Tripo
AI 3D workspace by VAST that generates textured, riggable models from text, images or sketches in seconds.
CSM (Common Sense Machines)
Image, sketch and text to 3D generator. Alphabet acquired the company in January 2026 and the service is no longer reachable.
AI 3D Generation Freemium
Overview
Common Sense Machines, usually shortened to CSM, was a Cambridge, Massachusetts company building generative models that turned a photo, a sketch or a text prompt into a game-ready mesh. Alphabet acquired it in a deal that closed on 24 January 2026, and roughly a dozen engineers moved to Google DeepMind. This entry is kept as a record rather than a recommendation: as of 2026-09-03 neither csm.ai nor docs.csm.ai answers a request, from two independent networks, so there is nothing to sign up for. It is worth reading anyway, because CSM documented the gap between a generated mesh and a usable one more openly than most of its competitors, and because its disappearance is a useful reminder about where the durable part of this workflow actually sits.
Key features
Cube foundation model, multi-input
The final generation of the model accepted sketches, text and images, exposed adjustable polygon targets, and embedded physics properties such as collision geometry in the exported file rather than leaving them to be authored later.
Part-based generation
Rather than returning one fused shell, CSM could decompose a single image into separate mesh parts and reassemble them. That matters downstream: separate parts can be moved, hidden or printed in different colours, where a fused mesh cannot.
API and Blender integration
A Python API and a Blender add-on speaking the Model Context Protocol let a script or a language model start a generation and pull the result into an open scene, which technical artists rated more highly than the web interface itself.
PBR maps, not just vertex colours
Exports carried albedo, normal and roughness maps, so assets reacted to scene lighting properly. Several contemporaries only baked colour into vertices, which looks correct in a preview and wrong in an engine.
Pros and cons
What users consistently praise
- Turning a concept image into a usable blockout in under a minute was genuinely fast, and reviewers consistently rated it as a real gain for pre-production.
- The Python API and the Blender add-on let technical artists skip the web interface entirely, which was the most praised part of the product.
- Exports carried proper PBR maps rather than baked vertex colour, so assets lit correctly in Unreal and Unity.
- Part-based generation returned separable pieces instead of one fused shell, which is the harder and more useful behaviour.
What users consistently criticise
- Non-manifold geometry in the output was the most consistent complaint. Slicers frequently refused the raw mesh, and the community workflow was to run Merge by Distance, recalculate normals and seal holes in Blender before printing.
- UV islands came out badly fragmented, so editing a texture in Substance Painter usually meant unwrapping the model again from scratch.
- Automatic retopology rounded off hard edges, which made it acceptable for rocks and foliage and unsuitable for vehicles, weapons or anything mechanical.
- No real-world scale awareness: a coffee mug could arrive in Blender several metres across, so every asset needed manual scale normalisation.
- The service ended without a public migration path, stranding work that lived only in the platform.
Recurring themes across third-party user reviews (G2, Capterra, TrustRadius).
Videos
Pricing
Is CSM (Common Sense Machines) free?
CSM (Common Sense Machines) is free to start. There is a free tier you can use indefinitely, and paid plans above it unlock further capability.
| Plan | Price | What you get |
|---|---|---|
| Tinkerer (historical, free) | USD 0 | About 10 generations. Output was published to the public gallery under CC BY 4.0, so anything uploaded on this tier stopped being private. |
| Maker (historical) | USD 20/month | Around 100 generations, private processing, multi-view upload and PBR textures. |
| Creative Pro (historical) | USD 60/month | Around 400 generations with priority processing. |
| Startup (historical) | USD 499/seat/month | Team features for small studios. |
| Enterprise (historical) | Custom | AI retopology, fine-tuning and custom model work. |
Best for
Anyone researching what happened to CSM
The company was acquired and the product went away without a public migration path. This page records the dates, what the tool did, and where its output tended to break, because none of that is on the vendor site any more.
People choosing a generator today
CSM is a worked example of platform risk in this category. The meshes people exported still open; the accounts, credits and cloud history did not survive. That is an argument for exporting early and keeping your own copies, whichever tool you pick.
Who should look elsewhere
Anyone looking for a tool to use now
There is nothing to sign up for. Neither csm.ai nor docs.csm.ai responds as of 2026-09-03. Tripo, Meshy, Rodin and Hunyuan3D cover the same job.
Frequently asked questions
Can I still use CSM?
No. Alphabet acquired Common Sense Machines in a deal that closed on 24 January 2026 and the engineering team moved to Google DeepMind. As of 2026-09-03 neither csm.ai nor docs.csm.ai answers a request, checked from two independent networks on the same day that other vendor sites responded normally.
Was there an official shutdown announcement?
None that we can find. Coverage of the acquisition noted that no statement addressed product direction, and we have located no vendor notice about the service ending. What is verifiable is the acquisition date and that the domains no longer respond.
What happened to models people generated on the platform?
Anything already exported to a GLB, FBX, OBJ or USDZ file is unaffected, because those files sit on the user's own disk. Anything that existed only as cloud history, credits or gallery entries went with the service. This is the practical lesson from the whole episode: the file you exported is the part you own.
What should I use instead?
Tripo, Meshy, Rodin and Hunyuan3D all cover image-to-3D generation and all export a mesh you can download. None of them removes the downstream work CSM users described, so the checks worth running on the output are the same whichever you choose.
Why keep a page for a tool that no longer exists?
Because the question people arrive with is what happened to it, and the vendor site that would have answered is gone. The technical record is also still useful: CSM's failure modes, from non-manifold geometry to fragmented UVs to meshes arriving at the wrong scale, are the same ones every generator in this category produces.
Alternatives to CSM (Common Sense Machines)
See all AI 3D Generation ›
Meshy
AI 3D model generator that turns text prompts or images into textured, riggable meshes in about a minute.
Rodin (Hyper3D)
Deemos' AI 3D generator focused on high resolution quad geometry and PBR textures from text or multi image input.
Hunyuan3D
Tencent's AI 3D generator with open weights for self hosting and a commercial cloud API, strongest on hard surface geometry.