An AI 3D model generator that does not stop at the mesh.

Start from an image when you already know what the subject looks like. Concept art, a clean render or a product shot fixes the silhouette, proportions and colours, and the seven-view comparison tells you quickly whether the generator read it correctly. This is the most controllable route and the one the image to 3D page describes in detail.

Start from a photograph when the subject exists in the real world: a toy, a figurine, a tool, a pet. Photographs bring lighting, cast shadows and backgrounds that the generator has to separate from the object, so the photo to 3D model page covers how to shoot one that reconstructs well.

Start from text when there is no picture yet. A paragraph describing the subject, its parts, materials and pose becomes a mesh through the same generator. Results vary more between runs than with an image, so plan to iterate on the mesh before rigging. The text to 3D page has prompt guidance.

Whichever you choose, the mesh is the stage to get right. Rigging and motion are built on it and cannot add geometry that was not generated. If you only need the geometry, for printing or for your own rigging pipeline, turn rigging off and pay for the mesh alone; the 3D mesh generation page explains what the mesh contains.

A generated model can look convincing from the front and still be wrong. Orbit it. Count the appendages and compare them with the input. Check the underside and the back, because those are the surfaces the generator inferred. Look for parts that fused (two legs as one column, a tail merged into a leg) and for parts that vanished (thin straps, whiskers, antennae).

Then read the rig reports rather than trusting the skeleton overlay. The placement score says how well joints sit inside the parts they drive; the rig check card lists what the stage verified; the deformation report measures compression and stretch on the skinned surface across a test motion. The rig inspection guide explains each number and what to do when one is poor.

Finally test with an isolated motion before a complex one: turn the head, bend one leg, rotate one wheel. Nearby surfaces should follow only when that movement makes sense. If they do not, re-rig with a hint, or regenerate the mesh; both are cheaper than discovering the problem after building a scene around the asset.

The fox on this page passed those checks and is shown live from its exported GLB. That is evidence for one subject, and the validation notes describe which structures the regression fixtures cover: repeated limbs, different scales, posture changes, rigid wheel and panel assemblies, multipart meshes.

A credit is about two cents at pack rates. A mesh is 25 credits, a rig 10, a described motion 5 and a critique round 2; presets, the introductory loop, the free converters and repeat exports cost nothing. A complete rigged model is therefore 35 credits, roughly seventy cents.

New accounts get 60 credits without a card. Packs start at $10 for 500 credits and do not expire. Pro ($24 a month, 1,500 credits) adds a commercial license, rig quality reports and API and MCP access; Studio ($79, 6,000 credits) adds unlimited critique rounds. Charges apply when a job is accepted and are refunded if it fails. The pricing page has the table, plan features and a calculator.