hyperframes Hit 49,000 Stars by Treating Video as HTML You Have Not Rendered Yet

The Number That Made Us Look

hyperframes sits at 49,228 stars on GitHub. It was created on March 10, 2026 and gained 5,084 stars in the week of September 5 to 11 alone, ranking 10th on GitHub's star growth chart. That is not a one-time launch spike. The repo is 6 months old and still accelerating, with roughly 270 stars per day averaged across its lifetime.

The tagline is three words: "Write HTML. Render video. Built for agents." We read the source, checked the contributor data, and traced what sits underneath 49,000 stars.

What hyperframes Actually Does

hyperframes takes an HTML file with timing attributes and renders it into a deterministic MP4. The pipeline has three steps: parse the HTML composition, drive headless Chrome to capture each frame using the beginFrame CDP (Chrome DevTools Protocol) API, then encode the frames with FFmpeg. Audio tracks are mixed separately and muxed in at the end.

The key word is deterministic. Because the engine seeks to each frame mathematically (frame = floor(time * fps)) rather than recording live playback, identical input always produces byte-identical output. No dropped frames, no timing drift, no wall-clock dependency. That makes it CI/CD-friendly: you can diff video output the same way you diff screenshots.

<div data-composition-id="launch" data-start="0"
     data-width="1920" data-height="1080">
  <video class="clip" data-start="0" data-duration="6"
         src="intro.mp4" muted></video>
  <h1 id="title" data-start="1" data-duration="4">Launch</h1>
  <audio data-start="0" data-duration="6" src="music.wav"></audio>
</div>

It supports GSAP, CSS animations, Lottie, Three.js, Anime.js, the Web Animations API, and WebGL shader transitions. The animation framework is pluggable: you register a custom frame adapter and hyperframes calls it.

The Agent Angle Is Not Marketing

The repo ships with 20 pre-built skills for Claude Code, Cursor, Gemini CLI, and Codex. It includes .claude/, .cursor-plugin/, .codex/, and .agents/ directories, plus a skills-manifest.json that maps agent capabilities to hyperframes operations.

The insight is structural: LLMs are trained on more HTML than any other markup format. An agent that can write a web page can write a video composition without learning React, a timeline DSL, or a proprietary format. The composition is just HTML with data-start and data-duration attributes. That is a lower bar than any competing approach requires.

We could not find another video framework that ships agent integration as a first-class feature rather than a community add-on. Remotion, the closest competitor, requires React, which is a framework agents can write but do not prefer.

hyperframes vs. Remotion

Dimension	hyperframes	Remotion
Render (10s @1080p)	35 to 50 seconds (M2 Pro)	~25 seconds (Lambda)
Build step required	None	Yes (bundler)
Framework lock-in	Framework-agnostic HTML	React only
License cost (team of 10)	$0 (Apache 2.0)	$100/month
Ecosystem maturity	6 months old, pre-1.0	5+ years, stable API
Component library	Transitions, overlays, charts	Deep React ecosystem
Cloud rendering	Lambda/Cloud Run adapters (community)	First-party Lambda support

hyperframes wins on openness and agent ergonomics. Remotion wins on speed, maturity, and ecosystem depth. If you already have a working Remotion pipeline, there is no reason to switch.

What It Does Not Do

• No avatar generation. hyperframes renders compositions. HeyGen's commercial product handles AI avatars, voice synthesis, and translation. The open-source tool is the rendering primitive, not the full product.
• No audio-reactive animation. You hand-calculate data-start values. Syncing a beat drop to a visual requires manual timing.
• No first-party React adapter. An experimental Remotion integration exists but loses the determinism guarantees.
• Slow local renders. A 30-second 1080p composition takes roughly 3 minutes on a laptop. Cloud rendering via the Lambda or Cloud Run adapter packages is possible but not first-party.
• Pre-1.0. The current version is 0.8.36, with daily patch releases. APIs may shift between versions.

Who Built It

HeyGen, the AI video startup valued at $500 million after a $60 million Series A in June 2024. The company reported roughly $205 million in ARR as of mid-2026, according to Sacra estimates. hyperframes is the rendering engine underneath their commercial avatar product.

The repo has 59 contributors, but three HeyGen engineers account for 98.6% of commits in the past year: miguel-heygen (1,772 commits, 47% of total), vanceingalls (1,020), and jrusso1020 (907). Community contributions exist but are minor. The bus factor is 3.

One telling number: the Show HN post by maintainer Miguel07Code received only 3 points and 2 comments. The 49,000 stars came from other channels: the HeyGen community, AI agent developer circles, and organic GitHub trending. This is not an HN-driven star surge.

Verdict

Adopt now if: you are building agent-generated video, your team does not want to learn React, or your license budget is zero. The Apache 2.0 license and the HTML-as-composition model are genuine differentiators that no competitor matches.

Wait if: you need production-grade render speed, a stable API contract, or a deep component ecosystem. Remotion is more mature, and the $100/month is cheap insurance against breaking changes in a pre-1.0 tool.

What would change our mind: a 1.0 release with a stability commitment, first-party cloud rendering that matches Remotion Lambda's speed, and contributor growth outside of HeyGen's payroll. Right now, the project's future depends on one company's strategic interest in keeping it open. That interest is real today (it funnels users toward the paid avatar product), but interests change.