Forum
The Garbled Label Won’t Save Us: Competitions After Midjourney V8.1

Hasselblad Masters is one of the photography world’s best-known competitions, and this spring it is back for the first time since 2023. When the company published its 70 finalists in late April, online commenters quickly focused on one image in the Street category. The giveaway, they argued, was a soft drink bottle whose label was a garbled mess: the kind of lettering that image generators have produced for years.
As of this week, the accusation has been widely reported and discussed. The competition’s rules state that entrants must ensure the originality and authenticity of their submissions, and that the use of AI synthesis leads to disqualification.
Meanwhile, on 30 April, Midjourney released V8.1, a point update to the alpha model it introduced in March. Among the improvements it highlights are faster generation, native high-resolution output and better handling of text in images, including a way to request specific readable words.
Put those two stories side by side and the problem for every competition becomes obvious.
Detection by mistake has an expiry date
For the past few years, many AI images have been caught because of their errors: extra fingers, melted architecture, nonsensical lettering. Communities of careful viewers have become very good at spotting them, and that vigilance has real value.
But every model release closes some of those gaps. Hands have improved. Lighting has improved. Now text is improving in several major models at once. A future entry with a perfectly legible label will not be caught by the method that flagged this one.
Automated detectors share the weakness in a different form. Software trained to recognize the fingerprints of today’s models has to be retrained for tomorrow’s, and its scores are rarely conclusive enough, on their own, to justify taking an award away from someone.
Competitions that rely on spotting flaws are relying on the tools staying flawed. They will not.
What works instead
The answer is not to give up on authenticity, but to shift from inspecting the result to examining the process. Photography competitions already have some of the tools for this, and AI-focused competitions need them even more.
- Original files. Raw files, uncompressed originals and full-resolution frames are much harder to fake convincingly than a final JPEG.
- Context. Contact sheets, sequences shot before and after the winning frame, and location details make a fabricated scene harder to sustain.
- Provenance records. Cameras and phones that sign images at capture create evidence that survives editing, as long as it is preserved.
- Conversation. Asking a finalist to talk through how a picture was made remains one of the most effective checks there is.
Fairness to entrants
There is another side to this. Online accusations can be wrong, and a photographer accused of using AI deserves a fair process before anyone reaches a verdict. Odd lettering can have innocent causes too, from motion blur to a worn or damaged label. Clear rules about what evidence finalists must provide protect honest entrants as much as they deter dishonest ones. They also take the judgment out of comment threads and put it back with the organizers.
Where AI awards fit
Competitions that welcome AI work face a different version of the same problem. The question is not whether AI was used, but whether its use was described honestly. The International AI Award asks entrants to explain their process, name the models and versions they used and, for code-based work, submit their source code. Its 2025 rules also set a final round of live interviews with finalists. Requirements like these test the account of the work, not just its surface.
The takeaway
The garbled label was useful while it lasted. Generative tools are now improving faster than viewers’ eyes can adapt, and text was one of the last obvious tells. Competitions in every field, with or without AI categories, should move now from hunting for mistakes to asking for evidence. The entrants with nothing to hide will be glad they did.



