How to Make a Launch Video with HyperFrames and a Coding Agent

Peter Yang talks with HeyGen’s Bin Liu and Jake Moran about HyperFrames, a tool for creating videos with coding agents. The guests describe free tooling, with optional external model and media services that may cost money; current availability and terms can change.

A HyperFrames composition combines HTML, CSS, JavaScript, footage, audio, and animation. Its renderer turns that composition into a video. The guests show ways to give a coding agent the relevant skill or use an available integration, then demonstrate a launch video built from a website and its assets.

Their website-to-video workflow gathers source material, writes a storyboard, and builds the video scene by scene. The team also mentions local text-to-speech and optional external providers. A simple initial request can produce a draft, but good results still depend on the source material and subsequent review.

Jake explains that a new launch may not have a finished website. He starts a project folder with product context, interface screenshots, visual references, and a design document. A frame-specific style file can translate brand colors, type, and visual guidance into directions better suited to motion.

Before building the full video, Jake asks the agent for a table of story beats and refines the message and copy. He reuses code for existing visual components, such as a prompt box or text effect, adapting the style and timing to the new project. The guests say their examples and components are available in their open-source materials.

Next he asks for a static HTML storyboard with a representative frame from each scene. Reviewing those frames helps settle the aesthetic before spending time generating a full animation. Once the direction works, he asks the agent to build the composition and preview it in the studio.

The studio breaks the video into scenes and lets a person inspect and adjust elements such as text, layout, and timing. The guests say edits are reflected in the underlying code, and describe exporting formats including MP4 and WebM. Specific features and formats should be checked in the current version.

Bin explains that the team moved toward web code after finding that a structurally valid data representation did not necessarily produce appealing visuals. He distinguishes spatial aesthetics—how a static page is arranged—from temporal aesthetics, or how information unfolds over time. The team uses evaluations and review loops to improve the latter, he says.

The guests mention launch videos, education, internal updates, and short summaries of agent work as possible uses. Jake recommends learning with very short clips and saving effective prompts and animations for reuse. Their main practical lesson is to combine clear product context, early visual review, and repeated iteration rather than expect a single prompt to guarantee a finished video.