Zing-0.5: Toward Playable Worlds with Real-Time Joint Action and Text Control
Zing-0.5 is a 5B world model built to let users steer generated scenes with both keyboard input and live text prompts.
The paper says the system combines magnitude-aware keyboard controls with temporally aligned text instructions, so navigation and event changes can happen in the same generated sequence. It uses event-scale supervision and distillation from connected multi-prompt videos to support incremental generation. The authors report 832 x 480 inference at 24 FPS, with estimated server rental cost around $0.009 per stream-minute. They also say they are releasing model weights, inference code, and serving implementation. HF Daily Papers' note
The paper says the system combines magnitude-aware keyboard controls with temporally aligned text instructions, so navigation and event changes can happen in the same generated sequence. It uses event-scale supervision and distillation from connected multi-prompt videos to support incremental generation. The authors report 832 x 480 inference at 24 FPS, with estimated server rental cost around $0.009 per stream-minute. They also say they are releasing model weights, inference code, and serving implementation. HF Daily Papers' note
score 6