When a competition judge marks an image for the shortlist, they’re responding to something in the frame — a quality of light, the emotional charge of a fleeting moment, a compositional decision that feels exact. What they’re rarely told, at least in most competitions today, is how many of those qualities were authored by a neural network rather than a photographer. That gap between what a judge sees and what a judge knows is where the real argument about AI and photography is happening right now.
The question isn’t whether AI editing tools produce images that can win awards. They clearly can, and increasingly do. The question is what disclosure — or the absence of it — actually changes about how those images get judged.
What AI Editing Does to a Photograph
To understand why disclosure matters technically as well as ethically, it helps to be precise about what AI-assisted editing is actually doing to a file.
Traditional adjustments — exposure compensation, curve edits, hue shifts — are deterministic operations applied uniformly or via a user-drawn mask. A human chose the value; the math executed it. AI tools operate differently. Generative fill, sky replacement, subject-aware sharpening, and neural upscaling all involve a model making probabilistic decisions about pixels that weren’t originally in the image, or weren’t originally the way they now appear.
When an AI upscaler increases an image’s resolution, it’s not interpolating between neighboring pixels the way bicubic resampling does — it’s hallucinating plausible detail based on patterns learned from training data. When a generative fill tool extends a sky or removes a person from a background, it’s synthesizing content that was never photographed. The resulting JPEG or TIFF contains pixel values that no photon from the original scene ever produced. That is a meaningful distinction from dodging and burning, which alter tonal relationships between pixels that actually existed.
AI masking tools in Lightroom have also shifted what counts as a photographer’s “decision” — the model selects subjects, skies, and backgrounds automatically, collapsing what was once a skilled manual task into a single click. Whether that represents assistance or authorship depends heavily on the context in which the image is being judged.
Disclosure as a Variable in Aesthetic Judgment
Research in adjacent fields has long shown that labeled context changes perceived quality. Wine tasters rate the same wine differently depending on price information; gallery viewers interpret photographs differently once they know they were staged. The same effect applies here.
When a competition entrant discloses AI-assisted generation, judges inevitably process the image through a different interpretive frame. The question shifts from “how did this photographer see the world?” to “how well did this photographer use these tools?” That is not a lesser question — but it is a different one, and different competitions have different answers about which question they want to be asking.
Some competitions have drawn lines by creating separate AI-assisted categories, effectively accepting that disclosure changes the competitive category rather than simply the outcome within one. Others have banned outright any AI manipulation that introduces pixels not captured by the camera. Still others have issued disclosure requirements without enforcing them, which is arguably more damaging to trust than either of the cleaner positions.
The inconsistency creates a structural problem: photographers competing honestly by disclosing AI work are, in competitions that lack separate categories, being judged against photographers who did not disclose equivalent or more extensive edits. The disclosed edit is visible; the undisclosed one is invisible. Judges end up penalizing transparency rather than rewarding it, at least in aggregate.
Why Judges Can’t Just Look Harder
The obvious answer — “judges should be trained to spot AI edits” — is less workable than it sounds. Some AI artifacts are detectable: unnatural texture in synthesized bokeh, repeated patterns in generative fills, implausibly smooth gradients in AI-replaced skies. An experienced retoucher can catch some of these, particularly when they know what to look for and are comparing at full resolution.
But the best AI editing tools, run carefully by someone who understands their failure modes, produce output that is genuinely indistinguishable from an unaltered capture — or at minimum, indistinguishable from a heavily but manually edited one. The goalposts moved with every model release in the last two years, and the gap between detectable and undetectable AI output narrows continuously.
Regulatory frameworks are beginning to catch up. The EU’s AI Act, for instance, includes provisions touching on synthetic content — coverage we’ve already looked at in the context of what the EU’s transparency rules mean for photo editing. But competition bodies are not regulators, and their enforcement capacity is close to zero absent technical detection tools, which are themselves arms-racing against generative model improvements.
What Honest Disclosure Policy Looks Like
For a disclosure policy to do actual work rather than perform good intentions, it needs to clear a few practical bars:
- Specificity about what requires disclosure. “AI editing” is too broad. Competitions should distinguish between AI tools that operate on captured pixels (noise reduction, sharpening, sky selection masking) and tools that generate or synthesize new content (generative fill, neural upscaling beyond a defined threshold, background replacement).
- Submission metadata requirements. Requiring entrants to submit the original RAW file alongside the edited version gives judges a baseline comparison. A responsible disclosure framework asks not just “did you use AI?” but “here is what changed between capture and submission.”
- Category structure that doesn’t punish disclosure. If disclosed AI work competes in the same category as undisclosed AI work, the policy is functionally broken. Separate categories or explicit judging criteria that account for AI assistance are more honest to everyone involved.
- Clear consequences and a mechanism to apply them. A disclosure requirement with no enforcement is worse than none, because it creates the appearance of governance without the reality.
Some of this mirrors how documentary and photojournalism competitions have long handled color-grading and manipulation — accepting tonal adjustments while drawing the line at content changes. The underlying logic is the same: the photograph is meant to be evidence of something that existed in front of the lens.
The Actual Stakes
Awards create careers. Winning a major competition changes what a photographer can charge, which clients approach them, what galleries show their work. When the competitive field is opaque about AI involvement, that career-shaping function operates on incomplete information. Judges think they’re selecting for one kind of excellence; they may be selecting for a combination of that excellence and undisclosed technical augmentation.
Disclosure doesn’t level the playing field automatically — someone skilled with AI tools will still produce better results than someone who isn’t. But it makes the field legible. Judges can make category-appropriate decisions. Audiences can interpret winning images accurately. Photographers who choose not to use generative tools know they’re being assessed on comparable terms to those who do.
The photograph that won deserved scrutiny on its own terms. Disclosure ensures that scrutiny is applied to the right question.