlNet, depth, inpainting and other blocks connect in a single graph. As a result, a workflow becomes not a throwaway prompt but a reproducible process that can be saved, moved and improved.

The new round was led by Craft Ventures, with participation from Pace Capital, Chemistry and TruArrow. According to the company, ComfyUI has more than 4 million users, over 60,000 nodes created by the community, and more than 150,000 downloads per day. These figures matter not only for PR but also for understanding the product: ComfyUI is developing not as a closed service with a fixed set of features, but as an ecosystem where the platform's capabilities grow along with the community. The faster new models and techniques appear, the more valuable a tool becomes that can quickly pick them up and assemble them into working chains.

The bet on ComfyUI is also understandable because generative content is moving beyond the experimental stage. The tool is now used not only by enthusiasts but also by studios, advertising agencies, VFX specialists and industrial design teams. In its official materials, the company cites Black Math and Silverside AI: some build modular creative pipelines for clients, while others use the platform in large advertising campaigns where repeatability, production speed and a consistent visual standard matter. For technical artists, ComfyUI works almost like a development environment: a workflow can be assembled from components, run on hundreds of assets and saved as a JSON file.

Against this background, the contrast with prompt-first products such as Midjourney, or with generation through chat interfaces, is especially visible. According to ComfyUI CEO Yoland Yan, an ordinary prompt often gets the result only 60–80% of the way to the desired state, and the remaining 20% becomes a lottery: any refinement can spoil parts of the image or video that were already good. This is where ComfyUI sells not "simplicity for everyone" but control for those who need a professional result. And the market seems to understand this: the company says job boards already list roles such as ComfyUI artist or ComfyUI engineer.

The broader market context matters too. Last year, Figma bought the startup Weavy, which also worked at the intersection of creative workflows and AI. This shows that the competition is not only for the best base models but also for the layer on top of them: interfaces, automation, collaboration and production infrastructure. If a model produces the raw material, workflow tools turn it into a managed process suited to studios, agencies and large teams.

For ComfyUI itself, the new money means a chance to speed up development, strengthen its cloud offering, improve the local experience and make the node ecosystem more stable. But for the market, something else matters more: investors valued at half a billion dollars not yet another AI generator, but a platform that makes generation controllable and reproducible. If this trend holds, the next big winner in AI creativity will not be whoever writes prompts best, but whoever best assembles models into a full production system.