Open source

TripoSR

Stability AI + Tripo AI · Mar 2024 · MIT · site · source

Sub-second image→3D reconstruction — the speed champion of open 3D.

What it is

TripoSR is a transformer-based image→3D model released by Stability AI in collaboration with Tripo AI. It reconstructs a textured mesh from a single image in under a second on a modern GPU, which makes it a strong fit for rapid iteration loops and in-browser demos. Quality is ceiling-limited compared to diffusion-based rivals, but nothing else is this fast.

Generated with TripoSR

We don't run TripoSR on OKSLOP yet — samples will appear here once we do.

Honest take

Strengths & limitations

Strengths
  • +Sub-second reconstruction on consumer GPUs
  • +MIT-licensed — fully commercial-friendly
  • +Tiny VRAM footprint (runs in the browser via WebGPU forks)
  • +Great for interactive tools and previewing
Limitations
  • Geometry detail below diffusion-based models
  • Vertex-colored textures only — no PBR material maps
  • Single-view only — no multi-reference conditioning

Prefer a model that's already running?

OKSLOP runs production 3D generation on Hunyuan3D today. Start a brief and get a GLB without picking a provider.