Apple LiTo rebuilds 3D from one image

This is a meaningful step up for image-to-3D: Apple is treating appearance as part of the 3D representation, not just a texture layer slapped on afterward.

• –LiTo models geometry and view-dependent appearance in one latent space, which is how it keeps specular highlights and Fresnel reflections believable
• –Single-image conditioning makes the workflow much easier, but it also makes the reconstruction problem much harder than multi-view approaches
• –The comparison against TRELLIS suggests the battleground is now physical consistency, not just whether the object vaguely resembles the source image
• –If Apple turns this into a product feature, it could be highly relevant for AR, spatial computing, and asset creation workflows
• –For now, it reads more like a strong research milestone than a shipping product feature