TripoSR

The model that made image-to-3D feel real. TripoSR reconstructs a 3D mesh from a single photo in well under a second on a consumer GPU — fast enough to sit inside an interactive tool.

Type · feed-forward reconstruction Licence · MIT (open weights) By · Stability AI + Tripo AI Speed · <1s on an A100-class GPU
TL;DR

TripoSR turns one image into a textured 3D mesh almost instantly, with open MIT weights you can run locally. It is not the highest-fidelity generator anymore — newer models beat it on detail — but its speed, openness and simplicity make it the best starting point for image-to-3D and the reference others are measured against.

What it is

TripoSR is a feed-forward single-image 3D reconstruction model built on the LRM (Large Reconstruction Model) approach, released by Stability AI and Tripo AI. Unlike diffusion-based generators that iterate, it predicts geometry in a single forward pass, which is why it is so fast. You give it one image of an object on a clean background and get back a mesh you can clean up and print.

Where it wins

Where it still hurts

The AI angle

TripoSR is itself the AI — it is a useful anchor in this category and a clean benchmark target. On 3d.2nth.ai it also doubles as an eval subject: it is the kind of open model whose quality, latency and cost we can compare directly against newer image-to-3D systems, which is exactly the testbed this site runs.

Start here

← More AI 3D & scanning tools