AI metadata from the image
A vision model reads each image and writes a title, a description and an ordered keyword list. Titles stay within 70 characters, descriptions within 150, and keywords within 20–49 entries (20–40 is the recommended range).
AI-assisted titles, descriptions & keywords for Adobe Stock and Shutterstock. Your API keys never leave the browser.
Everything the generator does, in one page. Metadata Studio turns a folder of stock images into editable titles, descriptions, keywords and Shutterstock categories — then exports a CSV your marketplace accepts.
A vision model reads each image and writes a title, a description and an ordered keyword list. Titles stay within 70 characters, descriptions within 150, and keywords within 20–49 entries (20–40 is the recommended range).
Queue a whole folder, run two or three images at a time, stop the batch, cancel a single item, regenerate one result, or retry only the failures.
Gemini, OpenRouter and Groq are tried top to bottom. Add more than one key per provider and the queue rotates through them, cooling a key down after quota or transient errors.
Refresh the curated model list against your own key. Vision-capable models are labelled so you know which ones can actually read an image.
On the Shutterstock tab, each image is classified against the 26 official categories by a dedicated vision call, and every category stays editable.
Every field is editable inline, and each result shows the provider and model that produced it, so you can fix a weak keyword list without regenerating the whole batch.
Download an Adobe Stock CSV (Filename,Title,Keywords,Category,Releases) or a
Shutterstock CSV, up to 2,000 rows per export.
Your API keys live in this browser's localStorage and are forwarded per request through a stateless relay. They are never stored on the server and never written to logs.
One request produces four things: a title, a description, an ordered keyword list, and — for Shutterstock — a set of categories. The model is instructed to describe what is visibly in the image and to avoid hype words and invented locations. Output is cleaned on the way back: smart quotes are normalised, banned words are removed, singular and plural duplicates are merged, and each field is capped to the length you chose in the sidebar.
If a model rejects an image, the relay retries that item once in text-only mode using the filename. Those items are flagged text-only so you can spot them and try a vision model instead.
This matters when a provider rate-limits you mid-batch: failed items stay visible with their error reason instead of silently disappearing, so a retry is a single click.
The topbar toggle switches between Adobe Stock and Shutterstock, and the export button follows the active tab.
| Export | Columns | Filename |
|---|---|---|
| Adobe Stock | Filename, Title, Keywords, Category, Releases | Adobe_Stock_CSV_<DD-MM-YYYY>.csv |
| Shutterstock | Filename, Description, Keywords, Categories, Editorial, Mature Content, Illustration | Shutterstock_CSV_<DD-MM-YYYY>.csv |
A Shutterstock export needs at least one valid category per row. If a row has none, the export stops and names the offending files instead of producing a CSV the marketplace will reject. See templates and column references for the exact formats.
Metadata Studio is bring-your-own-key. Keys are stored only in your browser, sent with each request through a local relay, and redacted from logs and error messages. Uploads are capped at 25 MB and validated by MIME type and magic bytes. The app ships a strict Content Security Policy, request validation, rate limiting, and an optional auth token for public deployments.
Metadata Studio generates metadata for raster images only — JPG, JPEG, PNG, WEBP and GIF. It is not a text-to-image tool, it does not accept PDF, EPS, AI or SVG files, and it only exports Adobe Stock and Shutterstock CSVs. There is no server-side history or database: clear your browser storage and the keys are gone.
No account, no upload quota from us. Add a provider key and generate your first batch.