Hunyuan 3D World Model 2.0 - Tencent Hunyuan Open Source Multimodal World Model

What is Hunyuan 3D World Model 2.0?

HY-World 2.0 is an open-source, multimodal world model developed by Tencent's Hunyuan platform. It supports the generation of roamable 3D Gaussian sputtering (3DGS) scenes from text, single images, multiple views, or videos. The model employs a four-stage architecture: panoramic generation → trajectory planning → world expansion → 3D reconstruction, achieving a unified approach to "generating a 3D world from text/images" and "reconstructing a 3D world from video." HY-World 2.0's generated scenes support physical collision and character exploration, and can be exported to Unity/UE engines, with performance comparable to the closed-source commercial product Marble.

Main functions of Hunyuan 3D World Model 2.0

• World GenerationGenerate 360° roamable 3DGS/Mesh immersive scenes based on text or a single image.
• World ReconstructionReconstruct a high-fidelity 3D digital twin space from multi-view image or video input.
• Panoramic image generationConvert images or text from any perspective into 360° panoramic images (HY-Pano 2.0).
• Trajectory Intelligent Planning: Analyze scene semantics and plan exploration paths to avoid unreasonable behaviors such as passing through walls (WorldNav).
• Role-playing Adventure ModeSupports controlling characters to freely walk and explore in generated scenes.
• Export in multiple formatsSupports integration with various formats such as 3DGS, Mesh, point cloud, and video, as well as mainstream engines.

Technical Principles of Hunyuan 3D World Model 2.0

• Panoramic image generation (HY-Pano 2.0)Employing an end-to-end implicit learning approach, the system autonomously learns the spatial mapping from perspective view to 360° isometric cylindrical projection (ERP) using a multimodal Diffusion Transformer (MMDiT) without requiring camera metadata. Circle padding and pixel blending techniques are introduced to address the discontinuity issue at ERP boundaries, and training is performed using a hybrid approach combining real panoramic and user-generated (UE) synthetic data.
• Trajectory Planning (WorldNav)It understands spatial structure through geometric and semantic scene analysis (point cloud, Mesh, semantic segmentation, NavMesh), intelligently plans roaming trajectories that maximize information coverage, ensures natural paths and avoids obstacles, and supports diverse trajectory types such as regular, surround, reconstruction perception, roaming, and aerial.
• WorldStereo 2.0Based on keyframe generation rather than continuous video generation, global geometric memory and spatial stereo memory mechanisms are introduced to ensure consistency across multiple perspectives. Efficiency is improved through model distillation, enabling scene expansion and detail completion along the planned path.
• World Mirror 2.0A feedforward 3D prediction model is employed, with geometric accuracy optimized through normalized position encoding, explicit normal supervision, and depth mask prediction. 3D Gaussian sputtering (3DGS) is combined for final scene fusion and optimization, supporting robust reconstruction from both generated views and real inputs.

How to use the Hunyuan 3D World Model 2.0

• Access PlatformAccess the Tencent Hunyuan 3D official website and log in with your account.
• Select input methodUpload a single image (any angle is supported) or enter a text description (e.g., "Cyberpunk-style street with flashing neon lights").
• Parameter configurationSelect parameters such as generation mode (world generation/world reconstruction), scene style, and roaming range.
• Generation and ExplorationAfter clicking "Generate", the system will automatically execute a four-stage process (panorama → trajectory → expansion → reconstruction).
• ExportSupports downloading 3DGS/Mesh files or directly generating asset packages that can be imported into the Unity/UE engine.

Key information and usage requirements of Hunyuan 3D World Model 2.0

• Development TeamTencent Hunyuan was developed and launched.
• Open source situationThe model weights, training code, and complete technical report have been fully open-sourced and are available for free download and study.
• Input modeIt supports four input methods: text prompts, single images, multi-view images, and video streams.
• Output formatIt can export 3D Gaussian sputtering (3DGS), mesh models, point cloud data, 360° panoramic images and video sequences.
• Engine supportThe generated assets are natively compatible with Unity and Unreal Engine (UE) and can be directly imported for secondary editing.
• Hardware RequirementsLocal deployment requires an NVIDIA GPU, and it is recommended to use an environment with ≥16GB of video memory to ensure generation efficiency.

The core advantages of Hunyuan 3D World Model 2.0

• Unified Generation and ReconstructionThe first open-source framework to support both "creating something from nothing" and "reconstructing from reality" paradigms.
• Physically InteractiveThe generated scene has spatial logic and physical collision detection, and supports character-controlled exploration.
• No camera parameters requiredThe panoramic generation stage does not require metadata such as focal length and FOV, adapting to real-world scene images.
• High-quality boundary handling360° panoramic seamless integration, eliminating the left and right boundary breakage problem of traditional ERP images.
• Memory consistencyThe world expansion phase introduces a memory mechanism to ensure visual continuity during long-trajectory roaming.
• Industrial grade compatibleIt natively supports mainstream game engines and can be directly used for game development and robot simulation.

Project address for Hunyuan 3D World Model 2.0

• Project official website: https://3d-models.hunyuan.tencent.com/world/
• GitHub repository: https://github.com/Tencent-Hunyuan/HY-World-2.0
• HuggingFace model libraryhttps://huggingface.co/tencent/HY-World-2.0
• Technical Papers: https://3d-models.hunyuan.tencent.com/world/world2_0/HY_World_2_0.pdf

Comparison of Hunyuan 3D World Model 2.0 with similar competing products

Application Scenarios of Hunyuan 3D World Model 2.0

• Game development fieldDevelopers can use the model to quickly generate interactive 3D scene prototypes with physical collisions, significantly reducing the time and technical barriers of traditional level design.
• Virtual reality applicationsThe system can construct high-fidelity immersive VR/AR spaces, allowing end users to walk freely and explore in real time within the generated environment.
• Digital Twin ConstructionUsers only need to provide a few photos or short videos to automatically reconstruct a high-precision 3D digital copy of the real physical world.
• Robot simulation trainingIt provides a physically consistent, interactive virtual training environment for embodied intelligence and autonomous driving systems, supporting algorithm security verification.
• Virtual film productionThe model can generate 360° panoramic virtual background assets for movie pre-visualization, virtual studio shooting, and post-production special effects compositing.