Image to 3D (Idol)

Turn a single object or character image into a rotatable 3D Gaussian model, with batch processing and PLY downloads.

60 credits/useImage inputMax 40 MB

Input

Original image · PNG / JPEG / WebP

Image formats

pngjpgjpegwebp

Output

3DGS PLY · Camera parameters · Generation record · ZIP

Use case

3D displays of character art

Models, parameters, and sources

Idol encodes images with DINOv3 and FLUX.2 VAE in FP32. Generation retains native FP16/FP32 precision; PLY retains all Gaussians and FP32 attributes.

Idol

No paper links are provided.

Defaults: 50 steps, guidance 3 and time shift 3. Keep the input and seed fixed when comparing settings.
More Gaussians can represent finer structures but require more time, storage and browser GPU memory. They do not improve every image.
The result is a 3D Gaussian model without mesh topology, UVs or bones. Unseen surfaces are inferred and may differ from the original character.

Use a complete subject with a clear silhouette and a simple background. Transparent backgrounds are retained.
Defaults: 50 steps, guidance 3 and time shift 3. Keep the input and seed fixed when comparing settings.
The result is a 3D Gaussian model without mesh topology, UVs or bones. Unseen surfaces are inferred and may differ from the original character.

Turn a single object or character image into a rotatable 3D Gaussian model, with batch processing and PLY downloads.

This tool is temporarily unavailable for new tasks. Generation will resume when it is available.

Use a complete subject with a clear silhouette and a simple background. Transparent backgrounds are retained.

The result is a 3D Gaussian model without mesh topology, UVs or bones. Unseen surfaces are inferred and may differ from the original character.

Include the tool name, steps and the problem in your feedback. Screenshots can help us locate it.