James Russo, HyperFrames co-creator and engineering lead at HeyGen, describes using HTML, CSS and JavaScript as an authoring format for agent-generated videos. He contrasts this thin format with specialized languages and framework-specific instruction, arguing that familiar web code gives models more room to compose animation and visual media. These are lessons from his team's experiments rather than a controlled comparison proving HTML universally superior.
The rendering layer addresses the browser's asynchronous behavior. HyperFrames freezes its clock, advances through individual frames, waits for fonts and other assets to load, captures the page and encodes the result. This connects browser previews with deterministic video export and supports browser-native graphics such as SVG, charts and WebGL.
Russo separates a usable first draft from a polished video. Creative skills guide motion, pacing and branding, while storyboards, keyframes and a studio editor let people refine the output. He acknowledges that models still struggle with creative quality and describes work toward a code-to-video benchmark. Reported adoption numbers and output examples are the presenter's own claims, not independently verified performance measurements.
Watch on YouTube
