Idol: Turn one picture of an object into a 3D Gaussian splat

What it does

Idol is an image-to-3D model from Sperid Labs that turns one picture of an object into a 3D Gaussian splat (.ply), including the sides the picture never showed. It removes the background from your picture, frames the object, and generates up to 1,048,576 Gaussians, which are small 3D blobs that together form the object. The result is saved as a standard 3DGS .ply file that opens in any Gaussian splat viewer.

How to use it

Accept the terms of the two gated image encoder models on Hugging Face and log in. Then install the idol-sdk code from GitHub and run the sample command on a picture, choosing real mode for photographs or synthetic mode for renders. Idol's weights, about 3 GB, and the two image encoders download automatically into the Hugging Face cache on the first run.

Common use cases

• Give it a photograph in real mode, and it writes a .ply 3D splat, the cropped input image and a predicted camera file.
• Point it at a folder of rendered images in synthetic mode, and it writes one .ply file for each image into the output folder.
• Pass a generated image from Python with a fixed seed, and you get the same 3D object again, saved as a .ply file.