Gallery Overview

The gallery system manages images and videos downloaded from CivitAI for LoRA model galleries. It provides local storage, indexing, and serving of media files so the LoRA manager can display example images for each model.

Design Principles

• Two storage areas -- original/card images belong to a specific model and are grouped by model_id. Community images are numerous and cross-version, so they are sharded for scale.
• Co-located files -- for each media item, the media file, thumbnail, and metadata JSON are stored in the same directory
• SQLite index -- a single gallery.db database indexes all files. The DB is rebuildable from files on disk.
• Cross-endpoint dedup -- the same CivitAI image can have different IDs in the REST vs tRPC API, but the CDN URL UUID is always unique. The file_uuid field prevents duplicates.
• Thread safety -- WAL mode + busy_timeout, each thread gets its own connection via threading.local()

Components

Component	Purpose
gallery_db.py	SQLite database operations: CRUD, queries, tag caching
loras_api.py	REST endpoints for gallery download, serve, query, star, to-input
assets/images/	File storage root
gallery.db	SQLite database file