Nikon Strips Microscopy Contest Winner's Title Over Generative-AI Rules, Hands Gold to a Roundworm
Nikon says the first-place video broke its AI rules; the entrant says AI was used for postprocessing, and the contest now has to define 'real' footage.
Nikon disqualified the first-place video in its 2026 Small World in Motion competition on October 9, saying the entry “did not comply with the competition rules regarding generative AI.” The title moved to the runner-up, a video of a roundworm and a single-celled organism called Dileptus.
Let the record show: a roundworm has won a microscopy contest on the grounds that it is, at minimum, a roundworm. In a year of “is this real?” it is a refreshing credential.
First, the precise part, because this is a case where the details matter more than the jokes. Per DPReview, Nikon’s rules say AI-generated videos are not permitted and the sponsor can request the original video for verification. Nikon said its decision was based solely on the video’s eligibility under those rules, and not on the entrant’s reputation, scientific contributions or intent. The decision does not, per DPReview and Let’s Data Science, say what role generative AI played, and Nikon has not publicly described one.
The entry, by Ning Xu, an optical engineer at the National University of Singapore, per Gizmodo, was described as showing airway cilia movement associated with primary ciliary dyskinesia, a rare respiratory condition. Gizmodo reports Xu said the video began as real research footage and that he used generative AI for “postprocessing.” Other reporting cited by DPReview and others says that, according to the BBC, the winner acknowledged using AI. Gizmodo adds that experts found structures in the video that do not appear in biology, and that the video carried an AI-identifying SynthID watermark. DPReview, citing the BBC, quotes a primary-ciliary-dyskinesia expert saying the cells “look nothing like those from PCD patients.”
We will say plainly that we are not in a position to judge the science, the intent or the individual, and Nikon has declined to as well. This is a story about a rulebook.
The rulebook is the interesting part, because the line it draws is not obvious. “AI-generated” can mean at least three things in a microscopy workflow:
- A model invents footage from nothing.
- A model reconstructs or enhances real data, as in denoising or deconvolution.
- A model colors, segments or labels structures in real data.
Number one is plainly out under the contest’s rules. Number three is, broadly, what image processing has always done. Number two is where the argument lives, since a model that “improves” an image can supply detail the sample did not contain. Per Let’s Data Science, the ruling applies only to Nikon’s contest rules and does not set a general standard for microscopy AI, and it argues that preserving source data, processing histories and disclosed transformations helps assess cases like this one.
DPReview’s author makes a related point: AI-detection tools are often unreliable, since many rely on guesswork or on watermarks that can be bypassed. In this instance there was, per Gizmodo, a watermark, which is the rare case of the system working as designed and the story still being a mess.
The prize situation, for completeness, per DPReview. First place and the $3,000 prize go to Nguyen Nam Nhat of Vietnam for the roundworm-and-Dileptus video, previously second. Second place goes to Benedikt Pleyer for jellyfish larvae. Third is Dr. Andrew Moore, for synchronized cell division in neighboring cells.
That is, on paper, a clean and boring result, which in this genre is a form of good news. Nobody had to explain a cell dividing in sync with its neighbor. It simply did that, and then it was filmed.
Nikon says it needs to re-evaluate its rules and procedures. This is the right call and also the hard part. A contest that rewards beautiful footage of living things at the scale of a micron has always, quietly, rewarded processing: contrast, color, stacking, deconvolution. The art form is “what this looked like if you had better eyes.” The moment a generative model enters, the question becomes how much of “better eyes” is the eyes.
A fair policy probably looks like what Let’s Data Science suggests: show your source data, show your processing history, declare your transformations. It would make for a lot of paperwork for an event that is, remember, about a wriggling thing in a droplet. But the alternative is a world where a fabricated ciliated cell can beat a real roundworm. And the roundworm, as we’ve seen, will not accept that.
The wider lesson, if you want one, is the usual one. Rules written for a previous technology are being tested by the current one, and the test is whether the people who run contests, journals and newsrooms can say, in advance and in writing, what counts. Nikon’s rule said “AI generated videos are not permitted.” Fair enough. That sentence, at least, doesn’t spell out what to do with footage that was real before the AI touched it, if, as the entrant says, the AI was only processing it.
Which means the winner of this episode is not a person or a worm. It’s the sentence “the sponsor reserves the right to request the original video for verification.” In October 2026, it is the most useful sentence in microscopy.
Sources: