IPTC keyword generator
Generate the keywords and description for every photo in a shoot, review them, and export a file you can push straight into Lightroom, Bridge or your DAM. Nothing is uploaded.
No account. No upload. Runs in your browser.
The metadata nobody fills in
IPTC metadata is the set of fields that travel inside a photo file — the keywords, the description, the creator, the copyright, the location. It is what makes a photo findable in a stock library, attributable in a newsroom, and legally safe to publish.
It is also, on most drives, almost entirely empty. Not because photographers do not know it matters, but because filling it in for six hundred frames from a single shoot is an evening of typing that nobody has ever looked forward to.
The economics are brutal: the keywording takes longer than the edit, it earns nothing directly, and its absence only hurts you later, quietly, when a picture editor searches and your image does not come back.
What PicsTag generates
For each image, two things that map cleanly onto IPTC fields:
- Keywords (IPTC
Keywords) — a set of proposed terms, each with a confidence score, describing what is recognisably in the frame. - A description (IPTC
Description/Caption-Abstract) — one factual sentence about the image.
You review both, accept what is right, add your own terms for what the model cannot know, and export a CSV or JSON keyed on your own asset IDs.
Getting it into the files
Being straight about the boundary: PicsTag does not write into your files. It produces the metadata; you write it. That is a deliberate split — a browser page cannot safely rewrite thousands of files on your disk, and you almost certainly want the write step to run through the tool you already trust.
The usual routes, once you have the export:
- Lightroom Classic — its metadata import expects a CSV keyed on filename; several plugins handle the mapping.
- ExifTool — the command-line standard. It reads a CSV directly and writes IPTC/XMP into a whole folder in one pass.
- Your DAM — most accept a CSV or JSON bulk metadata import keyed on asset ID.
The generation is the part that used to cost you an evening. The write is a single command.
Keywords that get found
A keyword set is not a description of the photo, it is a prediction of the search that will one day need to find it. Those are different things, and confusing them is why so many keyword fields are full of words nobody types.
An AI pass gives you the literal layer — what is visibly in the frame — quickly and consistently. What it cannot give you is the conceptual layer (what the image is about), the practical layer (copy space, crop, mood) or the provenance layer (shoot, client, rights). Those come from you, and they are where the value is.
The full method is in keywording a photo library people can actually search, and the field-by-field details are inIPTC and XMP metadata, explained.
Why the browser matters here
Photographers are the group with the strongest reason to refuse an upload: embargoed editorial work, client shoots under NDA, unpublished commercial campaigns. Sending those frames to a third-party AI service is often not a preference — it is a contract violation.
PicsTag runs the models in your browser. The files never leave your machine, because there is no server for them to go to.
Frequently asked questions
Does PicsTag write IPTC keywords into my image files?
Not yet — and we would rather say so plainly than pretend. PicsTag generates the keywords and captions and exports them to CSV or JSON, which you then import into Lightroom, Bridge, ExifTool or your DAM to write into the files. The generation is the slow part; the writing is a one-line command.
What is the difference between IPTC and XMP?
IPTC is the standard that defines the fields (Keywords, Description, Creator, Copyright…). XMP is the modern container that stores them inside the file. In practice you almost always write IPTC fields in XMP form, and every serious photo tool reads both. See the guide for the parts that actually matter day to day.
How many keywords should an image have?
Enough to be found, not so many that they stop meaning anything. Somewhere between 8 and 25 is typical for a stock or editorial workflow. Beyond that you are usually adding synonyms nobody searches for, and some agencies will penalise obvious keyword padding.
Can it match my controlled vocabulary?
Not automatically. The AI proposes terms from a general-purpose vocabulary — it does not know your taxonomy. Treat its output as candidates, add your own controlled terms in the custom tag field, and the export keeps both. Any tool claiming to auto-populate a controlled vocabulary correctly is overselling.
Do my photos get uploaded?
No. The models run inside your browser. For photographers under embargo, agencies with client NDAs, or anyone handling unpublished work, that is usually the deciding factor.