Meshy v7

Image-to-3D is a reconstruction problem with missing data: a single photo shows one side of an object, and everything behind it has to be inferred. Meshy v7 is good enough at that inference to make a single clean photo a legitimate input, and Ultra mode is the setting that spends more on the parts a single view has to guess — finer surface detail and tighter geometry, at the cost of a longer run. Ultra is v7-only and single-image only. When you have more than one view, use them instead: one to four photos of the same object from different angles switch Genfire to the multi-image endpoint, where the model measures the silhouette from several directions rather than inventing it. Meshy v6 is still selectable and takes exactly the same inputs minus Ultra, which is what you pin when you need to reproduce a model you generated earlier.

The controls are the part that decides whether the output is a curiosity or an asset. Target polycount runs from 100 to 300,000 with a default of 30,000, so the same source photo can produce a background prop and a hero object. Topology is a real choice — quad for surfaces that will be subdivided or deformed, triangle for detail that goes straight into an engine — and low-poly mode ignores both in favor of clean game geometry. Textures are optional, cheaper to skip, and steerable with a short texture prompt such as "weathered bronze, green patina in the recesses". PBR adds the metallic, roughness and normal maps a modern engine expects. Auto-rigging targets clearly limbed humanoids, takes an approximate height in meters, and can be pinned to an A-pose or T-pose so the skeleton lands where a rigger would put it, with an animation preset applied on top.

Because a run takes 5 to 10 minutes, everything about it is asynchronous by design: the studio hands the job to the server and shows the finished mesh in a model-viewer window when it lands, with a download button per format, and the API returns a queued run you poll for model_url and the full format set. The endpoint is POST /v1/models/3d/generations with the alias 3d.meshy_v7 (or 3d.meshy_v6 to pin the previous generation); the MCP tool is genfire_generate_3d_model; the workflow builder exposes it as a 3D Model node you can wire an image node into; the game builder uses it to turn planned props into GLB assets; and the Blender extension on /plugins generates models at the 3D cursor without leaving the file. For the full tool walkthrough — uploading, options, formats and what to do with the result — see the AI 3D model generator page. Genfire is pay-as-you-go: credit packs start at $19, purchased credits last 12 months, and every purchase includes a commercial license.