Tencent's Hunyuan team has released Hy-Nomas, an interactive world model that builds a world one video segment at a time. At each turn you can type a free-form instruction to add a character, start an action or change the style, or give a camera path to move through the scene. Earlier segments stay in memory to keep the world consistent.

It outputs 720p video at 24 FPS (the playback rate; generation speed depends on hardware). The standard setup uses eight GPUs, and a low-memory setup runs on a single RTX 4090. The release adds cinematic and robotics LoRAs, training code and Hy-Nomas-Bench, a test set of 280 cases for text control, camera navigation and spatial memory.

Model files are open under Apache 2.0 on Hugging Face.

@TencentHunyuan #WorldModel #OpenModels #AI

Model: https://huggingface.co/tencent/Hy-Nomas