Tencent releases Hy-Nomas, an open-source interactive world model runnable on an RTX 4090 Other

Tencent has released Hy-Nomas, an open-source foundation model for interactive video world generation that runs on consumer hardware down to a single NVIDIA GeForce RTX 4090. The release includes full training pipelines, inference code, interactive demos, and model weights under the permissive Apache-2.0 license.

Hy-Nomas constructs 3D environments sequentially across video segments while allowing real-time steering via text instructions and camera trajectories. At each segment transition, users can switch camera angles, scene elements, actions, or visual styles while maintaining spatial and contextual continuity in world memory. Generated outputs render at 720×1280 resolution at 24 frames per second.

Standard cluster deployments for Hy-Nomas use eight GPUs, but Tencent engineers optimized the model for single-card operation by pairing INT8 quantized weights and an INT8 KV cache with selective CPU offloading. For specialized domains such as robotics simulation and autonomous systems, the model supports parameter-efficient fine-tuning via LoRA adapters.

What it is

• Interactive world simulation. Generates cohesive multi-segment video scenes with continuous environmental state memory across transitions.
• Multimodal trajectory control. Accepts text commands and 6-DoF camera paths to guide camera angles, visual atmosphere, and scene events in real time.
• Single-GPU consumer inference. Operates on a single 24GB RTX 4090 using INT8 weight quantization, INT8 KV caching, and host CPU memory offloading.
• Permissive open source release. Distributed under the Apache-2.0 license on GitHub with model weights on Hugging Face.
• Robotics-ready adaptation. Includes LoRA fine-tuning recipes for domain-specific simulators, embodied AI, and robotics control environments.

Technical specifications

Sources

• Tencent Hy-Nomas GitHub Repository
• Tencent Hy-Nomas on Hugging Face
• Tencent Hy-Nomas Interactive Web Demo