As of Beijing time October 24, 2026, Tencent Hunyuan announced Hy-Nomas on X as an open-source free-form interactive video world model. The GitHub repository Tencent-Hunyuan/Hy-Nomas and the Hugging Face listing tencent/Hy-Nomas publish the code and weights; the official website hosts documentation and an interactive demo.
According to the model card and repository, Hy-Nomas generates a world one video segment at a time. You can introduce objects, change style or scene, describe actions, or supply camera trajectories as generation continues. A persistent key/value cache carries context between segments for longer-horizon consistency. Distilled inference uses eight denoising steps, with a documented default of 720×1280 output at 24 FPS.
The release includes optional LoRA adapters, including cinematic and robotics adapters that can be merged into a compatible floating-point DiT. The robotics LoRA is described as trained on a real-world robotic-arm dataset to strengthen robotics simulation in generated worlds. The INT8 single-GPU loader does not support loading a separate LoRA at inference time.
On hardware, the documentation separates two paths. The standard configuration uses eight-way sequence parallelism and is launched for eight GPUs in one inference group. The single-GPU low-memory configuration is documented for one NVIDIA GeForce RTX 4090: INT8 DiT weights and KV storage, an FP8 decoder, CPU offload, and sequence parallelism disabled. That 4090 path is a low-memory recipe—not a claim that the default eight-GPU job runs on one card. Related reading: EdenLive Is an Open-Source Real-Time World Model That Runs on One RTX 5090, Simular Agent S4 Evolves Interfaces Without Retraining the Model, and What HeyGen Video Does That Avatar V Doesn't.
As for local install cost, actual VRAM on your machine, demo latency, and how the robotics LoRA holds up on your clips—we don’t have that information yet. If you’ve tried it, you can send feedback through the site-owner contact link in the footer. Open weights and a public demo are not the same as a free hosted API.