Meshy 7 Brings 2D References Closer to Their 3D Reconstructions

Meshy 7 improves geometric fidelity between a reference image and its 3D model, with single-view and multi-view generation plus a new Ultra mode.

Meshy 7 focuses its improvements on one of the more sensitive aspects of Image-to-3D: preserving the geometry of the reference image more faithfully during 3D reconstruction. According to Meshy, the new model improves shape alignment on complex references and works with both single-image and multi-view inputs.

A new Ultra mode pushes this further by prioritizing geometric fidelity and surface detail. Based on Meshy’s launch materials, it is currently limited to single-view generation, with multi-view support planned for a later stage.

The rest of the pipeline builds on several components Meshy had already introduced before this release. The platform supports up to four images of the same object for multi-view reconstruction and includes controls for remeshing, face count, and topology. Smart Topology also uses a dedicated model capable of producing separate parts and targeting a specific face count.

Texturing can reach 8192 × 8192 for the base color map. This capability is not specific to Meshy 7, however, as it was added to the platform in July. At that resolution, PBR maps are still generated in 4K, no emission map is produced, and only triangular meshes are supported.

An asset can then continue through Meshy’s existing toolset with remeshing, auto-rigging, and animation before being exported in different formats depending on the workflow. The API documentation supports formats including GLB, FBX, OBJ, STL, USDZ, and 3MF, depending on the operation.

Meshy 7 is therefore less about introducing an entirely new end-to-end pipeline all at once and more about improving the reconstruction model itself, with a stronger focus on fidelity between the 2D reference and the resulting 3D geometry. High-resolution texturing, topology, animation, and 3D printing tools complement that workflow, but several of those capabilities were already available on the platform before this release.