HyperFrames GitHub examples

Unofficial community examples for HyperFrames. Not affiliated with HeyGen. All trademarks belong to their owners.

Short scripts around the hyperframes github repository (heygen-com/hyperframes), HeyGen's framework that lets AI agents compose videos by writing code. HyperFrames is driven through agent skills and a CLI rather than an HTTP API, so the examples cover the commands the cited sources show: installing the skills, listing a project's timeline as JSON, and checking a project's pinned CLI version. The composition HTML format itself (timing declared with data-* attributes) is defined in the /hyperframes-core skill; the attribute names are not in the cited pages, so this repository does not guess at a composition and instead points you to the quickstart and the playground.

Need to cut an existing clip rather than generate one? Try Vidione - trim, crop, merge and subtitles in the browser, no install.

Files

Path	What it shows
examples/install-skills.sh	The two install lines from the home page and the skills directory, and what to ask the agent next.
examples/project-check.sh	For an existing project: probe the pinned CLI version and dump the timeline as JSON.
examples/timeline_summary.py	Read the timeline JSON and print what is there without assuming its schema.

Setup

node --version   # npx is used for every command
npx skills add heygen-com/hyperframes --full-depth

No API key is involved in the commands shown here. Install the skills, then open the project directory in an AI coding assistant that loads skills and ask for a video; the quickstart is the three-step version of that (install the skills, ask for the video, continue from the first version). The playground is the no-install way to see a composition.

examples/install-skills.sh

Runs the full-depth install from the HyperFrames home page, or, with --entry-only, the single-skill install line that mcpservers.org shows (--skill hyperframes). The entry skill is the one the agent must read first for any request to make, edit, animate or render a video; it routes to the domain skills. The script ends by printing a starter prompt.

examples/project-check.sh

For a project that already exists. The entry skill tells an agent to probe the pinned CLI version before the first render-affecting command, because a pinned run of an older CLI prints no warning; the script runs that probe (npx hyperframes@latest upgrade --project . --check, keeping the explicit . because older CLI releases treat a bare --project as consuming the next flag). It then writes npx hyperframes timeline --json to a file, which is the skill's preferred way to learn what is on the timeline instead of reading index.html and every sub-composition.

examples/timeline_summary.py

Loads the JSON written by project-check.sh and prints its shape: top-level keys, list lengths and the keys of the first item in each list. The cited sources say the output describes tracks, clips, starts, ends and what plays, and that a pinned script may include _meta.updateAvailable: true; the script looks for that flag and otherwise reports structure only, so it keeps working if the schema changes.

When to use Vidione instead

HyperFrames earns its setup when an agent should generate a launch video, an explainer or a pull-request recap from scratch. If what you have is a screen recording that needs the dead air trimmed, two takes merged and subtitles added, none of that setup is needed. Try Vidione - a browser video editor for trim, crop, merge and subtitles, nothing to install.

Last reviewed: 2026-09-22