Tag and caption your images with AI — without uploading them
PicsTag reads your photos, suggests keyword tags and writes a caption for each one. The AI runs inside your browser, so your images never leave your machine. Review the suggestions, fix what's wrong, export to CSV or JSON.
No account. No upload. No server.
Untagged images are invisible images
A photo library without keywords is a folder you scroll through. The asset you need is in there somewhere, and finding it costs more than shooting a new one. This is the quiet failure of most digital asset libraries: the search box works fine, but nobody ever filled in the metadata for it to search.
The reason is boring and human. Tagging is slow, repetitive work that nobody volunteers for, and it has to happen for every single image before it pays off even once. So it gets postponed, and the library rots.
PicsTag does the first pass for you. It looks at each image, proposes the keywords it recognises with a confidence score, and writes a plain-language caption. What used to be an hour of typing becomes a few minutes of reviewing — accepting what's right, rejecting what isn't. The judgement stays yours; only the typing is automated.
How it works
- 1
Add your images
Drag and drop files from your disk, or upload a CSV of image URLs to pre-index a catalogue you already host somewhere.
- 2
The AI runs on your machine
Two open models — an image classifier and a captioning model — are downloaded once (about 250 MB, then cached) and run in your browser through WebGPU, falling back to WebAssembly. Nothing is sent anywhere. You press Start; you can press Stop at any time.
- 3
Review, one image at a time
Each suggested tag carries a confidence score. Accept it, reject it, or add your own. Bulk actions are there when you trust the model — auto-accept everything above 80% confidence and only look at the rest.
- 4
Export
CSV or JSON, ready to import into your DAM, your CMS, or a spreadsheet. Only the tags you accepted make it into the file.
Your images never leave your browser
Every other AI tagging tool works the same way: you upload your images to a server, a model runs there, and you get keywords back. That is fine for holiday snaps. It is a problem for unreleased product shots, client work under NDA, medical or legal imagery, or anything covered by a data processing agreement you would rather not have to amend.
PicsTag runs the model in the same place the image already is: your computer. There is no upload step because there is no server to upload to — the whole tool is a static page and the inference happens in a Web Worker in your tab. You do not have to take our word for it: open the Network tab in your browser's developer tools while you tag a batch. You will see the model files download on first use, and then nothing.
If you want a cloud model — they are more capable, and sometimes that's the right trade-off — you can add your own OpenRouter API key in the settings. It stays in your browser's local storage, the request goes straight from your browser to them, and we are not in the middle.
Who it's for
Photographers & image libraries
Keyword a shoot in minutes instead of an evening. Export and feed your catalogue.
SEO & accessibility
Generate alt text for a whole page of images at once, then edit the ones that matter.
E-commerce
Feed a CSV of product image URLs, get back keywords and descriptions for your catalogue.
DAM teams
Pre-index a backlog before it enters the DAM, without shipping assets to a third party.
What people use it for
Generate alt text →
Write the accessibility descriptions you were never going to write by hand — for a page, or for a whole site via CSV.
Caption images in bulk →
One factual sentence per image, for a catalogue, a CMS field or an image database.
Auto-tag a photo library →
Keyword a shoot or clear a backlog: accept the confident tags in one click, review only the rest.
Also: IPTC keywords for photographers, and bulk tagging from a CSV of URLs. New to image metadata? Start with ourguides.
Frequently asked questions
Are my images uploaded anywhere?
No. With the default local model, the AI runs inside your browser tab using WebGPU or WebAssembly. Your images are read from disk into memory and never sent over the network. You can verify it yourself: open your browser devtools, go to the Network tab, and watch — the only downloads are the model files themselves, on first use.
What does it actually generate?
Two things per image: a set of keyword tags with a confidence score each, and a one-sentence caption you can use as an alt text or a description. You review both before anything is exported — accept the tags that are right, reject the rest, add your own.
Is it free?
Yes. The local model costs nothing to run because it runs on your machine, not on our servers. If you would rather use a more capable cloud model, you can plug in your own OpenRouter API key in the settings — you pay that provider directly, we never see the key or the bill.
Why is the first run slow?
The AI models (about 250 MB) are downloaded once, then cached by your browser. The first image takes a while; every image after that is fast, and your next visit skips the download entirely.
Can I tag images that live on a URL rather than on my disk?
Yes. Upload a CSV with an image_url column and an asset_id column, and PicsTag will fetch and process each row. That is the fastest way to pre-index an existing catalogue.
What formats can I export?
CSV or JSON, with asset_id, filename, file size, dimensions, description, validated tags and source URL. Only the tags you explicitly accepted are exported — the AI never puts a keyword in your catalogue behind your back.
Tag your first batch now
Nothing to install, nothing to sign up for.
Open PicsTag