Spaces:
Running on Zero
Running on Zero
| title: RynnWorld-4D | |
| emoji: 🌀 | |
| colorFrom: gray | |
| colorTo: indigo | |
| sdk: gradio | |
| sdk_version: 6.20.0 | |
| app_file: app.py | |
| short_description: 4D world model — synced RGB, depth & optical-flow video | |
| python_version: "3.10" | |
| startup_duration_timeout: 1h | |
| # RynnWorld-4D | |
| Interactive demo of [Alibaba-DAMO-Academy/RynnWorld-4D](https://huggingface.co/Alibaba-DAMO-Academy/RynnWorld-4D), | |
| a 4D embodied world model that generates **synchronized RGB, depth, and optical-flow** | |
| videos from a reference image + text prompt using a tri-branch diffusion transformer | |
| built on [Wan2.2-TI2V-5B](https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B-Diffusers). | |
| Pick one of the built-in samples (each is a reference-image latent + prompt validated by | |
| the authors), preview its reference frame, and generate the three aligned streams. | |
| The inference path is a faithful port of the official `inference-sft.py` | |
| (`fusion_mode=joint`, 3D RoPE joint attention, frame-wise, layers 0–30 every 3). | |