Noiz AI open-sources HelixWorld, a real-time interactive audio-visual world model
SimplyAnnisa · x · 2026-10-14
- Noiz AI open-sourced HelixWorld 1.0 (Apache-2.0, GitHub), a real-time interactive audio-visual world model: give it an image and a prompt, then walk (WASD) and look around (arrow keys) while video and 48kHz stereo audio update together in real time — sound is spatial and turns with the camera, not dubbed in post.
- You can edit the prompt mid-exploration to reshape the scene. 24 FPS visuals; a live demo and Discord are available. The team also released HelixBench, a benchmark for interactive audio-visual world models, on Hugging Face, testing whether synthesized sound fields track dynamic viewpoints.
More from Multimodal
- NVIDIA's PhyProbe scores physical consistency of AI-generated videos, beating prior evaluators — nvidia · 2026-10-14
- 95% of new Chinese short dramas are AI-made; the format behind +780% ROAS Meta ads — CurieuxExplorer · 2026-10-14
- Singapore startup uses AI to turn novels into graphic novels, reaching 60-70% before humans finish — lmoroney · 2026-10-14
- AI-generated realistic portrait of Cleopatra VII goes viral on Reddit — MeroveeFrancSalien · 2026-10-14
- Runway CEO shares fashion-tech video demo made with Runway — c_valenzuelab · 2026-10-14
- What working with AI in a professional creative pipeline actually looks like, per a practitioner — nptacek · 2026-10-14