Automatic Photo Keywords: How to Generate Them at Scale (2026)

Your photo library has 12,000 images and maybe 300 of them have keywords. Typing the rest by hand, at the 100 to 200 photos per hour a skilled keyworder manages, is three to four weeks of full-time work, which is why the job never gets finished.

Quick answer: Automatic photo keywords come from a vision model that looks at each image and writes its own keyword list, no typing involved. The fast tools generate keywords for roughly 1,000 photos every 8 minutes, against 5 to 8 hours for a human at the same depth. You have three ways to get them: free and DIY options, buy-once desktop keyworders like ON1 or Excire, and connected catalogs like Tagrly that keyword photos in place in Google Drive or Dropbox. The decision that matters most is not the tool. It is whether the keywords should live inside your image files or in a searchable index.

What automatic photo keywords actually are

An automatic photo keyword is a descriptive word or phrase, "golden retriever", "sparkler exit", "linen blazer", generated by software that read the image's pixels. A vision model identifies what is in the frame and emits a keyword list, the same way a human keyworder would, just 30 to 60 times faster. For the full speed-and-cost math against a human, see AI photo tagging vs. manual keywording.

Three things that are not automatic keywording, because tool marketing loves to blur the line:

• Filename and folder matching. Search that only reads IMG_4827.JPG and the folder name has not keyworded anything.
• EXIF data. Camera model, capture date, GPS. Useful, but it says nothing about what the photo shows.
• Shallow category detection. "Person", "food", "outdoor" stamped on every image. Technically keywords, practically useless past a few hundred photos.

Photographers usually call this job keywording, because the words end up in Lightroom or a stock agency uploader. Newer tools call it tagging. Same job, different home for the output, and that difference is the first real decision. For the broader survey of the whole category, see the complete guide to AI photo tagging in bulk.

Decide where the keywords should live first

Before comparing any tools, answer one question: where do the keywords need to end up? There are exactly two homes, and most tools only write to one of them.

Embedded in the file. Desktop keyworders write keywords into the image's IPTC or XMP metadata fields, the standard maintained by the IPTC photo metadata working group. Embedded keywords travel with the file: Lightroom reads them, Bridge reads them, and stock agency uploaders read them on submission.

In a searchable index. Catalog tools store keywords in a database that points at the file instead of writing into it. Your originals stay untouched, and everyone on the team searches the same index from a browser instead of one person's desktop app.

Choose embedded if you submit to stock agencies or work alone in Lightroom. Choose an index if a team needs to search a shared Drive or Dropbox library. Studios that need both usually index first, then export a CSV for the subset that gets submitted.

Three ways to generate automatic photo keywords

Free and DIY

Google Photos and Google Drive both run shallow category detection you already have. It finds "beach" and "dog", it cannot find "sand linen blazer, on-model", and neither product lets you export its labels as keywords. Free desktop options like digiKam add basic auto-tagging with more setup. If you can write code, you can send images straight to a vision API like Claude's and store the results yourself; that is exactly the work the paid catalogs automate.

Free covers a one-off job under about 1,000 photos. Past that you hit the 1,000-photo wall, the point where a library becomes too big to navigate by memory, and shallow categories stop helping.

Desktop keyworders and Lightroom plugins

ON1 Photo Keyword AI and Excire Foto are buy-once desktop apps that generate keywords locally and write them into IPTC or XMP (pricing varies by version, check their current pages). They are the right call for a solo photographer whose library already lives in Lightroom, and they produce embedded keywords stock uploaders can read. Their limit is the machine they run on: the keywords sit in one catalog on one computer, which is fine for one person and a dead end for a team.

Connected catalogs

Connected catalogs skip the download entirely. They read photos from Google Drive or Dropbox through a read-only connection, run every image through a vision model, and store the keywords in one searchable index the whole team shares. Tagrly is one example in this tier; it uses Google's drive.readonly scope and never gets write access to your files.

Speed is the reason this tier exists. On a working production archive of roughly 19,000 wedding and event photos, the first full scan finished in about 9 hours overnight, roughly 8 minutes per 1,000 photos. Nobody typed anything.

Tip. If you want to see what automatic keywords look like on your own photos, Tagrly's free tier keywords the first 100 photos in any Drive or Dropbox folder, no credit card. Test it on a sample folder and read the output before deciding anything.

Put the focal subject first, or the keywords won't help

The failure mode of automatic keywording is not missing keywords, it is generic ones. A library where every image is keyworded "person, event, indoor" is just as unsearchable as one with no keywords at all.

The fix is focal-subject tagging: the first keyword names the single dominant element of the photo ("sleeping golden retriever"), context keywords come after ("navy couch", "brass floor lamp"), and search ranks the focal keyword above the rest. That ordering is why a query for "golden retriever" returns photos of the dog before every living-room shot that happens to contain one in the background.

The five-photo test works on any tool in any tier: run five of your own photos through it and look at the first keyword. If the first keyword is the focal subject, the tool understands your photos. If it is "person" or "indoor", keep shopping. Tagrly publishes the same 30 photos run through both of its output tiers on the output quality page, which is the level of transparency worth asking every vendor for.

One caution for stock shooters: agencies cap keyword counts, most sit around 50 per image (check your agency's current rules), so more keywords is not better. Fifty focal-first keywords beat two hundred generic ones.

Which option should you pick?

The honest matchups, by situation:

• Pick free or DIY if you have under 1,000 photos or a one-time job, and generic categories are good enough.
• Pick ON1 or Excire if you are solo, live in Lightroom, and need keywords embedded in the files, especially for stock submission.
• Pick a single-image generator like PhotoTag.ai if you keyword small batches for stock uploads and don't need a searchable library. Our photo metadata generators head-to-head covers that matchup in detail.
• Pick a connected catalog like Tagrly if your photos live in Google Drive or Dropbox and more than one person needs to find them.

Whichever tier you land in, the sequence is the same: decide where the keywords need to live, run the five-photo test, then let the tool do the typing. The math has not been close for a while, 8 minutes per 1,000 photos against 5 to 8 hours by hand. Automatic photo keywords are the base layer; you edit the 5 percent that need your private vocabulary, and the library finally searches like it should. If your photos are already in Drive or Dropbox, the first 100 are free to keyword, so the test costs you nothing but ten minutes.