Overview
TripoSR is an open-source, feed-forward model for fast 3D reconstruction from a single image, developed jointly by Tripo AI and Stability AI. Built on the Large Reconstruction Model approach, it generates a textured 3D mesh in under a second on an NVIDIA A100, inferring unseen geometry and texture automatically. The code is MIT-licensed on GitHub with weights on Hugging Face.
Our take
TripoSR reconstructs 3D objects from a single image, released openly under a permissive licence with code and weights available. For developers building 3D features it is a genuinely useful base — fast, free and modifiable. Output quality is what single-image reconstruction currently achieves, which means usable geometry and unresolved detail on unseen surfaces.
Worth it if
Developers integrating image-to-3D into their own products without licensing costs.
Look elsewhere if
Non-technical users — this is a model to build with rather than a product to use.
✨ Features
Single-Image to 3D
Reconstructs a full textured 3D mesh from just one input image.
Sub-Second Inference
Generates models in under 0.5 seconds on an NVIDIA A100 GPU.
Low Compute Footprint
Runs under modest inference budgets, even without a dedicated GPU.
Open Weights & MIT License
Code and model weights are freely available on GitHub and Hugging Face.
⚖️ Pros & Cons
Pros
Cons
💰 Pricing
TripoSR
Free
- MIT-licensed open-source model
- Weights available on Hugging Face
- Runs locally on GPU or CPU
Categories
❓ Frequently Asked Questions
How good is single-image reconstruction?
Reasonable on the visible geometry and necessarily invented on the back of the object, because the model has no information about it. That is inherent to the problem rather than a flaw in this implementation. Multi-view approaches produce better complete geometry when you can supply more than one photograph.
What does an MIT licence permit?
Commercial use, modification and redistribution with attribution, which makes it usable inside a commercial product without a licensing negotiation. That is rarer in AI models than it should be — many open-weight releases carry restrictions on commercial use that only appear on a careful reading.