Verifying Images From a Phone in the Field

Field verification has different constraints from desk verification: a small screen, a slow connection, a single device, and often very little time. Most published workflows quietly assume none of that applies.

Get the original, immediately

This matters more on mobile than anywhere else, because mobile messaging destroys files. A photo forwarded through a chat app has typically been re-compressed and stripped of metadata. If a source sends you an image, ask them straight away to also send it as a file or document attachment rather than as an image — most messengers preserve the original that way.

Do this before anything else. You usually get one chance before the source moves on.

Analysis works offline

Because our detector runs entirely in the browser, once the page has loaded it keeps working with no connection. On a patchy network this is a practical advantage rather than a privacy talking point: load the page while you have signal, then analyse as many images as you need without one.

You can also install the page to your home screen and it will behave like an app.

Read the results on a small screen

The layer breakdown is more useful than the ELA map on a phone, simply because a 6-inch display cannot show regional anomalies well. Practical order on mobile:

1. Read the composite score and band.
2. Read the four layer values and look for one that diverges sharply from the others.
3. Only zoom into the ELA map if a layer looks anomalous, and pinch-zoom generously.

Screenshot discipline

The single most common field mistake is screenshotting an image to share it with a colleague for a second opinion. That screenshot destroys the evidence you were trying to evaluate. Share the original file, always — and if you must screenshot for speed, say explicitly in the message that it is a screenshot.

Reverse search on mobile

Long-press an image in most mobile browsers for a reverse-search option. Results are weaker than desktop, so if a story turns on the result, flag it for a desk colleague with a proper browser rather than treating the mobile result as final.

Note-taking that survives

Record, in a single note, at minimum:

• Who sent the file and through which channel
• Whether you obtained the original or a forward
• Time received
• Composite score and the four layer values
• Anything visually anomalous

Two lines written in the field beat a perfect reconstruction attempted three days later.

What not to do in the field

• Do not publish a verdict from a single forensic run on a forwarded file.
• Do not treat a high score as verification — provenance still matters more.
• Do not delete the original after analysing it. It is the evidence.