Where was this taken? A ranked shortlist, from the scene itself.
Morrigan estimates likely capture locations for images and video — indoors as well as outdoors — and ranks them with confidence scores. It narrows a search; an analyst confirms it.
The question behind much of the image work in a case
Who is the child, where is the room, which country holds jurisdiction, is this footage from where the uploader says it is — a large share of investigative and intelligence work on imagery comes back to where. The usual tools wait for the obvious: a landmark, a street sign, a skyline. Most case imagery offers none of them. It is a bedroom, a hotel corridor, a stretch of road, a view out of a window.
Morrigan reads the whole frame instead — architecture and building materials, signage and script, vegetation, terrain, the light, the layout and furnishings of a room, what people are wearing — and reasons from the combination rather than a single clue.
We had a case here where we were able to locate a picture of a person sitting in a restaurant, based on the view out of the window.
— Senior Officer, European Law Enforcement Agency
Who uses it, and for what
| Team | Typical question | What Morrigan contributes |
|---|---|---|
| Victim identification | Which country, and which kind of place, is this room in? | A ranked set of candidate regions to narrow the search and route the referral to the right jurisdiction |
| OSINT and digital investigations | Is this image from where the post claims? | Independent candidate locations to test a claim against, before anyone spends time on manual verification |
| Intelligence and defence analysis | Where was this footage filmed? | Candidate locations from several frames read together, to prioritise which areas an analyst checks first |
| Cross-border casework | Whose case is it? | An early, documented basis for a jurisdictional assessment — to be confirmed, not relied on alone |
What comes back
- Candidate locations, ranked, with confidence scores — several of them, not one pin. A tight cluster is a strong lead; a spread across regions tells you the image does not carry enough signal, which is worth knowing too.
- Indoor and outdoor scenes. Interiors are where most landmark-based methods have nothing to work with.
- Joint estimation across several images. Select photos you believe come from the same place and Morrigan estimates them together — materially more accurate than reading any one of them alone.
- Video, frame by frame, with automatic frame selection, and a crop editor to point the analysis at the one detail that matters in a wide shot.
Results are estimates, not findings. A geolocation is a lead for an analyst to confirm against other sources. It does not establish where an image was taken, and a confident-looking cluster can still be wrong.
Where it runs
- Inside VST Teams, across every image in a case, with the strongest geographic leads surfaced first and each result linked to its source file.
- As a standalone REST API for teams with their own tooling.
- In Morrigan’s investigator UI in the browser, for analysts who want to work image by image without an integration. The interface is in English.
- On-premise and air-gapped as a container, with no external connectivity required and outputs under your control.
Capabilities and deployment modes available to an organisation depend on who it is. Tell us about your team and we will say what applies to you.
Try it before you talk to us
- The public demo on the image geolocation page reads one still image at a time, three runs a day. Images submitted there are kept for up to 7 days for trust and safety review, then deleted — do not use it for case material.
- From the demo you can request a trial key for the full investigator tool — joint estimation, video and cropping, on a private endpoint with nothing logged publicly. Any email address works.
- For a pilot on your own material, on your own infrastructure, request a briefing.