Tripo AI

Tripo AI is the product line of VAST, a generative 3D company whose research code ships from the VAST-AI-Research organisation on GitHub. Tripo turns text prompts, single images, multiple views and rough sketches into textured 3D assets, and it runs in the browser and over a REST API rather than on a local machine.

Tripo Studio is the creator facing workspace, where models are generated, edited and refined directly in a browser. It supports segmentation, symmetry, pose adjustment and detail tuning, and exports to GLB, FBX, OBJ, USD, STL and 3MF, with automatic rigging for characters and creatures. A DCC Bridge carries results into Blender, 3ds Max, Maya, Unity, Unreal, Godot, Cocos and ZBrush. The Tripo API exposes the same generation, refinement and export operations to outside applications. The API moved to a v3 architecture with a dedicated endpoint for each task in April 2026, and the older v2 API stopped receiving updates on 1 October 2026 and stops accepting requests on 1 November 2026, China Standard Time.

Tripo's generation models have moved through several numbered releases since Tripo 2.0 in September 2024. Version 2.5 arrived in January 2025, the same month as the 2.0 rebuild of the web app, and v3.0 reached general release in September 2025. Two model families now run side by side. The H series, led by Tripo v3.1, which reached the API in March 2026, targets high detail output of up to two million polygons for rendering, visual effects and 3D printing. The P series began with the P1 Smart Mesh model in March 2026 and moved to P2 in August 2026, generating clean, engine ready topology for games and real time work, with P2 adding quad meshes of 48 to 25,000 faces. A v3.5 texture model followed in September 2026, with an option that removes baked lighting from the reference image before texturing.

TripoSR is a single image to 3D reconstruction model released with Stability AI, capable of draft quality textured meshes in about half a second on an A100, and it was open sourced under MIT. VAST has since published TripoSG and TripoSF for shape generation and texture control.

TripoSplat is the Gaussian splatting model, released to GitHub on 1 June 2026 and licensed MIT with VAST as the copyright holder, verified from the LICENSE file. It converts a single image into a variable number of 3D Gaussians and carries roughly 1,380 stars and 145 forks as of October 2026. The model is described in the paper Generative 3D Gaussians with Learned Density Control, outputs up to 262,144 Gaussians as .ply or .splat files, and its training code followed in August 2026 in a separate MIT licensed repository from lead author Runjie Yan. It reached ComfyUI as a native generation path, has been compared with Apple's SHARP and Microsoft's TRELLIS, and supplies the splat inside third party tools such as LocalMesh, which runs it on a local GPU. Co-founder Dr. Yanpei Cao has said that treating the 3D Gaussian as the final delivered format is in development rather than shipped, so Tripo's production path still ends in a mesh.

Pricing runs on credits. The free tier gives 200 monthly credits, roughly thirteen models, one concurrent task and non-commercial use only, with models public. Pro at $20 a month gives 3,000 credits, ten concurrent tasks, private models and commercial use. Max at $90 a month gives 25,000 credits and unlimited 8K texture access, and Team is $55 per seat per month billed annually. Credit packs start at $10 for 1,000 credits. As of October 2026 Tripo's homepage cites more than 20 million creators and 200 million generated models.