Tencent Releases Hy-Nomas for Interactive Video World Generation

Tencent released Hy-Nomas, an interactive video world generation model, under the Apache 2.0 license, as detailed in its model card on Hugging Face. The model generates a world one video segment at a time and accepts new text instructions and camera controls between segments.

Users can add objects, change environments and visual styles, describe actions, and control camera movement as generation continues. Hy-Nomas uses a persistent key and value cache to retain context from earlier segments, while its distilled configuration generates 720 × 1280 video at 24 frames per second using eight denoising steps.

Tencent also provides a configuration that can run on one NVIDIA GeForce RTX 4090 using compressed model weights, a compressed cache, CPU offload, and an FP8 video decoder. The release includes Hy-Nomas-Bench, a suite of 280 test cases covering text interaction, camera navigation, combined text and camera control, and spatial memory.