Hugging Face has introduced Zing-0.5, a 5B autoregressive world model built around playability: users can explore generated worlds, influence unfolding events, and respond to the resulting feedback through combined keyboard and online text control.
The work rests on three technical contributions. First, unified action and text conditioning combines magnitude-aware keyboard inputs with temporally aligned text instructions and jointly annotated videos, so navigation and event control are learned within the same sequence. Second, event-scale supervision for incremental generation uses a segment-level teacher trained on connected multi-prompt videos to supervise a block-level causal student through distribution-matching distillation. Third, low-cost real-time interaction pairs four-step generation with context-preserving streaming to support 832 x 480 inference at 24 FPS, at an estimated server rental cost of roughly USD 0.009 per stream-minute.
On 158 WBench Navigation cases, Zing-0.5 reports an overall score of 81.0 and a consistency score of 88.5. A joint-control demonstration shows a text-directed event change occurring during continued navigation without restarting generation.
The model weights, inference code, and the Zing-SGLang serving implementation are being released to support further work on playable generated worlds.