Hunyuan3D V3

Text-to-3d generates geometry from scratch using natural language descriptions, whereas image-to-3d reconstructs a 3D mesh from an existing 2D concept image or photograph. Choose image-to-3d when you have a defined visual concept and require faithful shape reproduction; choose text-to-3d for rapid concept ideation when you do not have reference artwork.

Sketch-to-3d uses a 2D line drawing or CAD outline to define physical contours while using a text prompt as a material shader directive. This gives artists precise silhouette control that pure text prompts cannot match, turning hand-drawn outlines directly into textured 3D assets.

Select quadrilateral polygon layouts when importing meshes into DCC tools like Maya, Blender, or ZBrush for subdivision modeling, sculpting, and character skinning. Select triangle layouts when exporting directly into real-time engines such as Unreal Engine, Unity, or WebGL environments.

Target face counts range from 40,000 up to 1,500,000 polygons. Keep the budget between 40,000 and 100,000 faces for mobile games and AR previews, and scale up to 1,500,000 for high-detail hero assets, intricate organic sculpts, or cinematic renders.

The model specializes in single, isolated objects and struggles with complex multi-object scenes, interlocking mechanical assemblies, refractive transparent glassware, and legible typographic decals. Back-facing geometry on single-image inputs may also require light manual touch-up.