GitHub radar
Turn one photo of an object into a 3D model
speridlabs released Idol, a model that builds a 3D Gaussian-splat object from a single picture, including sides the photo never showed.
speridlabs is a small AI team that builds 3D and image tools. Their model Idol turns one photo of an object into a 3D Gaussian splat — a cloud of points that stores the object's shape and color and looks three-dimensional when you rotate it. The model removes the background itself, crops the object and fills in sides the original photo never showed, up to a million points in one file. It accepts a regular photo, an AI-generated image, or a render from 3D software. In the team's own tests, Idol was compared against three open rivals, TripoSplat, TRELLIS and LGM, and it scored best on almost every accuracy and sharpness metric. The one exception is a metric where a rival with a visibly blurrier result scores higher, because that particular metric does not penalize blur as strictly as the others.
Why a vibe-coder should care
This is not a replacement for careful 3D modeling, but a fast way to get a rough 3D object out of a single photo — useful for game prototypes, visualizations, or generating 3D training data for other models. It runs on a regular gaming GPU in a couple of seconds, so you can run through dozens of versions instead of waiting on a render farm.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up the Idol model from speridlabs for me, from Hugging Face: https://huggingface.co/speridlabs/idol — run it on my GPU, turn a photo of an object into a 3D Gaussian splat file, and show me how to open it.
A regular gaming GPU is enough: peak load is about 7 GB of VRAM. On an RTX 4090 the result is ready in under a second, on weaker cards it takes up to a minute.
Open on Hugging Face▌ More finds