Why the obvious rule does not work
Every competition now has a line about artificial intelligence, and most of them are unusable. The trouble is that generative technology is already inside ordinary photographic tools, doing things photographers have done by hand for a century.
Noise reduction is a trained model. Sharpening is a trained model. Automatic masking, sky selection, subject detection, lens correction and upscaling are all machine learning, and all of them ship inside mainstream editing software with no label.
So a photographer who removed sensor dust with a healing brush has used a generative tool, and a rule saying no AI has technically disqualified them. Nobody intends that, which means the rule is doing no work and the judges are improvising.
The competitions that have handled this well stopped writing about technology and started writing about what may be changed. That is a rule about the picture rather than about the software, and it survives the next release of everything.
The rule that works
Require the raw file, or the original camera file where raw is not available, submitted at entry rather than requested after a shortlist. That single requirement resolves most of what a detector would otherwise be asked to guess at.
| Enforceable | Survives new tools | Cost to entrants | |
|---|---|---|---|
| No AI permitted | No Undefined in practice | No | Partly Confusing |
| No generative fill or removal | Partly Detectable on the map | Partly | Yes Clear enough |
| Raw file required at entry | Yes Settles it directly | Yes Independent of tooling | Partly Excludes phone entrants |
| Edit list declared | Partly Relies on honesty | Yes | Yes Trivial |
The third row is the strongest and the one with a real cost: it excludes phone photographers who never had a raw file, which for some competitions is most of the entrants. The workable compromise is to require the original camera file, whatever form it takes.
What organisers should actually do
-
Write the rule about the picture
Name what may not be changed: no adding or removing elements, no combining exposures beyond a stated limit, no replacing skies. Do not name technologies.
-
Require the original file at entry
Collected up front rather than requested later. Asking only the shortlist for raw files invites a scramble and an appeal.
-
Screen the shortlist, not the entries
Running a check across a hundred finalists is proportionate. Running it across ten thousand entries produces a pile of false positives and no capacity to resolve them.
-
Compare the raw against the submission
This is the actual adjudication. A detector points at where to look; the two files answer the question.
-
Publish an appeal route
Entrants disqualified on an image analysis need somewhere to send their raw file and a named person who will look at it.
The fifth step is the one competitions keep learning the hard way. A public disqualification that turns out to be wrong does more reputational damage than any undetected entry, and the correction never travels as far as the accusation.
What entrants should keep
The defence is provenance, prepared before it is needed. A photographer who can produce the raw file, the edit history and the frames either side of the winning shot has an answer no analysis can override.
Keep the whole sequence rather than the selected frame. A burst of twelve exposures around the moment, with consistent metadata and consistent light, is close to unfakeable and takes no effort to retain.
Keep the layered editing file too. Version history in mainstream editors records which tools touched which regions, and that record answers the specific question a competition will ask far better than an argument about detectors.
Where a phone was used, the equivalent is the original file straight from the camera roll, before any share or export. That is a weaker artefact than raw and considerably better than nothing.
The categories that cause the most disputes
Complaints cluster in a few genres, and knowing which ones lets an organiser write better rules rather than adjudicating case by case.
Wildlife is the most contested. Long lenses, heavy cropping, aggressive noise reduction at high sensitivity and composite backgrounds are all common, and the line between acceptable processing and a fabricated encounter is genuinely hard to draw.
Astrophotography is the second. Stacking dozens or hundreds of exposures is the standard technique, and the finished image is a computed result by construction. Any rule about combining frames has to say what it means here or it will disqualify the whole category.
Photojournalism is the third, and the one where the rules are strictest and clearest. Most established competitions permit only global adjustments, prohibit anything that changes content, and require the raw file. Those rules predate generative tools and have aged well precisely because they describe the picture rather than the software.
Portrait and fashion sit at the other end. Retouching is expected, the rules are permissive, and almost nobody complains, which is a useful reminder that a rule is only worth writing where the audience cares about the answer.
One organisational point sits behind all of this. Whoever writes the rules should be somebody who edits photographs, not only somebody who administers competitions. Almost every unenforceable rule in this field was written by a person describing a technology they had not used.
The same applies to adjudication. A judge comparing a raw file against a submission needs to know what ordinary processing looks like in that genre, or every strong entry becomes a suspicious one.
A closing note for organisers with small budgets. None of this needs software. Requiring the original file at entry and comparing it against the submission for the shortlist is entirely manual, costs nothing, and is more reliable than any automated screen.