HeyGen has introduced frame.md, a brand-spec format aimed at letting AI agents generate branded video and motion content that stays on-brand. The company is positioning it as a video-focused counterpart to design.md, the spec format already used to keep visual identity consistent across static screens.
design.md works for webpages and decks, but agents that try to use it for video output keep collapsing back into webpage and deck layouts. frame.md is the proposed fix for that translation gap.
design.md held the screen, but agents couldn't carry the brand into motion
In its announcement, HeyGen describes design.md as the format that "kept your brand consistent across screens." That worked when the agent's output was a webpage, a slide, or a static layout. The brand spec told the agent what to honor, and the rendered surface obeyed.
The breakdown, according to HeyGen's post, came when agents were pointed at video. Given a design.md, agents "translated it back into webpages and decks," defaulting to the static surfaces the spec was originally written for. The brand stayed intact, but the output stopped being video.
That is the gap HeyGen is naming. A brand spec written for screens does not carry timing, motion, scene transitions, on-screen text behavior, or any of the other choices that make a piece of content read as video rather than a slide deck on autoplay. Agents reading design.md had no instructions for those choices, so they reached for the surface they did know how to render.
frame.md is positioned as the video-and-motion counterpart, not a replacement
HeyGen describes frame.md as "a spec built for videos & motion" that "teaches your agents how to make branded video." The company's call-to-action in the post is to "turn your design.md into frame.md," which suggests frame.md is meant to extend an existing brand spec rather than replace it.
The announcement itself stops there. HeyGen has not, in the post, published a schema, a sample document, or a property-by-property breakdown of what frame.md contains. The post links out to a HeyGen URL for more detail, but the on-tweet description is the only surface verified here. Anything beyond "video and motion brand spec for agents" is implication, not stated fact.
frame.md extends HeyGen's agent-orchestration push for videohttps://www.vp-land.com/p…
HeyGen has been framing itself less as an avatar generator and more as an agent-orchestration layer for video. We covered the HeyGen Video Agent launch, where CEO Joshua Xu described the product as a "Creative Operating System" that runs script-to-final-cut through agents rather than a single generation model.
A brand-spec format is the logical next layer for that stack. An agent that can write, generate, and assemble a video still needs an instruction set for what the finished piece is supposed to look and feel like for a specific company. frame.md is HeyGen's pitch for what that instruction set should look like, sitting one level above the model and one level below the brief.
