Ant Group's Duo v1: a reciprocal world model for real-time full-duplex interaction
inclusionAI · hf · 2026-11-13
Ant Group releases Duo v1, a Reciprocal World Model (RWM) that continuously perceives audiovisual input while responding to users in real time, enabling full-duplex interaction. It treats Intent (cognitive reasoning and participation decisions) and Realization (coherent audiovisual behavior) as equally fundamental world-model capabilities, whereas existing world models mostly simulate only the latter. Three key designs: (1) autoregressive modeling for continuous interaction reasoning rather than diffusion-limited understanding; (2) chunk-specific conditions with Intent/Realization tokens synchronized on a unified timeline, with module-specific attention masks enforcing intra- and inter-chunk causality; (3) dual rollout distillation to mitigate distribution shift in both modules, trained on generated Intent tokens and synthesized audiovisual histories.
More from Multimodal
- Someone rebuilt The Simpsons house in real life with AI and it looks oddly plausible — TommycoolGun · 2026-11-13
- Engineer builds 3D Ayodhya model with Chisel, shares the file publicly — gowthami_s · 2026-11-13
- 67B MiniMax H3-Image runs on a 16GB MacBook Air via MLX with ~85% less peak memory — FlimsyBrief4232 · 2026-11-13
- Viral image prompt: watch the alien invasion from your penthouse — charis_ai · 2026-11-13
- G-STAR from PKU & Meituan LongCat: 20x faster autoregressive image generation at 2.5s per image — meituan-longcat · 2026-11-13
- Dev builds ComfyUI nodes to add batch generation to MiniMax H3 workflows — an80sPWNstar · 2026-11-13