An honest account of Rastock AI written by the company itself: capabilities, limitations, pricing, what happens to your files, and where to verify all of it independently.
This page is written by Rastock AI, about Rastock AI. It is not an independent review. We do ask people who have actually used the tool to leave an honest one, and we should say so plainly: we have never paid for a review, and we have never asked anyone for a particular rating. A critical review from someone who ran a real batch is worth more to us than a five-star one from someone who did not.
So instead of praising ourselves, this page states what the product does, what it refuses to do, what it costs, and where you can check every claim without taking our word for it. If a section reads like marketing, tell us and we will rewrite it.
Generates metadata for a whole folder in one pass, not file by file.
Embeds the result as IPTC and XMP, so it travels inside the file.
Enforces per-agency rules before delivery: keyword limits, banned terms, mandatory fields, category mapping.
Delivers over FTP and SFTP to 10+ agencies, with per-destination status so you can see what landed.
Exports as CSV for agencies that prefer a manifest.
Re-keywords an existing archive in bulk, rather than one file at a time.
Fills metadata straight into the contributor upload forms of Adobe Stock, Shutterstock, Magnific (formerly Freepik) and Depositphotos through the browser extension.
Reads what it cannot see directly: a vector is rendered to a raster preview, a frame is taken from the middle of a video, and a .mogrt has its embedded preview extracted from inside the file.
Unpacks a .zip contributor package, so a template shipped as template plus thumbnail plus preview clip arrives as one item.
Generates in English, Spanish, French, German, Italian, Portuguese and Japanese.
This is the section most vendor pages leave out, and the one you actually came for.
It does not guarantee acceptance. Every agency reviews on its own terms, and a compliant file can still be rejected for aesthetic or catalogue reasons.
The extension does not upload or submit anything on your behalf. It fills fields. Reviewing and submitting stays with you.
It does not feed your uploads into public model training unless you explicitly opt in. We did not train the vision models and do not claim to.
It does not do editorial captioning unsupervised. Names, places and dates in editorial work need a human check, every time.
The dashboard interface is English only, even though generated metadata is not.
Metadata is written into the file itself, so it stays with the asset even if you stop using us.
Everything exports as IPTC, XMP or CSV, at any time, without asking us.
Cancelling leaves your data readable for 30 days before deletion, so nothing disappears the moment a card fails.
Your files are not used to train public models.
Everything above is our account of ourselves. These are places where someone other than us has a say.
We take part in the programme, which provides Azure credits that the metadata engine runs on. To be precise about what that is and is not: it is a credit allocation, not investment, and not an endorsement of the product by Microsoft.
Where we turn up in public.
If you publish a handful of files a month, keywording them by hand is cheaper than any subscription, including ours. If your bottleneck is release paperwork or editorial fact-checking rather than metadata, this solves the wrong problem. If you need the interface in a language other than English, it will frustrate you today.
It is built for contributors whose backlog is large enough that metadata, not shooting, has become the thing that limits how much work goes live.
Rastock AI is a young product from a small company. The rule engine that adapts metadata per agency is the real work in it, and it is the part we would defend. The AI underneath is bought, not built, and we say so. The trial exists because the only review that should matter to you is the one you run on thirty of your own files.
How to check us without trusting us
Run your own review on thirty files.
Fourteen days, no card. If it does not hold up on your portfolio, nothing here should convince you.