Skip to content

6. Smart Library

Lucas edited this page May 8, 2026 · 1 revision

Smart Library

Smart Library is a browsing mode for organizing large libraries by similarity and metadata-derived tags.

It is useful when your folder has too many images to curate one by one.

Main features

Smart Library includes:

  • Prompt clustering
  • Stack browsing
  • TF-IDF auto-tags
  • Deduplication helper
  • Cluster cache persistence
  • Similarity-based navigation

Prompt clustering

Prompt clustering groups similar images into stacks based on prompt similarity.

This helps when you have many near-duplicates, variations, or repeated prompt families.

The cluster view shows stack cards with a cover image, prompt preview, and image count.

Stack browsing

Stacks let you drill into a group of related images, browse inside that group, and return to the broader stacked result set.

Stacking can also be used from the normal Library view to group identical prompts.

Auto-tags

Auto-tags are generated from metadata such as:

  • Prompt text
  • Model/checkpoint
  • LoRA names
  • Workflow metadata
  • Other useful metadata terms

Auto-tags are separate from manual tags.

You can use them for discovery, filtering, and promotion into permanent manual tags.

Deduplication helper

The deduplication helper helps identify likely keep/archive candidates.

It can use heuristics such as:

  • Favorites
  • File size
  • Creation date
  • Duplicate-like prompt groups
  • Similarity stack context

It is intended as a helper, not an automatic deletion system.

Free and Pro behavior

Smart Library workflows exist in the core app.

Pro removes clustering scale limits for larger libraries.

Best practices

For large libraries:

  1. Let indexing finish first.
  2. Generate clusters.
  3. Review stacks rather than individual images.
  4. Use ratings and favorites to mark keepers.
  5. Promote useful auto-tags into manual tags.
  6. Use collections for project-level organization.
  7. Use deduplication suggestions only after reviewing the results.

Clone this wiki locally