Sketchfab updates exporter with on-device model inference for asset tagging

Tech · 3 min read

The exporter runs a small local model to infer semantic tags — for example, 'office chair,' 'ceramic vase,' or 'PBR metal' — before upload, enabling creators to opt into richer metadata without sending raw assets to cloud inference services. The tags can be edited prior to publishing.

Sketchfab noted this improves search relevance and helps marketplaces route assets to appropriate category feeds. Studios can also attach custom taxonomy rules that the exporter respects when generating tags.

Privacy-conscious teams welcomed the change since asset previews and tag generation happen locally; Sketchfab provides enterprise controls to toggle on-device inference and to enforce editorial workflows for published metadata.

Creators reported faster publishing flows and better initial discoverability. Sketchfab said future updates will expand language localization for tags and provide confidence scores so sellers can review automated labels efficiently.