Proposal: branching lesson exports from pre-rendered Manim scenes

[jabrailkhalil]

I would like to build a lesson exporter/player that connects pre-rendered Manim scenes with questions and explicit branching. A learner's answer could select either the next explanation or a short correction scene.

I would start with three scenes and one question, a small lesson-graph manifest, and a browser player that records the current attempt locally. I would reuse rendered clips and keep arbitrary Python execution or runtime rendering out of the browser.

I know slideshow and notebook integrations already exist; the specific addition is a portable lesson graph with answer-dependent transitions. Would you prefer an independent integration, and are there existing players I should extend before starting a prototype?

I previously proposed Qwen Code's Android work, adjusted the design with its maintainers, and contributed the merged initial shell/PWA. I would use the same small-first approach here.

[nikolajmunk]

Hi Jabrail, this sounds like a cool project. I suggest you develop it independently of Manim. Good luck, please share your progress!

[jabrailkhalil]

Thanks, I'll develop it as an independent player around pre-rendered Manim clips.

I'll keep the first demo to three scenes and one question: a correct answer continues the explanation, while an incorrect answer opens a correction scene. I'll put the transitions in a small manifest and save the current attempt locally. The browser will only play the clips and handle answers.

I'll share a runnable lesson here once I have the player and branching tests working.

[jabrailkhalil]

I've built the independent prototype: live lesson and source.

I rendered three Manim scenes for one gradient-descent question. A correct answer opens the downhill explanation; an incorrect answer opens a correction and lets the learner retry. I keep the transitions in a versioned manifest and save the attempt locally, resetting it when the lesson content changes.

I ran 48 unit tests and 21 browser cases across Chromium, Firefox and WebKit, including actual video playback, reload/restart, denied storage, keyboard navigation and automated accessibility checks. All five CI jobs passed, including re-rendering and decoding the clips. I also checked playback and both branches on the published page in all three engines.

I've included the scene source and transcripts. This is a small self-study example; I haven't evaluated it in a classroom or conducted screen-reader user testing. I'd welcome feedback on the lesson manifest and the branching interaction.