Monetizing Workflows Over GPU Seconds

Diving deeper into

Coco Mao, CEO of OpenArt, on building the TikTok for AI video

Interview
while infrastructure companies are in a race to the bottom, we can maintain better margins at the application layer
Analyzed 5 sources

The key margin advantage sits in packaging cheapening models into a workflow that saves users hours, not GPU seconds. OpenArt uses open models and ComfyUI workflows under the hood, then sells a simpler credit based product that handles scripting, storyboards, character consistency, image to video, and editing for creators who do not want to wire nodes together themselves. That lets it charge for convenience, curation, and outcomes while model providers compete on raw generation cost.

  • ComfyUI is a node based open source tool for advanced users. It is useful when someone wants to chain many model steps by hand, tweak every setting, and build reusable generation graphs. OpenArt integrated it because those workflows are a fast way to power product features in the backend without exposing that complexity to mainstream users.
  • That makes ComfyUI more complementary than directly competitive today. ComfyUI serves the builder who wants the workshop. OpenArt serves the creator, marketer, or small business that wants a finished video or visual story with much less setup. The overlap is real, but the buyer intent and workflow are different.
  • The broader market is splitting this way. Products like OpenArt and Photoroom bundle the best available models into push button creative software, while tools like Runway and Pika are closer to direct generation environments that still ask users to assemble clips and edit outputs. As models improve and costs fall, product layer companies keep the margin only if they own the full creation workflow and audience.

Going forward, ComfyUI is likely to become part of the production stack for many application companies rather than the final destination for most users. The winners at the application layer will be the ones that turn messy multi model pipelines into a repeatable consumer product, then use that product to expand from tool into template library, creator network, and eventually distribution.