World-model researchers
Study a released pipeline spanning panorama generation, navigation planning, consistent view synthesis, and 3D reconstruction.
Independent tool overview
HY-World 2.0 at a glance
HY-World 2.0 is Tencent Hunyuan's source-available research system for reconstructing, generating, and simulating navigable 3D worlds from multimodal inputs.
Visit the official HY-World 2.0 site ↗
- Product type
- Self-hosted multimodal 3D world model
- Outputs
- Navigable 3D Gaussian Splatting scenes or meshes
- Recommended hardware
- At least 4 GPUs for world generation; tested on 8 H20 GPUs
- License
- Tencent Hunyuan community license with territory and scale limits
- Last reviewed
- August 28, 2026
Overview
What HY-World 2.0 is
HY-World 2.0 is a research-oriented world model from Tencent Hunyuan. Its released pipeline combines panorama generation, camera-trajectory planning, view expansion, and 3D reconstruction so teams can turn visual inputs into scenes that can be explored from new viewpoints.
This is not a one-click hosted design app. The official repository is a self-hosted toolkit with model weights, inference code, multiple environments, and substantial GPU requirements. It is best evaluated as infrastructure for 3D and world-model teams rather than a finished consumer product.
Tencent calls the project open source, but the supplied community license has meaningful use and geography restrictions. Teams should review that license before committing the system to a commercial workflow.
Use cases
Who HY-World 2.0 is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
3D and spatial-computing teams
Prototype navigable environments for games, simulation, virtual production, or VR when the team can operate its own GPU stack.
Technical creators
Convert a panorama into an explorable 3D Gaussian Splatting scene or mesh with more control than a hosted one-click generator.
Capabilities
Core HY-World 2.0 features
HY-Pano 2.0 panorama generation
Creates panoramic starting views using either HunyuanImage-3's reasoning workflow or a lighter Qwen-Image-Edit backend.
WorldNav trajectory planning
Plans camera movement through a scene before the system expands the world along that path.
WorldStereo 2.0 expansion
Generates additional keyframes while using memory-guided conditioning to improve consistency across viewpoints.
WorldMirror 2.0 composition
Combines observations into the broader reconstruction and simulation pipeline described in the technical report.
3DGS and mesh output
Prepares frames, depth, normals, masks, and camera data for 3D Gaussian Splatting training, with mesh export also supported.
Interactive browser viewer
Includes a local viewer built around Viser and Nerfview for exploring reconstructed scenes.
Process
How the HY-World 2.0 workflow works
Step 1
Prepare the environment
Install the repository's CUDA and Python dependencies, download the required weights, and configure the separate environments used by the pipeline.
Step 2
Create or supply a panorama
Start with an existing panorama or generate one with HY-Pano 2.0.
Step 3
Plan and render a route
Use WorldNav to create a camera trajectory, then render and caption the path with a separately hosted vision-language model.
Step 4
Expand the scene
Run WorldStereo 2.0 to synthesize consistent keyframes beyond the original view.
Step 5
Build and inspect the 3D world
Prepare geometry signals, train the Gaussian Splatting representation, and inspect the result in the browser viewer.
Cost
HY-World 2.0 pricing and free plan
Tencent does not sell HY-World 2.0 as a hosted subscription. The repository and released weights can be downloaded under the project's community license, while GPUs, storage, model serving, and engineering time are paid separately.
Code and released models
$0 license fee
Download the released repository and model weights subject to Tencent's community license.
- Self-hosted rather than a managed service
- License territory excludes the EU, United Kingdom, and South Korea
- Redistribution and attribution conditions apply
Infrastructure
Variable
Run the system on your own hardware or cloud account.
- Official world-generation instructions recommend at least 4 GPUs
- The reference setup was tested on 8 NVIDIA H20 GPUs
- A separate VLM server is required for parts of the pipeline
Large-scale use
Contact Tencent
Organizations above the license's stated scale threshold must request a separate license.
- The condition applies when the licensee had more than 1 million monthly active users in the preceding month at the release date
- Legal review is appropriate before commercial deployment
Pricing checked . Check current pricing at the source ↗
Assessment
HY-World 2.0 strengths and limitations
Where it stands out
- Connects several difficult world-generation stages in one documented research pipeline.
- Provides released inference code and weights for major components rather than only a demo video.
- Supports both 3D Gaussian Splatting and mesh-oriented outputs.
- Gives technical teams direct control over models, data, trajectories, and reconstruction settings.
What to consider
- The setup is complex and assumes strong Python, CUDA, model-serving, and 3D graphics experience.
- World generation requires substantial compute; the official documentation recommends at least four GPUs and reports testing on eight H20 GPUs.
- The community license excludes use in the EU, United Kingdom, and South Korea and imposes additional scale and model-training restrictions.
- There is no hosted plan with predictable per-seat pricing, managed uptime, or customer support.
- Generated geometry and visual consistency still require human inspection before production use.
Compare
HY-World 2.0 alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Content Creator
World Labs Marble
Consider Marble for a more productized World Labs workflow that creates persistent 3D worlds from text, images, or video.
Explore World Labs Marble →Content Creator
Odyssey-2
Consider Odyssey 2 when real-time interactive generated video is more important than a self-hosted 3D reconstruction pipeline.
Explore Odyssey-2 →Design
Blockade Labs Skybox AI
Consider Skybox AI for a simpler hosted workflow focused on generating 360-degree environments.
Explore Blockade Labs Skybox AI →Questions
HY-World 2.0 FAQs
What is HY-World 2.0?
HY-World 2.0 is Tencent Hunyuan's multimodal research system for reconstructing, generating, and simulating navigable 3D worlds. Its released components cover panorama generation, camera planning, view expansion, and 3D scene construction.
Is HY-World 2.0 free?
The released code and weights have no listed download fee, but they are governed by Tencent's community license. You must also provide and pay for the required GPU infrastructure, storage, supporting models, and engineering work.
Is HY-World 2.0 a hosted app?
No. The official release is a self-hosted GitHub project, not a managed web app or subscription service.
What hardware does HY-World 2.0 require?
Requirements vary by component. For the full world-generation workflow, the official instructions recommend at least four GPUs and say the pipeline was tested on eight NVIDIA H20 GPUs.
Can HY-World 2.0 generate a 3D world from one image?
The released world-generation workflow starts from a panorama, which can be supplied or generated by HY-Pano 2.0, then expands it along a planned trajectory and reconstructs a navigable 3D scene.
Can HY-World 2.0 be used commercially?
The community license permits some uses but includes territory exclusions, a large-scale user threshold, redistribution conditions, and restrictions on using the materials or outputs to improve other AI models. Obtain legal advice for your specific deployment.
Bottom line
Our HY-World 2.0 verdict
HY-World 2.0 is a substantial research release for teams that want control over an end-to-end 3D world pipeline and already have serious GPU and engineering capacity. Creators seeking quick results will be better served by a hosted alternative, while commercial teams should scrutinize the license before adopting it.
Visit HY-World 2.0 website ↗Similar Miscellaneous Tools
View all Miscellaneous AI tools →
Gemini 3.1 Flash TTS
Gemini 3.1 Flash TTS - Google's new speech model with inline tags for voice direction across 70+ languages
Deep Max
Deep Max - Exa's new SOTA agentic search tool
Lyra 2.0
Lyra 2.0 - NVIDIA's new model that turns text and camera paths into explorable 3D scenes
xAI Custom Voices
Custom Voices - xAI's new tool to clone your voice from short clips for use within Grok's applications
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.