Per, founder of Scrimba (YC S20), says Scrimba Explain uses an LLM to create HTML-based explainer videos. HN.watch generates one the first time someone opens a Hacker News link. His hypothesis is that if video creation moves from “dollars and minutes” to “cents and seconds,” new use cases will unlock for people who prefer video to text.
What is Scrimba Explain? #
Scrimba spent the last decade teaching people how to code with an HTML-based video format. The team has now plugged an LLM into it so people can create explainer videos about anything.
HN.watch demonstrates the technology by replacing Hacker News articles with explainer videos generated on the fly.
Why does Scrimba use HTML-based video? #
Although HTML has obvious visual drawbacks compared with diffusion models, Scrimba identifies three major benefits:
- Speed: Much faster than pixel-based videos, taking just a few seconds from click to playback.
- Cost: Approximately**$0.04 per video** , excluding image generation, which some videos utilize and can quickly blow up the cost.
- Easy editing: These benefits also make AI-assisted editing cheap & fast.
What new use cases does Per see? #
Scrimba has already identified several possibilities:
- A video explanation of every single Pull Request, as Scrimba does internally
- A video for every page in internal or external documentation
- Turning a complex article into a video in ~4 seconds through its Chrome extension
- Helping course creators quickly draft lessons before recording the real thing
- Creating personal stories for children, wedding invitations, birthdays, and more
How does the custom stack work with LLMs? #
The stack is based on Imba, an open-source programming language created by Scrimba CTO Sindre Aarsæther. It compiles to JavaScript and interoperates fully with the npm + node ecosystem.
Scrimba also built its own sync engine, OP, and an agent context management system, Q. The team expected LLMs to struggle because these technologies were not well represented in training data, but LLMs have performed surprisingly well.
The stack is extremely dense and compact. In OP, a single declaration sets storage, sync, permissions, UI, and what the AI sees. This removes translations between frontend, API, db, and JSON that could confuse models and lead to errors.
Instead of using React.js, Express, Supabase, and LangChain, Scrimba built everything from scratch—admitting that it is suffering from the “not invented here” syndrome, lol.
Which models power Scrimba Explain? #
Scrimba uses Gemini, GPTs, Inworld, ElevenLabs, and a few others.
What can people try? #
People can access Scrimba Explain through:
- The Web UI: https://scrimba.com/explain
- [MCP](https://promptcube3.com/en/tags/mcp/) , which can be added to a coding agent
- [ChatGPT](https://promptcube3.com/en/tags/chatgpt/) Plugin
- Chrome Extension
A real pixel-based video demonstration is available on YouTube:
https://www.youtube.com/watch?v=k6rbHmBxSEs
More about Imba is available at:
https://imba.io/
What feedback is Per looking for? #
Per is interested in feedback and ideas for other use cases. He expects pushback from Hacker News because its community strongly favors text, and acknowledges that the tool is not for everyone. He also notes that many people prefer videos to text today, especially younger generations.
Next PaperMono e-ink magnet syncs shopping lists via phone web app →
All Replies （0） #
Want a live back-and-forth? Join the global AI chat room — login to talk. No replies yet — be the first!