#4545th of 70 in Open modelsFor image, video and design, 3d
hy-world-2.0
Generates and reconstructs 3D worlds (meshes, Gaussian splats) from text, images or video
- Stars
- 2.9k
- License
- custom license
- Last commit
- Aug 2026
- Language
- Python
1,485 on 21 Apr 20262,850 on 10 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
HY-World 2.0 is Tencent Hunyuan's PyTorch code and weights for two jobs: reconstructing 3D from multi-view images or video with WorldMirror 2.0 (~1.2B), and generating navigable 3DGS or mesh worlds from text or one image. Generation chains panorama synthesis (HY-Pano 2.0), trajectory planning, WorldStereo 2.0 expansion and 3DGS training. It includes a Gradio demo for reconstruction.
Who it is for: Researchers and 3D developers running world generation or reconstruction
Strengths
- Outputs editable 3DGS, meshes and point clouds that can be imported into Blender, Unity or Unreal
- WorldMirror 2.0 predicts depth, normals, cameras, point clouds and 3DGS in one forward pass
- Optional camera and depth priors can be injected to improve reconstruction accuracy
- Diffusers-like Python API, multi-GPU CLI, and Gradio app for reconstruction
Weaknesses
- Heavy install: CUDA 12.8, FlashAttention build, custom gsplat, git submodule and navmesh extension
- HY-Pano-2 is ~80B and WorldStereo-2 ~17B, so generation needs large multi-GPU hardware
- World generation is a five-script pipeline, not a single command
- License is not identified (NOASSERTION), so usage terms need manual review
What it needs
- GPU required
- Needs CUDA 12.8, Python 3.11+, PyTorch, FlashAttention, gsplat, recastnavigation
- Models: WorldMirror 2.0 (~1.2B), HY-Pano 2.0 (~80B), HY-Pano-2-Qwen LoRA (~425M), WorldStereo 2.0 (~17B), Hugging Face safetensors
Quick answers
Is hy-world-2.0 open source?
Its code is public on GitHub under its own license terms; read them before you use it at work.
Does hy-world-2.0 need a GPU?
Yes, it needs a GPU.
Also in Open models
See all 70| Rank | Project | Score |
|---|---|---|
| 1 | ruviewWiFi CSI sensing on ESP32 for presence, breathing, heart rate and pose | 94 out of 100 |
| 2 | layaSingle-pass multilingual classifier for typed choice, score and yes/no decisions | 88 out of 100 |
| 3 | rf-detrReal-time transformer models for object detection, segmentation and keypoints | 79 out of 100 |
| #4 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 78 out of 100 |
| #5 | yolov5PyTorch YOLOv5 object detection, segmentation and classification models with training code | 76 out of 100 |