Bottom Line Up Front (BLUF): Smart segmentation reads meaning, not connectivity: it understands that a hero and his sword share the frame and separates the two.
From a photo (PNG/JPEG/WebP): you get a mask plus auto-built 3D models of the selected parts. From a GLB model: the mesh is split respecting a 16-number transform matrix.
A text hint (subject, configuration) helps the model find the right parts.
The output includes the mask, the segmented model (GLB) and ids of the related tasks — all visible in your account.
Plain segmentation splits the whole model; smart segmentation can isolate one or two parts and build separate models of them.
The mask is the exact boundary of the selected part in the source image — it helps re-composite the frame with new textures.