Image Curation System

Image curation and review system for managing large generated image libraries at scale. Review workflow: keep, reject, or trash individual images with character assignment, quality scoring, and metadata tagging. Multi-character assignment mode supports batch operations — assign a character identity to hundreds of images in a single operation with per-character result counts.

Face-based search using combined embedding similarity — query by image to find every instance of a given identity across the full library regardless of art style, pose, lighting, or rendering approach. Vector similarity search powered by concatenated face recognition and vision encoder embeddings stored in PostgreSQL with pgvector indexing. Text-based semantic search finds images by description content.

Grid view interface with filtering by character, quality score, category, model used, LoRA applied, NSFW status, and free-text search. Pagination for libraries with tens of thousands of images. Trash system with soft delete, restore capability, and configurable auto-purge. Bulk operations for efficient review sessions — select ranges, apply actions to filtered sets.

Serves images from multiple nodes via network filesystem mounts — the curation interface presents a unified library regardless of which machine generated the image. Designed for the workflow gap between generation and use: when you're producing thousands of images and need efficient tooling to sort, classify, and retrieve them rather than scrolling through filesystem directories.