Hunyuan World 2.0 Beginner Tutorial: From Signup to Your First 3D World

Introduction

I previously wrote an overview of Hunyuan World 2.0 covering what it is and what it can do. This article is the hands-on tutorial — a step-by-step walkthrough from signup to generating your first 3D world.

Whether you are a game developer, designer, or just curious about AI, this guide gets you actually running the tool.

Step 1: Register and Access the Platform

HY-World 2.0 can be used two ways:

1. Online platform (recommended for beginners): Visit 3d-models.hunyuan.tencent.com/world
2. Local deployment (advanced): Clone from the GitHub repo and run it yourself

For first-time users, start with the online platform. After opening the site:

• Click "Register" or log in with WeChat QR code scan
• On first use, you must accept the terms of service
• ⚠️ Note: Users outside mainland China may need a VPN to access the platform

Step 2: Prepare Your Input Material

HY-World 2.0 supports several input types:

| Input Type | Description | Recommended Quality | | --- | --- | --- | | Text prompt | Describe the scene you want | Be detailed — include lighting, style, elements | | Single image | Upload one photo; AI fills in the other sides | High resolution, clear subject | | Multiple images | Photos from different angles of the same scene | 50%+ overlap between shots | | Video | Short video clip; AI extracts frames for reconstruction | Stable shooting, good lighting |

Beginners: start with a single image. Find a high-quality photo (1080p+ recommended) with a clear subject and not-too-complex background.

Step 3: Upload and Generate

1. After logging in, find the "Upload Image" button on the homepage
2. Select your image and upload it
3. Click "Start Generation"

The process takes about 20 minutes on the online platform. During this time the model:

• Analyzes the spatial structure of your input image
• Predicts the scene in all four directions (front, back, left, right)
• Generates point cloud data using 3D Gaussian Splatting (3DGS)
• Builds a fully explorable environment

You will get a notification when generation is complete.

Step 4: Explore Your 3D World

Once generated, explore directly in your browser:

• Mouse drag: Rotate the view
• WASD keys: Move forward, left, backward, right
• Shift + movement: Run faster
• Scroll wheel: Zoom in/out

Useful controls on the interface:

1. Character switching: Press Tab to swap in different characters (people, animals, robots) to explore the scene
2. Quality toggle: Switch between "HD" and "Smooth". Smooth mode has higher frame rate but less detail
3. Character scaling: Worlds are generated at human-eye height (~150-170cm) by default, adjustable as needed
4. Lighting/shadow toggle: Turn shadows on/off

Step 5: Export 3D Assets

Generated worlds can be exported in these formats:

| Format | Use Case | Import Into | | --- | --- | --- | | .splat (3DGS) | High-fidelity point cloud rendering | Blender, Unity, dedicated players | | .glb / .obj (Mesh) | Standard 3D mesh models | Blender, Unreal Engine, Three.js | | Depth + Normal maps | Post-processing and texture work | Any 3D software |

Export steps:

1. In the exploration interface, find the "Download" button
2. Choose your format (3DGS or Mesh)
3. After downloading, drag into Blender to preview

Advanced: Local Deployment (Optional)

If you want to run HY-World 2.0 locally:

Requirements

| Item | Minimum | Recommended | | --- | --- | --- | | GPU | RTX 4080 (16GB VRAM) | RTX 4090 (24GB VRAM) | | System RAM | 32 GB | 64 GB | | Disk Space | 80 GB | 200 GB | | CUDA | 12.1 | 12.4+ | | Python | 3.10 | 3.10 | | PyTorch | 2.4.0 | 2.4.0+ |

Installation

# 1. Clone the repo
git clone https://github.com/Tencent-Hunyuan/HY-World-2.0
cd HY-World-2.0

# 2. Create a conda environment
conda create -n hyworld2 python=3.10 -y
conda activate hyworld2

# 3. Install PyTorch (CUDA 12.4)
pip install torch==2.4.0 torchvision==0.19.0 \
  --index-url https://download.pytorch.org/whl/cu124

# 4. Install dependencies
pip install -r requirements.txt
pip install flash-attn --no-build-isolation

# 5. Launch the Gradio demo
python -m hyworld2.worldrecon.gradio_app

Then visit http://localhost:7860 in your browser.

Using the Python API

from hyworld2.worldrecon.pipeline import WorldMirrorPipeline

# Load the model (~1.2B params, BF16)
pipeline = WorldMirrorPipeline.from_pretrained('tencent/HY-World-2.0')

# Reconstruct a 3D scene from an image folder
result = pipeline('path/to/images')

# Or from a video
result = pipeline('path/to/video.mp4')

FAQ

The generated quality isn't great — what should I do?

• Make sure your input image is high resolution (1080p+)
• Keep the subject away from image edges
• Even lighting with no heavy shadows works best
• Simpler backgrounds produce cleaner results

Why can't I access the platform?

• Users outside mainland China may need a VPN to a CN node
• Make sure you accepted the terms of service (required on first use)
• The platform sometimes restricts new accounts temporarily — try again in a few days

Can I use generated worlds in my game?

At this stage, HY-World 2.0 outputs are great for prototyping and greyboxing, but not yet at AAA game quality. For indie games, concept validation, and rapid prototyping, they are more than sufficient.

Not enough VRAM for local runs?

• Lower the input resolution (minimum 50K pixels supported)
• Enable --enable_bf16 to reduce memory usage
• If you have a 16GB card, consider upgrading to a 24GB RTX 4090

Summary

Hunyuan World 2.0 is currently the closest tool to "describe a scene, get a walkable 3D world." For beginners, the simplest path is:

1. Online platform → register → upload image → wait 20 min → explore and export
2. Want more control? → Try local deployment

If you want the deeper technical breakdown of what it is, who it's for, and why it matters for small teams, check out my intro article.

---

Sources: Tencent HY-World 2.0 GitHub, HuggingFace Model Page, Official Technical Report