Real-time world model demo
Real-time interactive video generation beats waiting for static renders.

Just a tweet linking to someone else's demo with no code or project to evaluate.
AI researchers, developers interested in world models
Real-time interactive video generation beats waiting for static renders.
Curated GPT-Image-2 prompts linking back to original Twitter threads.
Yet another AI video generator wrapper when Runway, Pika, and Luma already dominate.
JEPA orchestration framework solving a real but niche bottleneck—execution.
Microphone audio is captured in the browser as PCM and streamed over WebSockets to a Node server that pipes live segments into Mistral's Voxtral realtime API, then immediately hits DeepL for translation — a straightforward, usable demo of live STT->MT. It isn't reinventing the wheel, but the repo bundles the full flow (browser keys, Docker, server.mjs, and UI) so you can spin up an end-to-end test quickly; main downside is it's an integration demo that depends on external API keys and limits for real-world scale.
Unofficial fan site embedding the demo directly when original pages block iframes.