Image to 3D Conversion

The core of SphereLinks is its image-to-3D pipeline. Upload any image and receive a textured, watertight 3D mesh within minutes.

The generation pipeline

Under the hood, SphereLinks runs a two-stage model:

1. Sparse-view reconstruction — The AI infers the 3D geometry from the single input image using an octree-based model (TRELLIS).

2. Texture baking — A texture is projected back onto the mesh surface based on the input image colours.

The result is a GLB file containing the mesh and embedded PBR textures.

Generation parameters

• Steps — More inference steps give a smoother mesh but take longer. Default: 50.

• SLAT steps — Controls detail resolution in the sparse latent space. Default: 25.

• Seed — Fixed seed for reproducible results. Randomise to explore variations.

• Texture size — 512 px for fast results, 1024 px for high-fidelity output.

• Simplification ratio — Reduce poly count (0.9 = light reduction, 0.5 = heavy reduction). Lower is better for real-time use.

Best practices

• Shoot or select images where the subject is fully visible — truncated objects produce incomplete geometry.

• Objects with distinct silhouettes reconstruct better than flat or very thin subjects.

• Natural lighting with soft shadows is better than harsh highlights or studio flash.

Generation time

A typical job takes 60 – 120 seconds. The queue page shows live progress. You can close the browser and come back — jobs run in the background and results are saved to your account history.