EntryWorks
PlatformIdeapadSecurityIntegrationsWorkGuidesAboutStart a conversation

Guide · Integrity

Keeping AI-generated images out of your photo competition

Generated images have already won real photography prizes. Here's the layered screening that catches them — and why the final call should always belong to a person.

6 min readUpdated 25 August 2026

Why this is now every organiser’s problem

Image generators have crossed the line where a casual look can tell. Competitions have already awarded prizes to generated images — sometimes discovering it only when the “photographer” confessed. For a sponsored competition the stakes are commercial: a sponsor’s name sits on that winners’ page, and a “prize went to an AI fake” story undoes everything the campaign built.

The good news: you don’t need to solve an arms race. You need a screening process that catches the overwhelming majority cheaply, documents its reasoning, and puts a person in front of anything doubtful.

Layer one: provenance credentials

The strongest signal, when it’s present. Under the C2PA / Content Credentials standard, generators increasingly sign their output with a manifest that says “made with AI” — and newer cameras sign “captured live”. Reading those credentials on the way in gives you a verdict that isn’t a guess: the file itself declares its origin.

Layer two: metadata forensics

Real photographs carry a camera’s fingerprints — make, model, lens, exposure, capture time. Generated images usually carry none of that, or worse for the lazy forger, they carry a generator’s own markers: tool names and prompt fragments left in the file. A metadata pass is essentially free, and it catches the careless fakes before anything expensive runs.

Layer three: trained detectors

For files that pass the first two layers, trained classifiers score how likely an image is to be generated, alongside content-safety checks in the same pass. Detectors are probabilistic — that’s precisely why they should be one voice among three, run cheapest first, and never left to reject anything on their own.

Put it in the rules — and keep the receipts

Screening works best when the terms back it up. State plainly that AI-generated and AI-composited images are ineligible, that entries may be screened by automated tools, and that the organiser’s decision is final. Then keep an audit trail: what was checked, what it concluded, and why. When a disqualified entrant disputes the call in October, you answer with a record from August.

What this looks like when it’s built in

On EntryWorks, all three layers run in the seconds after upload, the verdict rides with the entry into the moderation queue, and every automated assessment lands verbatim in its own audit log. Across a recent three-thousand-entry photo competition, that meant a winners’ page a sponsor could put their name on without a second thought.