How Does AI Annotation Work?

Plamen
Plamen
  • Updated
Feature availability 🛈
⭕ Personal ✔️ Suite

AI Studio's Annotation generates labels and descriptions based on the asset's visual context:

  • Images: Metadata is derived directly from the image.
  • 3d Models (e.g., 3ds Max, Cinema 4D, SketchUp): Embedded or Connecter-generated previews provide the visual representation.
  • Custom previews: If the assets have embedded previews, they'll be used directly for annotation. If present, custom previews are treated as separate assets and annotated independently. The advantage is that when you select an asset, you'll see all annotation labels generated for its embedded and custom previews combined. That gives you a comprehensive view of all relevant metadata, even though some of it is derived from the linked custom previews.
  • Preview snapshots: These are snapshots generated internally by Connecter for certain 3d formats using its viewer.
  • Previews generated with Dedicated Services: Temporary previews created via external host applications (e.g., Revit, 3ds Max), discarded after annotation. These are especially useful for formats lacking visual previews or for improving preview quality.
  • OS-generated thumbnails: Previews provided by the operating system, such as PDF thumbnails displayed in Explorer when a PDF reader is installed. These are used when other preview types aren't available.

TIP: Boost annotation accuracy by using custom previews for assets with low-quality or missing embedded previews. Annotating custom previews adds richer, more descriptive metadata displayed alongside the asset's own annotations, providing a more complete and accurate representation.

What annotation metadata is generated?

The AI Studio's annotation tools produce two annotation metadata types:

  • Labels: Short tags (one to two words) identifying recognizable features in an asset's image, such as "sports car." The annotation tool will generate up to ten relevant labels depending on each asset's visual specifics.
  • Descriptions: Concise summaries of the asset's visual content (e.g., "A sleek white sports car parked against a striking brutalist concrete backdrop, bathed in the warm light of the golden hour.").