EyeSayTag and sort a whole photo collection from a folder
Thousands of scans or photos and no catalogue yet? EyeSay tags every photo in a folder with everyday words, gives you the tags as a CSV, JSON or XMP file and sorts the folder into categories you name — by copying, never by moving your originals. 100,000 photos cost about €30; the prices are on one page.
How it works
- Whole folder inConnect EyeSay once to Claude Code or Codex and ask “use EyeSay to tag the photos in ~/Archive/1950s”. One command sends every JPEG, PNG, HEIC and WebP file in the folder and its sub-folders straight to EyeSay, never through the chat, typically 5 to 8 photos a second.
- Tags out as filesEveryday words for what is in each photo; tags the AI is unsure about are marked as suggestions. You get the most frequent tags and a CSV (file name, a fingerprint of the file, sure tags, suggestions) or a JSON file that also has each tag’s score.
- Sort into your own categoriesName them — “portraits, buildings, street scenes, documents” — and each photo is copied into
EyeSay/<category>/in your folder, keeping its sub-folders; a photo close to none of them goes toUnsorted. Originals are never moved, renamed or deleted, and deleting theEyeSayfolder undoes it. - Keywords for your catalogueThe tags can also be written as XMP sidecar files, one per photo next to the original, which Lightroom, Bridge and other tools that read XMP sidecars take as keywords (how). An existing
.xmpfile is never overwritten.
To connect Claude Code: claude mcp add --transport http eyesay https://eyesay.app/mcp, then sign in with /mcp. Codex and the other assistants are in the docs. Running the command again sends only the photos that are missing, and running the sort again skips what is already copied.
Find a photo again
Once a folder is tagged you can search the whole of it by description — “a market stall in the rain”, “group portrait in front of a church” — and get the closest photos first, without sending the folder again.
What it is, and is not
EyeSay writes everyday-word tags for what a photo shows: objects, places and activities, the words someone would type to find it again. It does not write catalogue-standard metadata: there is no controlled vocabulary or catalogue record, and the tags do not replace what a cataloguer knows about a photo.
The AI can be wrong. It reads and answers in English, is weakest on small text in a photo, and gives tags it is unsure of as suggestions. Check what you rely on before it goes into a catalogue or onto a finding aid. Files that are not JPEG, PNG, HEIC or WebP, such as TIFF or RAW scans, are skipped by the folder command: make JPEG copies first.
What 100,000 photos cost
A free account tags 1,000 photos every month. After that, tagging costs €0.30 per 1,000 photos, so 100,000 photos come to about €30 and 10,000 to about €3. Units are bought in packs that are paid once and never expire; the full table, with the free month, the packs and what other services list, is on the pricing page.
A photo sent in a folder job counts once, as one tagged photo, for its tags and search data together; the same photo sent again does not count.
Questions
Does it change or move my original photos?
No. Sorting copies each photo into an EyeSay folder inside the folder you sent; your originals are never moved, renamed or deleted. XMP sidecar files are written only if you ask for them, next to the photo, and never over an existing .xmp file.
How many photos can one folder job take?
One folder job takes up to 10,000 photos. For a larger collection, run the command on one sub-folder at a time, a job for each, and save each job’s CSV or JSON when it finishes.
How long are the results kept?
A folder job keeps each photo’s answer — its tags, search data, a fingerprint of the file and its name — in memory for 24 hours, or less if the server is restarted or updated. The links to the CSV and JSON last 15 minutes, so download them as soon as the job is done.
Are my photos stored or used for training?
No. A photo is shrunk and tagged the moment it arrives and then dropped; the photo itself is never kept. Nothing you send is used to train models or for advertising. The details are in the privacy policy.
Can I rely on the tags for cataloguing?
Treat them as a fast first pass. They are everyday words, not catalogue-standard metadata, and the AI can be wrong, so check them before you rely on them.
How is it priced?
By what is done with each photo, in units of €0.0001: tags or a sort are 3 units, a search vector 1, a description 9. A free account has 3,000 units every month, which is 1,000 tagged photos. Tagging 100,000 photos uses 300,000 units, about €30. Packs of €10, €40 and €150 are paid once and never expire; see pricing.
Photos are not kept. Each photo is looked at in memory and dropped once the answer is sent. Nothing you send is used to train models or for advertising.
A few hundred photos rather than a collection? See EyeSay for photographers. Writing your own script? Use the API.