AI art detector for contests: how to judge submissions fairly

Who needs to detect AI art submissions?

Anyone who rewards, sells or commissions art on the basis of how it was made needs a way to check. Contest organizers, juries, photography awards, art fairs, galleries, online art communities and clients commissioning illustrations all meet the question “is this art AI?”.

• Open-call competitions with prize money and hundreds of digital entries.
• Photography awards, where a generated image can pass as a captured one.
• Art communities that reserve some galleries or tags for human-made work.
• Commissions, where a client pays for an artist’s own hand and style.
• Artists who need to show that their own work is not AI after being accused.

The question is rarely whether any software touched a piece. Digital artists have used tools for decades, and many now include AI features. What matters is whether the entry meets the rules the contest set and the expectations the client agreed to.

Why did AI art become a contest problem?

Generated images started winning before rules caught up. In 2022, Jason Allen’s “Théâtre D’opéra Spatial”, made with Midjourney, won first place in the emerging artist digital arts category at the Colorado State Fair’s fine arts competition. In 2023, Boris Eldagsen declined the Creative Open award at the Sony World Photography Awards after revealing that his winning image was made with AI.

Both cases turned on rules that did not say clearly whether AI-made work was allowed. The lesson for organizers is to write that rule down before the entries arrive, not after a winner is announced. A clear rule also protects the jury: it can point to the published text instead of defending a judgment call under public pressure.

What should AI generated art contest rules say?

Good contest rules define AI use precisely, say how it will be checked and give entrants a fair way to respond. These clauses cover most competitions.

• A definition: separate generative AI (images made from prompts) from AI-assisted tools such as noise reduction, upscaling or content-aware fill, and say which are allowed.
• Categories: if you accept AI work, give it its own category with its own judging criteria.
• Disclosure: entrants state which tools they used and for what.
• Process evidence: finalists may be asked for sketches, work-in-progress files, layered source files, time-lapses or RAW camera files.
• Verification: the jury may use detection tools to decide which entries to ask about, and a tool result alone never disqualifies an entry.
• Due process: questions are raised privately, the entrant can respond, and decisions can be appealed.

What goes wrong when juries miss AI art, or accuse real artists?

Missing a generated entry is unfair to every artist who spent weeks on their work, and it damages the contest’s reputation once the truth comes out. Sponsors and future entrants notice.

Accusing a human artist is worse for the person involved. Clean digital painting, smooth rendering and polished photo editing can all look generated to a classifier and to a hurried viewer. A public accusation can follow an artist across platforms long after it is withdrawn.

How to check AI art submissions fairly

A fair process checks every finalist the same way and reserves questions for entries where several signals line up. Here is a sequence juries can follow.

1. 01Publish the rules above with the call for entries, and ask for full-resolution original files rather than social media exports.
2. 02Read provenance first. A content credentials checker reports C2PA data that may name the tool used. Missing credentials prove nothing: most real art carries none.
3. 03Inspect the work closely at full size, looking for the signs listed below.
4. 04Run each finalist through the AI art detector and read the listed signals, not only the verdict.
5. 05Where signals line up, ask the artist privately for process evidence: sketches, layers, a time-lapse or RAW files.
6. 06Let the jury decide on all the evidence, record the reasons and offer an appeal.

What does AnyFact’s AI art detector for contests return?

The photo detector handles art and photography entries. It accepts JPG, PNG, WebP and HEIC files up to 20 MB, or a link. It checks content credentials and EXIF/XMP metadata first, then scores the image with an open-source AI-image classifier, then lists visible anomalies such as hands, lettering, shadows, reflections and textures.

You get a verdict of probably human, uncertain or probably AI, with the signals behind it, a plain-language explanation, the limits of the method and a PDF report for the jury file. If provenance names a known AI generator, the verdict is probably AI and the reason is stated. A free account gets 10 credits a month with no credit card, and each image costs 1 credit.

Limits of AI art detection

Art is a hard case for detectors. Hybrid works, where an artist paints over a generated base or uses AI for one element, can come back uncertain or mixed in their signals. Photos of physical paintings say more about the camera than the brushwork. Highly stylized or flat digital art gives a classifier little texture to read.

That is why the detector should decide which entries to ask about, and process evidence should decide the outcome. For more on error rates, read how accurate AI detectors are.

Social media exports are another weak spot. Platforms resize images and strip metadata, including content credentials, so always ask entrants for the full-resolution file they made rather than checking a copy saved from a feed.

Frequently asked questions