HeyGen teaches brand video to AI agents with frame.md

HeyGen launches frame.md, an open-source Markdown format enabling AI agents like Claude or Gemini to generate brand videos via the HyperFrames framework.

Designed for the web, `design.md` files, which define colors, fonts, and spacing, poorly guide AI agents when it comes to video: these agents revert to their reflexes and produce animated pages or slides rather than true motion design.

HeyGen responds with `frame.md`, a Markdown specification that retains the same tokens but rewrites them for moving images: pacing, scale, dwell (the pause time on an element), transitions, and 16:9 composition. Summarized on 𝕏 as "taste, in a format an agent can understand," it is generated by pasting an existing `design.md` into the converter at `hyperframes.dev`, a template gallery that allows testing without an in-house design system. The entire system relies on HyperFrames, HeyGen's open-source framework (Apache 2.0): composition is written in HTML, CSS, and JS with temporal attributes, animations use GSAP, Lottie, or Three.js, and an engine coupling headless Chrome and FFmpeg renders a deterministic MP4, the same input always producing the same output. A single, one-time skill installation is then sufficient for an agent like Claude, Cursor, or Gemini to autonomously chain together shots, code, animations, and final rendering.