Face Tools

mere-face-tools turns native mere.run face detection and embeddings into a durable photo-library index and reference-face search. It traverses the library, tracks changed files, stores normalized face records in SQLite, and produces review artifacts. Core mere.run continues to own model download and inference.

Install and check readiness

mere.run plugin install mere-face-tools
mere.run model pull vision-face-buffalo-l
mere-face-tools doctor

doctor checks the local mere.run face-analysis surface and SQLite support. The plugin does not require provider credentials or create paid resources.

Index a photo library

mere-face-tools index \
  --photos /Volumes/Photos \
  --database ./faces.sqlite3 \
  --output-dir ./face-index

Indexing uses the warm-session mere.run vision face batch JSONL interface and stores normalized embeddings, bounding boxes, landmarks, source dimensions, and modification times. It writes run.json before execution and can continue an interrupted scan:

mere-face-tools resume ./face-index/run.json

Unchanged photos are skipped when the library is indexed again.

Search by reference face

mere-face-tools search \
  --database ./faces.sqlite3 \
  --reference ./reference.jpg \
  --output-dir ./searches/reference

Search embeds the reference image, ranks the best matching face per photo, and writes JSON, CSV, and a contact sheet. Symlink-only review folders group results under strong/, likely/, and review/ without copying the originals.

Privacy and source safety

Detection, embeddings, indexing, and search run locally. The plugin never modifies the source photo library. The SQLite database and review artifacts may contain sensitive biometric relationships, so keep the output directory under the same access controls as the original library and remove it deliberately when it is no longer needed.

cleanup records the local cleanup decision but does not delete source photos or the index automatically.