What it does
Scores sharpness, exposure, and composition, detects duplicates, and sorts photos into keep, trash, or culled (undecided). Optional thresholds are 0.0 to 1.0: a photo scoring at or above keep_threshold becomes keep, at or below trash_threshold becomes trash. Leave both unset to use the user's configured defaults. Can take a while on a large library; heavier AI models are skipped automatically when not installed. Returns analysis stats plus per-status library counts. Only pending photos are scored: already-analysed photos keep their stored score, even if the scorer has improved since. Use reanalyze_stale_photos for those. smart_rank (default false, experimental) adds a second pass after the fast cull: an image-understanding model rates which analysed photos are worth keeping and which are snapshots (signs, parked cars, floors), and the whole library is re-ranked from that. It also re-ranks photos analysed earlier, so it can be used on its own with no pending photos. Opt in only when the user wants better top picks over speed: it needs a one-time download of about 600 MB if this machine has never used photo search, and adds roughly two minutes per 1,000 photos. It never re-runs the analysis, never changes a manual keep/trash decision, and if it fails or is unavailable the fast cull's results stay as they were. The reply carries a smart_rank block with the estimate, status and counts. Remote: dispatched as a background job past the call budget, like scan_directory, poll get_job_status with the returned job_id.
Parameters
| Name | Type | Required |
|---|---|---|
keep_threshold |
float | None |
optional |
trash_threshold |
float | None |
optional |
smart_rank |
bool |
optional |
Asking for it
You do not call this by name. You say what you want, and the agent picks the tool:
Cull the wedding shoot in D:/Shoots/eva_tom, keep the best frames, and tell me what you dropped.
The rest of the culling tools
Workflows that use it
- Cull a wedding shoot with AI — AI wedding photo culling
- Find duplicate and near-duplicate photos — find duplicate photos automatically
- Score and rank photos automatically — automatic photo quality scoring
Where it runs
On your machine. imagic's MCP server is a local process that your AI client starts; the photos never leave your disk and nothing is uploaded. That is true of every tool in the reference, including this one.
To call it you need an MCP client connected to imagic. Setup takes about three minutes for Claude Code, Cursor, VS Code, Codex, Gemini CLI and ten others.
Give your AI the whole pipeline
analyze_photos is one of 30 tools in imagic Max. The free 500-export trial includes all of them.