# Tripo (3D generation)

> Source: https://aiwiki.ai/wiki/tripo
> Updated: 2026-06-09
> Categories: AI Tools & Products, Generative AI
> From AI Wiki (https://aiwiki.ai), a free encyclopedia of artificial intelligence. Quote with attribution.

# Tripo (3D generation)

**Tripo** is an artificial-intelligence platform for generating three-dimensional models from text prompts, single images, multi-view images, or sketches. The platform is developed by **VAST** (also styled "VAST AI"), a Chinese AI startup headquartered in Beijing and founded in 2023 by Simon Song (Song Yachen), a co-founder of the Chinese AI unicorn [minimax](/wiki/minimax).[^1][^2] Tripo's core product is a hosted web service (Tripo Studio) backed by a series of proprietary 3D foundation models that output textured meshes with PBR materials, rigging, and topology suitable for downstream use in game engines, 3D printing, and visual-effects pipelines.[^3][^4]

Since its public launch in 2024, Tripo has released a cadence of versioned foundation models (Tripo 1.0, Tripo 2.0 in September 2024, Tripo 2.5 in January 2025, Tripo 3.0 in September 2025, and the production-oriented Tripo H3.1 and [tripo p1](/wiki/tripo_p1) Smart Mesh P1.0 lines announced in 2026) alongside open-source research releases such as **TripoSR**, jointly published with [stability ai](/wiki/stability_ai) in March 2024, and the TripoSG and TripoSF models open-sourced in March 2025.[^5][^6][^7][^8] By early 2026 the company reported approximately 6.5 million users, more than 90,000 developers using its API, and roughly 100 million 3D assets generated on the platform.[^9][^10]

In March 2026 VAST announced a US$50 million Series A round led by [alibaba](/wiki/alibaba) and Baidu Ventures, alongside the Tripo H3.1 and Tripo P1.0 models, which the company described as a shift toward "production-ready" 3D generation in native spatial geometry rather than sequential token prediction.[^9][^10][^11]

## Key facts

| Field | Value |
|---|---|
| Developer | VAST (also "VAST AI") |
| Founded | 2023 (Beijing; Cayman Islands registration) |
| Headquarters | Beijing, China (additional office in Hangzhou) |
| Founder & CEO | Simon Song (Song Yachen) |
| CTO | Liang Ding |
| Head of research | Yanpei Cao |
| Product type | Generative 3D foundation models (text-to-3D, image-to-3D) |
| Platform name | Tripo Studio (formerly tripo3d.ai) |
| First version | Tripo 1.0 (2024) |
| Latest models (2026) | Tripo H3.1, Tripo P1.0 ([tripo p1](/wiki/tripo_p1)) |
| Open-source releases | TripoSR (2024), TripoSG, TripoSF (2025), UniRig (2025) |
| Series A | US$50 million, March 2026 (lead: [alibaba](/wiki/alibaba), Baidu Ventures) |

## Background and founders

VAST was founded in 2023 in Beijing, with corporate registration in the Cayman Islands.[^1][^2] The company was established shortly after Google's October 2022 announcement of DreamFusion, a text-to-3D research project that Song has credited with making the field feel commercially tractable.[^1] By 2025 VAST employed more than 100 people across offices in Beijing and Hangzhou.[^2]

The founding leadership comprises three people. **Simon Song** (also rendered as Song Yachen) serves as founder and chief executive. Prior to VAST he worked in the CEO's office at [sensetime](/wiki/sensetime) and was a co-founder of [minimax](/wiki/minimax), one of China's so-called "AI tiger" unicorns. **Liang Ding**, the chief technology officer, also previously worked at [sensetime](/wiki/sensetime) and is an alumnus of Tsinghua University.[^1][^12] **Yanpei Cao**, head of research, joined from [tencent](/wiki/tencent), where he was a principal researcher working on 3D reconstruction.[^1] Several core authors of Tripo's published research papers are co-listed with Tsinghua-affiliated coauthors, and news coverage has characterized the technical team as drawing heavily from Tsinghua University.[^12]

VAST's stated mission, articulated by Song in multiple interviews, is to "let everyone create 3D content with zero barriers and zero cost," with the longer-term aspiration of cultivating a user-generated 3D content ecosystem.[^1][^2] Although the company is China-founded, by 2026 more than 85 percent of its end-users were located outside China, with the United States and Europe its largest markets.[^2][^11]

## Products

### Tripo Studio

Tripo's user-facing product is **Tripo Studio**, a browser-based 3D workspace that wraps the company's foundation models with project management, editing, and export tooling. Users upload an image, type a text prompt, or import a sketch; the service returns a textured mesh that can be previewed in-browser and exported in formats including OBJ, FBX, GLB, and STL.[^3][^13]

Studio combines several generation primitives:

- **Image-to-3D**: single or multi-view image inputs are converted to a complete textured mesh.
- **Text-to-3D**: natural-language prompts produce geometry and materials end-to-end.
- **Sketch-to-3D**: a pipeline introduced with Tripo 3.0 that accepts hand-drawn or vector sketches.
- **Auto-texturing**: separate texturing of supplied or generated meshes, including high-fidelity PBR maps.
- **Auto-rigging and animation**: automatic skeleton generation and skinning (see *Technology* below).
- **Smart Part Segmentation** and **Magic Brush 2.0**: semantic editing of generated meshes.[^4][^13]

The platform integrates with [gpt 4o](/wiki/gpt_4o) and [flux](/wiki/flux) for prompt-to-image preconditioning prior to 3D generation.[^4]

### API and integrations

VAST exposes its models through a public API and ships official extensions for Blender and [comfyui](/wiki/comfyui), both maintained on GitHub under the `VAST-AI-Research` organization.[^14][^15] Tripo is integrated into workflows used by enterprise customers including [tencent](/wiki/tencent), NetEase, ByteDance, Microsoft, HTC, Pop Mart, and Sony.[^2][^11][^16] In the consumer 3D-printing ecosystem the company has partnerships with Bambu Lab, Anycubic, Creality, Elegoo, and Stratasys.[^3]

## Versions and releases

The Tripo platform has shipped through a series of numbered foundation-model releases:

- **Tripo 1.0**: the initial public model. According to a VoxelMatters profile, the first release generated a mesh in roughly eight seconds.[^3]
- **Tripo 2.0** (September 2024): combined a Diffusion Transformer (DiT) backbone with a U-Net for detail preservation, introduced high-fidelity PBR material generation, and added negative prompts, pose customization, and proportion controls.[^17][^13]
- **Tripo 2.5** (released as `v2.5-20250123` on 23 January 2025): focused on geometric detail and PBR fidelity for production-quality miniatures and game assets, with deeper [comfyui](/wiki/comfyui) and Blender integration.[^18]
- **Tripo 3.0** (September 2025): described by Song as the transition from "fascinating toy" to "pipeline-ready industrial tool." Tripo 3.0 raised model parameter count by approximately 20× over prior versions; its **Ultra Mode** outputs meshes of up to two million polygons, while **Standard Mode** produces topology-optimized assets for real-time rendering. The release also added T-pose character export with rigging, Magic Brush 2.0, Smart Part Segmentation, and a sketch-to-3D pipeline.[^4][^19]
- **Tripo H3.1** and **Tripo P1.0**: announced March 2026 alongside the Series A. H3.1 targets high-fidelity geometry for industrial design, 3D printing, and cinematic work; P1.0 (covered in detail at [tripo p1](/wiki/tripo_p1)) is optimized for real-time graphics and interactive environments, generating topology-aware meshes for game engines and robotics in as little as two seconds.[^9][^10][^11]

## Technology

Tripo's underlying models combine several architectural threads developed at VAST and in collaboration with academic and industrial partners. The Tripo 2.0 architecture publicly disclosed in 2024 mixed a **Diffusion Transformer (DiT)** for capturing long-range structural information with a **U-Net** for preserving fine geometric detail; this dual backbone was trained on what the company described as "tens of millions" of high-quality 3D assets.[^17][^13][^3] By the Tripo 3.0 release, VAST reported a training dataset of approximately 40 million curated 3D assets, expanding to roughly 50 million by the time of the March 2026 announcement.[^19][^9]

Two innovations introduced during 2025 are central to the current model family:

- **SparseFlex**, the representation underpinning the open-source TripoSF and the Tripo 3.0 family, replaces a dense voxel grid with sparse voxels concentrated near object surfaces. This allows the network to allocate capacity to surface detail, supports arbitrary topology (including non-manifold and open-surface assets such as fabrics and foliage), and according to VAST's benchmarks reduces Chamfer Distance by approximately 82 percent and improves F-score by approximately 88 percent over prior open baselines.[^7]
- **Rectified Flow with Mixture-of-Experts**, used in the open-source TripoSG, replaces standard diffusion denoising with a more direct noise-to-data path and adds MoE routing, the first reported application of MoE to 3D generation, according to VAST's research notes.[^7]

For animation, VAST released **UniRig** at SIGGRAPH 2025 as a unified framework for skeletal rigging across human, animal, stylised, and inorganic models. The Tripo platform's auto-rigging feature, which finds joints and bones and performs AI-driven skinning, is built on this research.[^20][^21]

## Open-source releases

Although Tripo's flagship models are proprietary, VAST has been an active open-source contributor. In late 2025 the company stated that it had published more than 50 research papers and open-sourced more than 20 projects, accumulating roughly 18,000 GitHub stars in aggregate.[^3]

### TripoSR

**TripoSR** is a single-image-to-3D reconstruction model released on 5 March 2024 in partnership with [stability ai](/wiki/stability_ai).[^5][^6] Drawing on the Large Reconstruction Model (LRM) line of research, TripoSR produces textured meshes from a single image in under 0.5 seconds on an Nvidia A100, runs without a GPU at higher latency, and was trained on a curated CC-BY subset of Objaverse.[^5][^22] Authors include Dmitry Tochilkin, David Pankratz, Zexiang Liu, Zixuan Huang, Adam Letts, Yangguang Li, Ding Liang, Christian Laforte, Varun Jampani, and Yan-Pei Cao; the paper is available as arXiv:2403.02151.[^22] Code is released under the MIT licence on the `VAST-AI-Research/TripoSR` repository, with weights on Hugging Face.[^5][^22]

### TripoSG and TripoSF

In March 2025 VAST launched what it branded a **Technology Open-Source Month**, beginning with TripoSG and TripoSF.[^7]

- **TripoSG** is a 1.5-billion-parameter rectified-flow Transformer with mixture-of-experts routing. It was trained on roughly two million curated image-SDF pairs and uses signed-distance-function encoding rather than occupancy grids.[^7]
- **TripoSF** introduces the SparseFlex representation described above and is presented as state-of-the-art among open models on standard 3D-generation benchmarks.[^7]

### UniRig and Blender/ComfyUI extensions

VAST maintains an official **Blender extension** (`tripo-3d-for-blender`) and **ComfyUI** custom nodes (`ComfyUI-Tripo`), both under the `VAST-AI-Research` GitHub organization, allowing artists to generate and import Tripo models without leaving their existing tools.[^14][^15] The **UniRig** SIGGRAPH 2025 paper and codebase provide automatic skeletal rigging for diverse mesh types.[^20]

## Reception

Industry reviews have generally placed Tripo among the strongest commercial 3D-generation services. A 2025 comparative benchmark by Ideate ranked Tripo's image-to-3D generation as the fastest among major providers, completing typical jobs in roughly ten seconds, and praised its handling of organic subjects such as fantasy characters and complex natural geometry.[^23] Coverage by Skywork.ai's review of Tripo 3.0 described it as the first AI 3D generator that "actually feels professional," noting strong topology and reliable PBR materials for production pipelines.[^24] VoxelMatters characterized VAST as on a "fast track to an AI-native 3D future," highlighting its dual focus on research and a commercial platform.[^3]

Media coverage in the South China Morning Post and KrASIA has framed VAST as one of a small group of Chinese AI startups competing directly with [tencent](/wiki/tencent)'s Hunyuan 3D effort and with Western players, while operating predominantly on a global user base.[^1][^2][^12]

## Competition

Tripo competes in the rapidly developing market for AI-generated 3D assets. Its principal competitors include:

- **[hunyuan 3d](/wiki/hunyuan_3d)** (Tencent): an open-and-closed family of 3D generation models from [tencent](/wiki/tencent) that, by 2025, included Pro and Rapid editions supporting text, image, and multi-view inputs. Reviews consistently praise Hunyuan 3D for hard-surface mechanical geometry and high-resolution PBR texture maps.[^23]
- **Meshy**: a competing commercial platform widely used by indie game developers, noted for SOC2/ISO 27001 compliance and a Smart Remesh feature aimed at low-poly mobile assets.[^23]
- **Rodin** (Hyper3D / DeemoS): another Chinese-founded 3D-generation product targeted at digital content creators.
- **TRELLIS** (Microsoft): a research-derived open model that has been widely benchmarked against Tripo on image-to-3D tasks.[^25]

Comparative reviews across 2025-2026 broadly position Tripo as the speed leader and strongest performer on organic subjects, Hunyuan 3D as strongest on hard-surface and photorealistic work, and Meshy as the most documented and enterprise-compliant offering.[^23][^25]

## Funding

Public funding details for VAST are limited; the company has not disclosed all rounds and PitchBook and Tracxn carry incomplete information.[^26][^27]

- **Series A, March 2026.** VAST announced a US$50 million Series A round on 25 March 2026 led by [alibaba](/wiki/alibaba) and Baidu Ventures. The company stated the capital would support continued development of large-scale 3D foundation models and expansion of its global developer platform across gaming, robotics, manufacturing, and immersive media. The announcement was paired with the Tripo H3.1 and Tripo P1.0 model launches.[^9][^10][^11]

At the time of the Series A, VAST reported that the company was profitable, with annual revenues of approximately US$12 million and a monthly revenue run-rate near US$1 million.[^3]

## See also

- [tripo p1](/wiki/tripo_p1): Tripo Smart Mesh P1.0, the real-time/topology-aware model announced in March 2026
- [hunyuan 3d](/wiki/hunyuan_3d): competing Chinese 3D-generation family from [tencent](/wiki/tencent)
- [stability ai](/wiki/stability_ai): collaborator on TripoSR
- [minimax](/wiki/minimax): earlier company co-founded by Tripo CEO Simon Song
- [sensetime](/wiki/sensetime): previous employer of Song and Liang Ding
- [comfyui](/wiki/comfyui): supported workflow integration

## References

[^1]: Coco Feng, "Ex-MiniMax co-founder's new venture Vast aims to challenge Tencent, Google in 3D AI models," *South China Morning Post*, 2025. https://www.scmp.com/tech/tech-trends/article/3323106/ex-minimax-co-founders-new-venture-aims-challenge-tencent-and-google-3d-ai-models
[^2]: "China-founded Tripo AI updates 3D content creation platform as users more than double," *South China Morning Post*, December 2025. https://www.scmp.com/tech/tech-trends/article/3338202/china-founded-tripo-ai-updates-3d-content-creation-platform-users-more-double
[^3]: "Tripo AI's fast track to an AI-native 3D future," *VoxelMatters*. https://www.voxelmatters.com/tripo-ais-fast-track-to-an-ai-native-3d-future/
[^4]: "Tripo v3.0 Ultra Official Release: A New Benchmark in AI 3D Creation," Tripo AI blog. https://www.tripo3d.ai/blog/introducing-tripo-new-algorithm3
[^5]: "Introducing TripoSR: Fast 3D Object Generation from Single Images," Stability AI, 5 March 2024. https://stability.ai/news/triposr-3d-generation
[^6]: "Stability AI and Tripo AI release image-to-3D AI model TripoSR," *CG Channel*, March 2024. https://www.cgchannel.com/2024/03/stability-ai-and-tripo-ai-release-image-to-3d-ai-model-triposr/
[^7]: "VAST Open Source Month | TripoSG & TripoSF, Setting a New SOTA in 3D Generation," Tripo AI blog, March 2025. https://www.tripo3d.ai/blog/vast-open-source-month
[^8]: "Introducing TripoSR by Tripo and Stability AI: High-Quality 3D Models in Under a Second," Tripo AI blog. https://www.tripo3d.ai/blog/stabilityai-with-tripo
[^9]: "Tripo AI Announces $50 Million in Funding and New Models for Production-Ready 3D Generation," PR Newswire, 25 March 2026. https://www.prnewswire.com/news-releases/tripo-ai-announces-50-million-in-funding-and-new-models-for-production-ready-3d-generation-302724894.html
[^10]: "Tripo AI Announces $50M in Funding and New Models for Production-Ready 3D Generation," The AI Insider, 27 March 2026. https://theaiinsider.tech/2026/03/27/tripo-ai-announces-50m-in-funding-and-new-models-for-production-ready-3d-generation/
[^11]: "Tripo AI Bets on US Market as $50 Million Backing Puts AI-Powered 3D Creation in Play," *Benzinga*, March 2026. https://www.benzinga.com/news/gaming/26/03/51529484/tripo-ai-bets-on-us-market-as-50-million-backing-puts-ai-powered-3d-creation-in-play
[^12]: "Tsinghua Team's New Work Tripo 2.0 Ignites the Generative 3D Field with More Realistic Textures!," AIbase. https://www.aibase.com/news/11900
[^13]: "Vast's Tripo AI brings generative AI to 3D modeling," KrASIA. https://kr-asia.com/vasts-tripo-ai-brings-generative-ai-to-3d-modeling
[^14]: "tripo-3d-for-blender: Official extension for Blender," VAST-AI-Research GitHub. https://github.com/VAST-AI-Research/tripo-3d-for-blender
[^15]: "ComfyUI-Tripo: Official custom nodes for using Tripo in ComfyUI," VAST-AI-Research GitHub. https://github.com/VAST-AI-Research/ComfyUI-Tripo
[^16]: "Sony × VAST: Opening the Future of 3D Creation," Tripo AI blog, 2025. https://www.tripo3d.ai/blog/sony-and-vast-3d-business-collabration
[^17]: "Tsinghua Team's New Work Tripo 2.0 Ignites the Generative 3D Field with More Realistic Textures!," news.aibase.com. https://news.aibase.com/news/11900
[^18]: "TRIPO 2.5: A Deep Dive into AI-Powered 3D Modeling," Toolify AI. https://www.toolify.ai/ai-news/tripo-25-a-deep-dive-into-aipowered-3d-modeling-3447774
[^19]: "VAST Launches Tripo 3.0! The New King of AI 3D Modeling Emerges, A 40 Million Data Set Shakes the World," AIbase. https://www.aibase.com/news/20740
[^20]: "UniRig: [SIGGRAPH 2025] One Model to Rig Them All: Diverse Skeleton Rigging with UniRig," VAST-AI-Research GitHub. https://github.com/VAST-AI-Research/UniRig
[^21]: "One Model to Rig Them All: VAST/Tripo Introduces UniRig for Diverse, Automated 3D Rigging," Tripo AI blog. https://www.tripo3d.ai/blog/unrig-automated-3d-rigging
[^22]: "TripoSR: Fast 3D Object Reconstruction from a Single Image," VAST-AI-Research/TripoSR GitHub. https://github.com/VAST-AI-Research/TripoSR (paper: https://arxiv.org/abs/2403.02151)
[^23]: "AI 3D Model Comparison: Trellis vs Tripo vs Meshy vs Rodin vs Hunyuan," Ideate. https://ideate.xyz/blogs/posts/ai-3d-model-comparison-trellis-tripo-meshy-rodin-hunyuan
[^24]: "Tripo AI 3.0 In-Depth Review: The AI 3D Generator That Actually Feels Professional," Skywork AI. https://skywork.ai/skypage/en/Tripo-AI-3.0-In-Depth-Review-The-AI-3D-Generator-That-Actually-Feels-Professional/1974872733789122560
[^25]: "Best AI 3D Model Generators in 2026: TRELLIS vs Meshy vs Tripo vs Hitem3D," trellis2.app. https://trellis2.app/blog/best-ai-3d-model-generator
[^26]: "Tripo: 2026 Company Profile, Team, Funding & Competitors," Tracxn. https://tracxn.com/d/companies/tripo/__036uK0gopl9hkHyDnvulm4mmIjAq_1eqHgQumDjUF-w
[^27]: "Tripo AI 2026 Company Profile: Valuation, Funding & Investors," PitchBook. https://pitchbook.com/profiles/company/589956-94