[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-687":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":25,"hasPages":23,"topics":26,"createdAt":10,"pushedAt":10,"updatedAt":30,"readmeContent":31,"aiSummary":32,"trendingCount":16,"starSnapshotCount":16,"syncStatus":33,"lastSyncTime":34,"discoverSource":35},687,"HY-World-2.0","Tencent-Hunyuan\u002FHY-World-2.0","Tencent-Hunyuan","HY-World 2.0: A Multi-Modal World Model for Reconstructing, Generating, and Simulating 3D Worlds","https:\u002F\u002F3d-models.hunyuan.tencent.com\u002Fworld\u002F",null,"Python",2201,184,41,9,0,22,50,344,66,28.8,"Other",false,"main",true,[27,28,29],"3d","ai","worldmodel","2026-06-12 02:00:17","\n\n\u003Ch1>HY-World 2.0: A Multi-Modal World Model for Reconstructing, Generating, and Simulating 3D Worlds\u003C\u002Fh1>\n\n[English](README.md) | [简体中文](README_zh.md)\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Fteaser.png\" width=\"95%\" alt=\"HY-World-2.0 Teaser\">\n\u003C\u002Fp>\n\n\u003Cdiv align=\"center\">\n  \u003Ca href=https:\u002F\u002F3d.hunyuan.tencent.com\u002FsceneTo3D target=\"_blank\">\u003Cimg src=https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FOfficial%20Site-333399.svg?logo=homepage height=22px>\u003C\u002Fa>\n  \u003Ca href=https:\u002F\u002Fhuggingface.co\u002Ftencent\u002FHY-World-2.0 target=\"_blank\">\u003Cimg src=https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F%F0%9F%A4%97%20Models-d96902.svg height=22px>\u003C\u002Fa>\n  \u003Ca href=https:\u002F\u002F3d-models.hunyuan.tencent.com\u002Fworld\u002F target=\"_blank\">\u003Cimg src= https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPage-bb8a2e.svg?logo=github height=22px>\u003C\u002Fa>\n  \u003Ca href=https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.14268 target=\"_blank\">\u003Cimg src=https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FReport-b5212f.svg?logo=arxiv height=22px>\u003C\u002Fa>\n   \u003Ca href=https:\u002F\u002Fmodelscope.cn\u002Fmodels\u002FTencent-Hunyuan\u002FHY-World-2.0 target=\"_blank\">\u003Cimg src=https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FModelScope-Models-624aff.svg height=22px>\u003C\u002Fa>\n  \u003Ca href=https:\u002F\u002Fdiscord.gg\u002FdNBrdrGGMa target=\"_blank\">\u003Cimg src= https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDiscord-white.svg?logo=discord height=22px>\u003C\u002Fa>\n  \u003Ca href=https:\u002F\u002Fx.com\u002FTencentHunyuan target=\"_blank\">\u003Cimg src=https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FTencent%20HY-black.svg?logo=x height=22px>\u003C\u002Fa>\n \u003Ca href=\"#community-resources\" target=\"_blank\">\u003Cimg src=https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FCommunity-lavender.svg?logo=homeassistantcommunitystore height=22px>\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003Cbr>\n\u003Cp align=\"center\">\n  \u003Ci>\"What Is Now Proved Was Once Only Imagined\"\u003C\u002Fi>\n\u003C\u002Fp>\n\n## 🎥 Video\nhttps:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fb56f4750-25c9-48fb-83ff-d58526711463\n\n## 🔥 News\n\n- **[April 16, 2026]**: 🚀 Release HY-World 2.0 technical report & partial codes!\n- **[April 16, 2026]**: 🤗 Open-source WorldMirror 2.0 inference code and model weights!\n- **[Coming Soon]**: Release Full HY-World 2.0 (World Generation) inference code.\n- **[Coming Soon]**: Release ![Panorama Generation](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPanorama_Generation-4285F4?style=flat-square) (HY-Pano 2.0) model weights & code.\n- **[Coming Soon]**: Release ![Trajectory Planning](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FTrajectory_Planning-EA4335?style=flat-square)（WorldNav） code.\n- **[Coming Soon]**: Release ![World Expansion](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FWorld_Expansion-FBBC05?style=flat-square)(WorldStereo 2.0) model weights & inference code.\n\n\n## 📋 Table of Contents\n- [📖 Introduction](#-introduction)\n- [✨ Highlights](#-highlights)\n- [🧩 Architecture](#-architecture)\n- [📝 Open-Source Plan](#-open-source-plan)\n- [🎁 Model Zoo](#-model-zoo)\n- [🤗 Get Started](#-get-started)\n- [🔮 Performance](#-performance)\n- [🎬 More Examples](#-more-examples)\n- [📚 Citation](#-citation)\n\n\n## 📖 Introduction\n\n**HY-World 2.0** is a multi-modal world model framework for **world generation** and **world reconstruction**. It accepts diverse input modalities — text, single-view images, multi-view images, and videos — and produces 3D world representations (meshes \u002F Gaussian Splattings). It offers two core capabilities:\n\n- **World Generation** (text \u002F single image → 3D world): syntheses high-fidelity, navigable 3D scenes through a four-stage method —— a) ![Panorama Generation](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPanorama_Generation-4285F4?style=flat-square) with HY-Pano 2.0, b) ![Trajectory Planning](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FTrajectory_Planning-EA4335?style=flat-square) with WorldNav, c) ![World Expansion](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FWorld_Expansion-FBBC05?style=flat-square) with WorldStereo 2.0, and d) ![World Composition](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FWorld_Composition-34A853?style=flat-square) with WorldMirror 2.0 & 3DGS learning.\n- **World Reconstruction** (multi-view images \u002F video → 3D): Powered by WorldMirror 2.0, a unified feed-forward model that simultaneously predicts depth, surface normals, camera parameters, 3D point clouds, and 3DGS attributes in a single forward pass.\n\nHY-World 2.0 is an **open-source state-of-the-art** world model.  We will release all model weights, code, and technical details to facilitate reproducibility and advance research in this field.\n\n### Why 3D World Models?\n\nExisting world models, such as Genie 3, Cosmos, and HY-World 1.5 (WorldPlay+WorldCompass), generate pixel-level videos — essentially \"watching a movie\" that vanishes once playback ends. **HY-World 2.0 takes a fundamentally different approach**: it directly produces editable, persistent 3D assets (meshes \u002F 3DGS) that can be imported into game engines like Blender\u002FUnity\u002FUnreal Engine\u002FIsaac Sim — more like \"building a playable game\" than recording a clip. This paradigm shift natively resolves many long-standing pain points of video world models:\n\n|  | Video World Models | 3D World Model (HY-World 2.0) |\n|--|---|---|\n| **Output** | Pixel videos (non-editable) | Real 3D assets — meshes \u002F 3DGS (fully editable) |\n| **Playable Duration** | Limited (typically 1 min) | Unlimited — assets persist permanently |\n| **3D Consistency** | No (flickering, artifacts across views) | Native — inherently consistent in 3D |\n| **Real-Time Rendering** | Requires per-frame inference; high latency | Consumer GPUs can render in real time |\n| **Controllability** | Weak (imprecise character control, no real physics) | Precise — zero-error control, real physics collision, accurate lighting |\n| **Inference Cost** | Accumulates with every interaction | One-time generation; rendering cost ≈ 0 |\n| **Engine Compatibility** | ✗ Video files only | ✓ Directly importable into Blender \u002F UE \u002F Isaac Engine |\n| | $\\color{IndianRed}{\\textsf{Watch a video, then it's gone}}$ | $\\color{RoyalBlue}{\\textbf{Build a world, keep it forever}}$ |\n\n\n\u003Ctable align=\"center\" style=\"border: none;\">\n  \u003Ctr>\n    \u003Ctd align=\"center\" width=\"50%\">\u003Cimg src=\"assets\u002Fscreenshot_1.gif\" width=\"100%\">\u003C\u002Ftd>\n    \u003Ctd align=\"center\" width=\"50%\">\u003Cimg src=\"assets\u002Fscreenshot_2.gif\" width=\"100%\">\u003C\u002Ftd>\n  \u003C\u002Ftr>\n  \u003Ctr>\n    \u003Ctd align=\"center\" width=\"50%\">\u003Cimg src=\"assets\u002Fscreenshot_7.gif\" width=\"100%\">\u003C\u002Ftd>\n    \u003Ctd align=\"center\" width=\"50%\">\u003Cimg src=\"assets\u002Fscreenshot_8.gif\" width=\"100%\">\u003C\u002Ftd>\n  \u003C\u002Ftr>\n\u003C\u002Ftable>\n\n\u003Cp align=\"center\">\u003Cem>All above are \u003Cstrong>real 3D assets\u003C\u002Fstrong> (not generated videos) and entirely created by HY-World 2.0 -- captured from live real-time interaction.\u003C\u002Fem>\u003C\u002Fp>\n\n## ✨ Highlights\n\n- **Real 3D Worlds, Not Just Videos**\n\n  Unlike video-only world models (e.g., Genie 3, HY World 1.5), HY-World 2.0 generates **real 3D assets** — 3DGS, meshes, and point clouds — that are freely explorable, editable, and directly importable into **Unity \u002F Unreal Engine \u002F Isaac**. From a single text prompt or image, create navigable 3D worlds with diverse styles: realistic, cartoon, game, and more.\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Fmesh_en.gif\" width=\"95%\">\n\u003C\u002Fp>\n\n\n- **Instant 3D Reconstruction from Photos & Videos**\n\n  Powered by **WorldMirror 2.0**, a unified feed-forward model that predicts dense point clouds, depth maps, surface normals, camera parameters, and 3DGS from multi-view images or casual videos in a single forward pass. Supports flexible-resolution inference (50K–500K pixels) with SOTA accuracy. Capture a video, get a digital twin.\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Frecon_en.gif\" width=\"95%\">\n\u003C\u002Fp>\n\n- **Interactive Character Exploration**\n\n  Go beyond viewing — **play inside your generated worlds**. HY-World 2.0 supports first-person navigation and third-person character mode, enabling users to freely explore AI-generated streets, buildings, and landscapes with physics-based collision.  Go to [our product page](https:\u002F\u002F3d.hunyuan.tencent.com\u002FsceneTo3D) for free try (![Very Crowded Now](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FVery_Crowded_Now,_Be_Patient-EA4335?style=flat-square)). \n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Finteractive.gif\" width=\"95%\">\n\u003C\u002Fp>\n\n\n## 🧩 Architecture\n- **Refer to our tech report for more details**\n\n  A systematic pipeline of HY-World 2.0 — *Panorama Generation* (HY-Pano-2.0) → *Trajectory Planning* (WorldNav) → *World Expansion* (WorldStereo 2.0) → *World Composition* (WorldMirror 2.0 + Splattings Learning) — that automatically transforms text or a single image into a high-fidelity, navigable 3D world (3DGS\u002Fmesh outputs).\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Foverview.png\" width=\"95%\">\n\u003C\u002Fp>\n\n## 📝 Open-Source Plan\n\n- ✅ Technical Report\n- ✅ WorldMirror 2.0 Code & Model Checkpoints\n- ⬜ Full Inference Code for World Generation (WorldNav + World Composition)\n- ⬜ Panorama Generation (HY-Pano 2.0) Model & Code — [HunyuanWorld 1.0](https:\u002F\u002Fgithub.com\u002FTencent-Hunyuan\u002FHunyuanWorld-1.0) available as interim alternative\n- ⬜ World Expansion (WorldStereo 2.0) Model & Code — [WorldStereo](https:\u002F\u002Fgithub.com\u002FFuchengSu\u002FWorldStereo) available as interim alternative\n\n\n## 🎁 Model Zoo\n\n### World Reconstruction — WorldMirror Series\n\n| Model | Description | Params | Date | Hugging Face |\n|-------|-------------|--------|------|--------------|\n| WorldMirror-2 [new] | Multi-view \u002F video → 3D reconstruction | ~1.2B | 2026 | [Download](https:\u002F\u002Fhuggingface.co\u002Ftencent\u002FHY-World-2.0\u002Ftree\u002Fmain\u002FHY-WorldMirror-2.0) |\n| WorldMirror-1 | Multi-view \u002F video → 3D reconstruction (legacy) | ~1.2B | 2025 | [Download](https:\u002F\u002Fhuggingface.co\u002Ftencent\u002FHunyuanWorld-Mirror\u002Ftree\u002Fmain) |\n\n### Panorama Generation — HY-Pano Series\n\n| Model | Description | Params | Date | Hugging Face |\n|-------|-------------|--------|------|--------------|\n| HY-Pano-2 [new] | Text \u002F image → 360° panorama | — | Coming Soon | — |\n\n### World Expansion — WorldStereo Series\n\n| Model           | Description | Params | Date | Hugging Face |\n|-----------------|-------------|-----|------|--------------|\n| WorldStereo-2 [new] | Panorama →  3DGS world |  —  | Coming Soon | — |\n\n### Spatial Planning — WorldNav Series\n| Algorithm           | Description | Params | Date |\n|-----------------|-------------|-----|------|\n| WorldNav [new] | Panorama →  Camera Traj. |  —  | Coming Soon | \n\nWe recommend referring to our previous works, [WorldStereo](https:\u002F\u002Fgithub.com\u002FFuchengSu\u002FWorldStereo) and [WorldMirror](https:\u002F\u002Fgithub.com\u002FTencent-Hunyuan\u002FHunyuanWorld-Mirror), for background knowledge on 3D world generation and reconstruction. \n\n## 🤗 Get Started\n\n### Install Requirements\n\nWe recommend CUDA 12.4 for installation.\n\n```bash\n# 1. Clone the repository\ngit clone https:\u002F\u002Fgithub.com\u002FTencent-Hunyuan\u002FHY-World-2.0\ncd HY-World-2.0\n\n# 2. Create conda environment\nconda create -n hyworld2 python=3.10\nconda activate hyworld2\n\n# 3. Install PyTorch (CUDA 12.4)\npip install torch==2.4.0 torchvision==0.19.0 --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu124\n\n# 4. Install dependencies\npip install -r requirements.txt\n\n# 5. Install FlashAttention\n# (Recommended) Install FlashAttention-3\ngit clone https:\u002F\u002Fgithub.com\u002FDao-AILab\u002Fflash-attention.git\ncd flash-attention\u002Fhopper\npython setup.py install\ncd ..\u002F..\u002F\nrm -rf flash-attention\n\n# For simpler installation, you can also use FlashAttention-2\npip install flash-attn --no-build-isolation\n```\n\n### Code Usage — Panorama Generation (HY-Pano-2)\n\n*Coming soon.*\n\n### Code Usage — World Generation (WorldNav, WorldStereo-2, and 3DGS)\n\n*Coming soon.*\n\n**We recommend referring to our previous work, [WorldStereo](https:\u002F\u002Fgithub.com\u002FFuchengSu\u002FWorldStereo), for the open-source preview version of WorldStereo-2.**\n\n### Code Usage — WorldMirror 2.0\nWorldMirror 2.0 supports the following usage modes:\n\n- [Code Usage](#code-usage--worldmirror-20)\n- [Gradio App](#gradio-app--worldmirror-20)\n\nWe provide a `diffusers`-like Python API for WorldMirror 2.0. Model weights are automatically downloaded from Hugging Face on first run.\n\n```python\nfrom hyworld2.worldrecon.pipeline import WorldMirrorPipeline\n\npipeline = WorldMirrorPipeline.from_pretrained('tencent\u002FHY-World-2.0')\nresult = pipeline('path\u002Fto\u002Fimages')\n```\n\n**With Prior Injection (Camera & Depth):**\n\n```python\nresult = pipeline(\n    'path\u002Fto\u002Fimages',\n    prior_cam_path='path\u002Fto\u002Fprior_camera.json',\n    prior_depth_path='path\u002Fto\u002Fprior_depth\u002F',\n)\n```\n\n> For the detailed structure of camera\u002Fdepth priors and how to prepare them, see [Prior Preparation Guide](DOCUMENTATION.md#prior-injection).\n\n**CLI:**\n\n```bash\n# Single GPU\npython -m hyworld2.worldrecon.pipeline --input_path path\u002Fto\u002Fimages\n\n# Multi-GPU\ntorchrun --nproc_per_node=2 -m hyworld2.worldrecon.pipeline \\\n    --input_path path\u002Fto\u002Fimages \\\n    --use_fsdp --enable_bf16\n```\n\n> **Important:** In multi-GPU mode, the number of input images must be **>= the number of GPUs**. For example, with `--nproc_per_node=8`, provide at least 8 images.\n\n### Gradio App — WorldMirror 2.0\n\nWe provide an interactive [Gradio](https:\u002F\u002Fwww.gradio.app\u002F) web demo for WorldMirror 2.0. Upload images or videos and visualize 3DGS, point clouds, depth maps, normal maps, and camera parameters in your browser.\n\n```bash\n# Single GPU\npython -m hyworld2.worldrecon.gradio_app\n\n# Multi-GPU\ntorchrun --nproc_per_node=2 -m hyworld2.worldrecon.gradio_app \\\n    --use_fsdp --enable_bf16\n```\n\nFor the full list of Gradio app arguments (port, share, local checkpoints, etc.), see [DOCUMENTATION.md](DOCUMENTATION.md#gradio-app).\n\n\n\n## 🔮 Performance\n\nFor full benchmark results, please refer to the [technical report](https:\u002F\u002F3d-models.hunyuan.tencent.com\u002Fworld\u002F).\n\n### WorldStereo 2.0 — Camera Control\n\n\u003Ctable>\n  \u003Cthead>\n    \u003Ctr>\n      \u003Cth rowspan=\"2\">Methods\u003C\u002Fth>\n      \u003Cth colspan=\"3\" align=\"center\">Camera Metrics\u003C\u002Fth>\n      \u003Cth colspan=\"4\" align=\"center\">Visual Quality\u003C\u002Fth>\n    \u003C\u002Ftr>\n    \u003Ctr>\n      \u003Cth>RotErr ↓\u003C\u002Fth>\u003Cth>TransErr ↓\u003C\u002Fth>\u003Cth>ATE ↓\u003C\u002Fth>\n      \u003Cth>Q-Align ↑\u003C\u002Fth>\u003Cth>CLIP-IQA+ ↑\u003C\u002Fth>\u003Cth>Laion-Aes ↑\u003C\u002Fth>\u003Cth>CLIP-I ↑\u003C\u002Fth>\n    \u003C\u002Ftr>\n  \u003C\u002Fthead>\n  \u003Ctbody>\n    \u003Ctr>\u003Ctd>SEVA\u003C\u002Ftd>\u003Ctd>1.690\u003C\u002Ftd>\u003Ctd>1.578\u003C\u002Ftd>\u003Ctd>2.879\u003C\u002Ftd>\u003Ctd>3.232\u003C\u002Ftd>\u003Ctd>0.479\u003C\u002Ftd>\u003Ctd>4.623\u003C\u002Ftd>\u003Ctd>77.16\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>Gen3C\u003C\u002Ftd>\u003Ctd>0.944\u003C\u002Ftd>\u003Ctd>1.580\u003C\u002Ftd>\u003Ctd>2.789\u003C\u002Ftd>\u003Ctd>3.353\u003C\u002Ftd>\u003Ctd>0.489\u003C\u002Ftd>\u003Ctd>4.863\u003C\u002Ftd>\u003Ctd>82.33\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>WorldStereo\u003C\u002Ftd>\u003Ctd>0.762\u003C\u002Ftd>\u003Ctd>1.245\u003C\u002Ftd>\u003Ctd>2.141\u003C\u002Ftd>\u003Ctd>4.149\u003C\u002Ftd>\u003Ctd>\u003Cb>0.547\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>5.257\u003C\u002Ftd>\u003Ctd>89.05\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>\u003Cb>WorldStereo 2.0\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>\u003Cb>0.492\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>\u003Cb>0.968\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>\u003Cb>1.768\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>\u003Cb>4.205\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>0.544\u003C\u002Ftd>\u003Ctd>\u003Cb>5.266\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>\u003Cb>89.43\u003C\u002Fb>\u003C\u002Ftd>\u003C\u002Ftr>\n  \u003C\u002Ftbody>\n\u003C\u002Ftable>\n\n### WorldStereo 2.0 — Single-View-Generated Reconstruction\n\n\u003Ctable>\n  \u003Cthead>\n    \u003Ctr>\n      \u003Cth rowspan=\"2\">Methods\u003C\u002Fth>\n      \u003Cth colspan=\"4\">Tanks-and-Temples\u003C\u002Fth>\n      \u003Cth colspan=\"4\">MipNeRF360\u003C\u002Fth>\n    \u003C\u002Ftr>\n    \u003Ctr>\n      \u003Cth>Precision ↑\u003C\u002Fth>\n      \u003Cth>Recall ↑\u003C\u002Fth>\n      \u003Cth>F1-Score ↑\u003C\u002Fth>\n      \u003Cth>AUC ↑\u003C\u002Fth>\n      \u003Cth>Precision ↑\u003C\u002Fth>\n      \u003Cth>Recall ↑\u003C\u002Fth>\n      \u003Cth>F1-Score ↑\u003C\u002Fth>\n      \u003Cth>AUC ↑\u003C\u002Fth>\n    \u003C\u002Ftr>\n  \u003C\u002Fthead>\n  \u003Ctbody align=\"center\">\n    \u003Ctr>\n      \u003Ctd align=\"left\">SEVA\u003C\u002Ftd>\n      \u003Ctd>33.59\u003C\u002Ftd>\n      \u003Ctd>35.34\u003C\u002Ftd>\n      \u003Ctd>36.73\u003C\u002Ftd>\n      \u003Ctd>51.03\u003C\u002Ftd>\n      \u003Ctd>22.38\u003C\u002Ftd>\n      \u003Ctd>55.63\u003C\u002Ftd>\n      \u003Ctd>28.75\u003C\u002Ftd>\n      \u003Ctd>46.81\u003C\u002Ftd>\n    \u003C\u002Ftr>\n    \u003Ctr>\n      \u003Ctd align=\"left\">Gen3C\u003C\u002Ftd>\n      \u003Ctd>\u003Cu>46.73\u003C\u002Fu>\u003C\u002Ftd>\n      \u003Ctd>25.51\u003C\u002Ftd>\n      \u003Ctd>31.24\u003C\u002Ftd>\n      \u003Ctd>42.44\u003C\u002Ftd>\n      \u003Ctd>23.28\u003C\u002Ftd>\n      \u003Ctd>\u003Cstrong>75.37\u003C\u002Fstrong>\u003C\u002Ftd>\n      \u003Ctd>35.26\u003C\u002Ftd>\n      \u003Ctd>52.10\u003C\u002Ftd>\n    \u003C\u002Ftr>\n    \u003Ctr>\n      \u003Ctd align=\"left\">Lyra\u003C\u002Ftd>\n      \u003Ctd>\u003Cstrong>50.38\u003C\u002Fstrong>\u003C\u002Ftd>\n      \u003Ctd>28.67\u003C\u002Ftd>\n      \u003Ctd>32.54\u003C\u002Ftd>\n      \u003Ctd>43.05\u003C\u002Ftd>\n      \u003Ctd>30.02\u003C\u002Ftd>\n      \u003Ctd>58.60\u003C\u002Ftd>\n      \u003Ctd>36.05\u003C\u002Ftd>\n      \u003Ctd>49.89\u003C\u002Ftd>\n    \u003C\u002Ftr>\n    \u003Ctr>\n      \u003Ctd align=\"left\">FlashWorld\u003C\u002Ftd>\n      \u003Ctd>26.58\u003C\u002Ftd>\n      \u003Ctd>20.72\u003C\u002Ftd>\n      \u003Ctd>22.29\u003C\u002Ftd>\n      \u003Ctd>30.45\u003C\u002Ftd>\n      \u003Ctd>35.97\u003C\u002Ftd>\n      \u003Ctd>53.77\u003C\u002Ftd>\n      \u003Ctd>42.60\u003C\u002Ftd>\n      \u003Ctd>53.86\u003C\u002Ftd>\n    \u003C\u002Ftr>\n    \u003Ctr>\n      \u003Ctd align=\"left\">WorldStereo 2.0\u003C\u002Ftd>\n      \u003Ctd>43.62\u003C\u002Ftd>\n      \u003Ctd>\u003Cu>41.02\u003C\u002Fu>\u003C\u002Ftd>\n      \u003Ctd>\u003Cu>41.43\u003C\u002Fu>\u003C\u002Ftd>\n      \u003Ctd>\u003Cu>58.19\u003C\u002Fu>\u003C\u002Ftd>\n      \u003Ctd>\u003Cstrong>43.19\u003C\u002Fstrong>\u003C\u002Ftd>\n      \u003Ctd>\u003Cu>65.32\u003C\u002Fu>\u003C\u002Ftd>\n      \u003Ctd>\u003Cstrong>51.27\u003C\u002Fstrong>\u003C\u002Ftd>\n      \u003Ctd>\u003Cstrong>65.79\u003C\u002Fstrong>\u003C\u002Ftd>\n    \u003C\u002Ftr>\n    \u003Ctr>\n      \u003Ctd align=\"left\">WorldStereo 2.0 (DMD)\u003C\u002Ftd>\n      \u003Ctd>40.41\u003C\u002Ftd>\n      \u003Ctd>\u003Cstrong>44.41\u003C\u002Fstrong>\u003C\u002Ftd>\n      \u003Ctd>\u003Cstrong>43.16\u003C\u002Fstrong>\u003C\u002Ftd>\n      \u003Ctd>\u003Cstrong>60.09\u003C\u002Fstrong>\u003C\u002Ftd>\n      \u003Ctd>\u003Cu>42.34\u003C\u002Fu>\u003C\u002Ftd>\n      \u003Ctd>64.83\u003C\u002Ftd>\n      \u003Ctd>\u003Cu>50.52\u003C\u002Fu>\u003C\u002Ftd>\n      \u003Ctd>\u003Cu>65.64\u003C\u002Fu>\u003C\u002Ftd>\n    \u003C\u002Ftr>\n  \u003C\u002Ftbody>\n\u003C\u002Ftable>\n\n### WorldMirror 2.0 — Point Map Reconstruction\n\n**Point Map Reconstruction on 7-Scenes, NRGBD, and DTU.** We report the mean Accuracy and Completeness of WorldMirror under different input configurations. **Bold** results are best. \"L \u002F M \u002F H\" denote low \u002F medium \u002F high inference resolution. \"+ all priors\" denotes injection of camera extrinsics, camera intrinsics, and depth priors.\n\n\u003Ctable>\n  \u003Cthead>\n    \u003Ctr>\n      \u003Cth rowspan=\"2\">Method\u003C\u002Fth>\n      \u003Cth colspan=\"2\" align=\"center\">7-Scenes \u003Csub>(scene)\u003C\u002Fsub>\u003C\u002Fth>\n      \u003Cth colspan=\"2\" align=\"center\">NRGBD \u003Csub>(scene)\u003C\u002Fsub>\u003C\u002Fth>\n      \u003Cth colspan=\"2\" align=\"center\">DTU \u003Csub>(object)\u003C\u002Fsub>\u003C\u002Fth>\n    \u003C\u002Ftr>\n    \u003Ctr>\n      \u003Cth>Acc. ↓\u003C\u002Fth>\u003Cth>Comp. ↓\u003C\u002Fth>\n      \u003Cth>Acc. ↓\u003C\u002Fth>\u003Cth>Comp. ↓\u003C\u002Fth>\n      \u003Cth>Acc. ↓\u003C\u002Fth>\u003Cth>Comp. ↓\u003C\u002Fth>\n    \u003C\u002Ftr>\n  \u003C\u002Fthead>\n  \u003Ctbody>\n    \u003Ctr>\u003Ctd colspan=\"7\">\u003Cem>WorldMirror 1.0\u003C\u002Fem>\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>  L\u003C\u002Ftd>\u003Ctd>0.043\u003C\u002Ftd>\u003Ctd>0.055\u003C\u002Ftd>\u003Ctd>0.046\u003C\u002Ftd>\u003Ctd>0.049\u003C\u002Ftd>\u003Ctd>1.476\u003C\u002Ftd>\u003Ctd>1.768\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>  L + all priors\u003C\u002Ftd>\u003Ctd>0.021\u003C\u002Ftd>\u003Ctd>0.026\u003C\u002Ftd>\u003Ctd>0.022\u003C\u002Ftd>\u003Ctd>0.020\u003C\u002Ftd>\u003Ctd>1.347\u003C\u002Ftd>\u003Ctd>1.392\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>  M\u003C\u002Ftd>\u003Ctd>0.043\u003C\u002Ftd>\u003Ctd>0.049\u003C\u002Ftd>\u003Ctd>0.041\u003C\u002Ftd>\u003Ctd>0.045\u003C\u002Ftd>\u003Ctd>1.017\u003C\u002Ftd>\u003Ctd>1.780\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>  M + all priors\u003C\u002Ftd>\u003Ctd>0.018\u003C\u002Ftd>\u003Ctd>0.023\u003C\u002Ftd>\u003Ctd>0.016\u003C\u002Ftd>\u003Ctd>0.014\u003C\u002Ftd>\u003Ctd>0.735\u003C\u002Ftd>\u003Ctd>0.935\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>  H\u003C\u002Ftd>\u003Ctd>0.079\u003C\u002Ftd>\u003Ctd>0.087\u003C\u002Ftd>\u003Ctd>0.077\u003C\u002Ftd>\u003Ctd>0.093\u003C\u002Ftd>\u003Ctd>2.271\u003C\u002Ftd>\u003Ctd>2.113\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>  H + all priors\u003C\u002Ftd>\u003Ctd>0.042\u003C\u002Ftd>\u003Ctd>0.041\u003C\u002Ftd>\u003Ctd>0.078\u003C\u002Ftd>\u003Ctd>0.082\u003C\u002Ftd>\u003Ctd>1.773\u003C\u002Ftd>\u003Ctd>1.478\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd colspan=\"7\">\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd colspan=\"7\">\u003Cem>WorldMirror 2.0\u003C\u002Fem>\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>  L\u003C\u002Ftd>\u003Ctd>0.041\u003C\u002Ftd>\u003Ctd>0.052\u003C\u002Ftd>\u003Ctd>0.047\u003C\u002Ftd>\u003Ctd>0.058\u003C\u002Ftd>\u003Ctd>1.352\u003C\u002Ftd>\u003Ctd>2.009\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>  L + all priors\u003C\u002Ftd>\u003Ctd>0.019\u003C\u002Ftd>\u003Ctd>0.024\u003C\u002Ftd>\u003Ctd>0.017\u003C\u002Ftd>\u003Ctd>0.015\u003C\u002Ftd>\u003Ctd>1.100\u003C\u002Ftd>\u003Ctd>1.201\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>  M\u003C\u002Ftd>\u003Ctd>0.033\u003C\u002Ftd>\u003Ctd>0.046\u003C\u002Ftd>\u003Ctd>0.039\u003C\u002Ftd>\u003Ctd>0.047\u003C\u002Ftd>\u003Ctd>1.005\u003C\u002Ftd>\u003Ctd>1.892\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>  M + all priors\u003C\u002Ftd>\u003Ctd>0.013\u003C\u002Ftd>\u003Ctd>0.017\u003C\u002Ftd>\u003Ctd>\u003Cb>0.013\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>\u003Cb>0.013\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>0.690\u003C\u002Ftd>\u003Ctd>0.876\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>  H\u003C\u002Ftd>\u003Ctd>0.037\u003C\u002Ftd>\u003Ctd>0.040\u003C\u002Ftd>\u003Ctd>0.046\u003C\u002Ftd>\u003Ctd>0.053\u003C\u002Ftd>\u003Ctd>0.845\u003C\u002Ftd>\u003Ctd>1.904\u003C\u002Ftd>\u003C\u002Ftr>\n    \u003Ctr>\u003Ctd>  \u003Cb>H + all priors\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>\u003Cb>0.012\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>\u003Cb>0.016\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>0.015\u003C\u002Ftd>\u003Ctd>0.016\u003C\u002Ftd>\u003Ctd>\u003Cb>0.554\u003C\u002Fb>\u003C\u002Ftd>\u003Ctd>\u003Cb>0.771\u003C\u002Fb>\u003C\u002Ftd>\u003C\u002Ftr>\n  \u003C\u002Ftbody>\n\u003C\u002Ftable>\n \n### WorldMirror 2.0 — Prior Comparison\n\n**Comparison with Pow3R and MapAnything under Different Prior Conditions.** Results are averaged on 7-Scenes, NRGBD, and DTU datasets. Pow3R (pro) refers to the original Pow3R with Procrustes alignment.\n\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Fprior_comparison2_wm2.png\" width=\"85%\">\n\u003C\u002Fp>\n\n\n\n\n## 🎬 More Examples\n\n\u003Ctable align=\"center\" style=\"border: none;\">\n  \u003Ctr>\n    \u003Ctd align=\"center\" width=\"50%\">\u003Cimg src=\"assets\u002Fscreenshot_3.gif\" width=\"100%\">\u003C\u002Ftd>\n    \u003Ctd align=\"center\" width=\"50%\">\u003Cimg src=\"assets\u002Fscreenshot_4.gif\" width=\"100%\">\u003C\u002Ftd>\n  \u003C\u002Ftr>\n  \u003Ctr>\n    \u003Ctd align=\"center\" width=\"50%\">\u003Cimg src=\"assets\u002Fscreenshot_5.gif\" width=\"100%\">\u003C\u002Ftd>\n    \u003Ctd align=\"center\" width=\"50%\">\u003Cimg src=\"assets\u002Fscreenshot_6.gif\" width=\"100%\">\u003C\u002Ftd>\n  \u003C\u002Ftr>\n  \u003Ctr>\n    \u003Ctd align=\"center\" width=\"50%\">\u003Cimg src=\"assets\u002Fscreenshot_9.gif\" width=\"100%\">\u003C\u002Ftd>\n    \u003Ctd align=\"center\" width=\"50%\">\u003Cimg src=\"assets\u002Fscreenshot_10.gif\" width=\"100%\">\u003C\u002Ftd>\n  \u003C\u002Ftr>\n\u003C\u002Ftable>\n\n\n## 📖 Documentation\n\nFor detailed usage guides, parameter references, output format specifications, and prior injection instructions, see **[DOCUMENTATION.md](DOCUMENTATION.md)**.\n\n\n## 📚 Citation\n\nIf you find HunyuanWorld 2.0 useful for your research, please cite:\n\n```bibtex\n@article{hyworld22026,\n  title={HY-World 2.0: A Multi-Modal World Model for Reconstructing, Generating, and Simulating 3D Worlds},\n  author={Team HY-World},\n  journal={arXiv preprint},\n  year={2026}\n}\n\n@article{hunyuanworld2025tencent,\n    title={HunyuanWorld 1.0: Generating Immersive, Explorable, and Interactive 3D Worlds from Words or Pixels},\n    author={Team HunyuanWorld},\n    year={2025},\n    journal={arXiv preprint}\n}\n```\n\n## 📧 Contact\n\nPlease send emails to tengfeiwang12@gmail.com for questions or feedback.\n\n\n## 🙏 Acknowledgements\n\nWe would like to thank [HunyuanWorld 1.0](https:\u002F\u002Fgithub.com\u002FTencent-Hunyuan\u002FHunyuanWorld-1.0), [WorldMirror](https:\u002F\u002Fgithub.com\u002FTencent-Hunyuan\u002FHunyuanWorld-Mirror), [WorldPlay](https:\u002F\u002Fgithub.com\u002FTencent-Hunyuan\u002FHY-WorldPlay), [WorldStereo](https:\u002F\u002Fgithub.com\u002FFuchengSu\u002FWorldStereo), [HunyuanImage](https:\u002F\u002Fgithub.com\u002FTencent-Hunyuan\u002FHunyuanImage-3.0) for their great work.\n","HY-World 2.0 是一个多模态世界模型框架，用于3D世界的生成、重建和模拟。其核心功能包括通过文本、单视图图像、多视图图像和视频等多种输入模式来创建或复现3D环境。该项目采用Python语言开发，利用先进的AI技术实现对复杂场景的高度还原与创新构建。适用于需要高质量3D内容生成的领域，如虚拟现实、游戏开发、建筑设计等。此外，HY-World 2.0还计划逐步开源更多组件，包括全景生成、轨迹规划及世界扩展等功能模块，为用户提供更全面的世界建模解决方案。",2,"2026-06-11 02:38:39","CREATED_QUERY"]