Using Hyperframes to export AI-generated classrooms to video — architecture feedback welcome

[cosarah]

Hi Hyperframes team 👋

I'm a maintainer of OpenMAIC, an open-source project that generates interactive AI "classrooms" — a lesson is authored as structured data (slide DSL, scenes, timed teacher actions like narration / spotlight / laser / embedded video, plus generated media and TTS audio) and played back in the browser.

A frequently requested feature is exporting a generated classroom as a narrated video (eventually with subtitles). After evaluating options, we've decided to build this on Hyperframes as the rendering backend rather than growing a custom static-frame + FFmpeg pipeline. Your HTML/CSS/media + seekable-animation → MP4 model with deterministic headless-Chrome frame capture fits us well: we keep OpenMAIC classroom data as the single source of truth and add a compiler that emits a Hyperframes composition (an intermediate VideoTimeline IR → HTML + paused GSAP timelines + a self-contained asset ZIP), then render out-of-app via the producer.

I ran a small end-to-end spike (PNG base + a GSAP-driven SVG spotlight mask + dual audio tracks) and it rendered cleanly to MP4 — doctor / lint / validate all green on Node 22, and the paused window.__timelines + frame-seek contract worked exactly as documented. Really nice tool. 🙏

Our full design and phased plan live in this issue, which should be self-explanatory: THU-MAIC/OpenMAIC#854

I'd value your perspective on a few points:

1. Pre-1.0 stability — we plan to pin a version and isolate API churn behind our IR/emitter. Anything on the roadmap (especially the frame-adapter API, which is marked experimental) we should design around?
2. Determinism — we enforce no network and no Date.now / Math.random at render time, and vendor GSAP locally. Any other determinism footguns for frame-seek fidelity we should be aware of?
3. Producer as a service — we'd like to run the producer's HTTP server (POST /render + SSE) in an isolated Node 22 container for in-app one-click export. Is that a supported/recommended deployment shape, and any guidance on concurrency / resource sizing?
4. Anything you'd approach differently given the architecture in Hot Reload is not working #854.

Thanks for building and open-sourcing Hyperframes — happy to share back anything useful we learn from the integration.