Model release · Tencent
Hy-Nomas
- Released
- Review updated
Tencent Hunyuan has released Hy-Nomas, an interactive video world model controlled through multi-turn text prompts and camera trajectories. It uses world memory for continuity and offers cinematic and robotics LoRA adapters. Training, inference and data-annotation code, along with model weights, are available under the project’s Apache 2.0 license. A low-memory configuration supports a single RTX 4090 using compatible INT8 weights, an FP8 decoder and CPU offloading with sufficient system RAM; GPU memory use depends on resolution and segment length. The technical report is still forthcoming.
Availability
Code and model weights are available. The project uses Apache 2.0 and provides an online demo and local deployment instructions.