The model is presented as an interactive video world generator: users can continue a scene across video segments with text instructions and camera trajectories, while a persistent key/value cache carries prior context forward.
The published inference configuration uses eight denoising steps and produces 720 by 1280 video at 24 FPS. Tencent also provides a single-NVIDIA GeForce RTX 4090 configuration using INT8 DiT weights, an INT8 KV cache, an FP8 decoder, and CPU offload. The architecture combines a segment-autoregressive diffusion transformer, a Qwen text encoder, camera-conditioning modules, and a PixelUMM video decoder.
Tencent released Hy-Nomas-Bench alongside the model. It contains 280 test cases across four areas: multi-turn text interaction, camera navigation, interleaved text and camera control, and spatial memory.
The model card describes the release and its configurations, but does not establish comparative quality against other video models, reliable long-session performance, or throughput on a particular RTX 4090 system.