Noiz AI and the Hong Kong University of Science and Technology (HKUST) announced on Monday, October 12, 2026, Helix World, a real-time interactive world model that generates video and spatial stereo sound simultaneously.
The system starts from a single image. The user can move with the WASD keys and look around with the arrow keys. The model produces the next stretch of video and the matching stereo audio at the same time, at 24 frames per second on a single NVIDIA H800 GPU. When the camera turns, the sound field follows.
Why sound matters in world models
Most current interactive world models generate silent video only. Sound, however, carries information the picture does not: the texture of a surface, the intensity of an impact, or events off-screen. In an immersive experience, turning your head should move the sound between your ears.
Adding audio after the fact does not solve the problem. Cascaded video-to-audio models never see the camera trajectory or the user’s actions and add extra latency. Helix World generates both modalities natively, conditioned on camera motion and keyboard input.
Performance and benchmarks
On the WBench navigation split, Helix World scores 79.9, close to the strongest silent models, with a real-time factor (RTF) of 0.77 — faster than real time. On HelixBench, the audio-visual benchmark released with the model, its spatial-acoustic consistency score reaches 41.8, compared with 12.7 for EchoWM, the only other world model in the comparison that generates sound.
Generation runs in chunks of about 1.3 seconds. New keyboard inputs or text prompts take effect from the next chunk, allowing the scene to be redirected while it is being generated. Output resolution is 768×512.
How to try it
An interactive demo is open at helixworld.org. The technical paper (arXiv:2609.38123) is available on arXiv, the code on GitHub, and a preview checkpoint on Hugging Face.
The model builds on the open-source LTX-2.3, with added 6-DoF camera control and keyboard actions, distilled into a causal student that generates in streaming without significant drift on long rollouts.
Limitations
Real-time speed was measured on an NVIDIA H800 data-center GPU; the paper does not report performance on consumer GPUs. The public checkpoint is a preview. On pure visual quality, the model sits slightly behind the best silent models, and its main advantage comes from the audio.
Interactive world models with camera-aligned spatial audio open paths to more immersive experiences in games, simulations, and real-time generated content. Helix World is one of the first to deliver this combination in a controllable and accessible way via a public demo.
Sources
- Official Noiz AI announcement
- Paper on arXiv
- Code on GitHub
- Interactive demo
- Preview checkpoint on Hugging Face
Transparency: This content was created, edited or reviewed with the aid of artificial intelligence. Information was cross-checked with public posts on X and sources available on the internet. Consult the original sources for the full context.
By GeekikiBot