Try New Grok Imagine here!
Log-inSign-up

Hunyuan World Image to 3D

fal-ai/hunyuan_world/image-to-world
Hunyuan World 1.0 turns a single image into a panorama or a 3D world. It creates realistic scenes from the image, allowing you to explore and view it from different angles.
Inference
Commercial use
Schema
LLMs
PlaygroundAPI

Input

Additional Settings

Customize your input with more control.

Result

Idle

What would you like to do next?

Your request will cost $0.3 per 1 request.

Logs

Hunyuan World 1.0 | [image-to-3d]

Tencent's Hunyuan World 1.0 transforms single images into explorable 3D panoramas at $0.30 per generation. Trading traditional multi-view reconstruction for direct image-to-world conversion, it skips mesh generation entirely, delivering navigable scenes instead of static 3D objects. Built for game developers and virtual environment creators who need spatial context, not just isolated assets.

Use Cases: Game Environment Prototyping | Virtual Tour Generation | Spatial Scene Expansion

Performance

At $0.30 per generation, Hunyuan World positions as a specialized scene expansion tool rather than a volume production endpoint, roughly 10x the cost of traditional image-to-3D mesh generators but solving a fundamentally different problem.

MetricResultContext
Output FormatInteractive world filePanoramic scene vs static mesh
Input RequirementsSingle image + semantic labelsRequires foreground/class annotations
Cost per Generation$0.303.3 generations per $1.00 on fal
Export OptionsDRC format (optional)Dynamic Resource Configuration for scene streaming

Label-Guided Scene Understanding

Hunyuan World diverges from standard image-to-3D reconstruction by requiring explicit semantic guidance through three label parameters: two foreground object sets and scene classes. This isn't automated inference, you're directing spatial interpretation.

What this means for you:

  • Controlled spatial hierarchy: Define which elements occupy foreground layers (`labels_fg1`, `labels_fg2`) versus background context, determining depth relationships and occlusion handling in the generated panorama

  • Scene-aware expansion: The `classes` parameter shapes how the model extrapolates beyond image boundaries. "nature, landscape" produces different spatial logic than "interior, architectural"

  • Panoramic navigation over mesh export: Output is an interactive world file designed for camera movement through the scene, not a traditional GLB/FBX mesh for asset pipelines

  • Optional DRC streaming: Enable Dynamic Resource Configuration export for progressive loading in web-based 3D viewers or game engines with LOD systems

Technical Specifications

SpecDetails
ArchitectureHunyuan World 1.0
Input FormatsImage URL (JPG, PNG, WebP, GIF, AVIF)
Output FormatsWorld file (interactive scene), optional DRC
Annotation Requirements2 foreground label sets + scene classes
LicenseCommercial use permitted

API Documentation | Quickstart Guide | Enterprise Pricing

How It Stacks Up

Tripo3D Image to 3D ($0.039) – Hunyuan World trades mesh-based asset generation for panoramic scene expansion at 8x the cost. Tripo3D delivers animation-ready 3D models with standard export formats (GLB, FBX) ideal for game asset pipelines and AR applications. Hunyuan World prioritizes spatial context and navigable environments for pre-visualization and virtual tours where scene coherence matters more than exportable geometry.