GLOSS: Geometric Local Self-Similarity Learning for Faithful Reference-Guided Texture Fill

GLOSS, short for Geometric Local Self-Similarity Learning for Faithful Reference-Guided Texture Fill, takes a more local approach to texture generation than the large-dataset, whole-object systems that dominate current research. Instead of asking a model to invent an entire material from broad guidance, it learns from one target shape and a set of geometry-aware reference patches, then fills the surface through patchwise inpainting. The goal is to better respect small-scale structure and the way geometry and texture tend to correlate on real assets.

That matters because texture artists still run into a familiar problem: strong generative tools can produce plausible results, but they often drift away from the reference when the mesh has tricky detail or when the artist wants to steer the result from multiple sources. GLOSS is designed to be more controllable and less data-hungry, which should make it easier to adapt to a specific asset instead of relying on a generic model trained on huge 3D corpora.

The system also supports local geometry-conditioned inpainting, can transfer novel references across the full object, and reportedly generalizes to PBR materials and unseen meshes. In comparisons, it matched or improved on strong image-conditioned texture-generation baselines, which is a useful signal for teams looking at AI-assisted texturing without giving up art direction.

The most practical detail for production teams is the Blender addon pilot. Several 3D texturing professionals gave positive feedback on its controllability, usefulness, and creative flexibility, which...