Nikon Small World in Motion’s disqualified AI video was judged as art, not data

Nikon has thrown out the winning entry of its Small World in Motion competition, and the official reason fits in two words: generative AI. The video, submitted by optical engineer Ning Xu, claimed to show cilia, the hair-like structures that sweep mucus out of human airways, beating abnormally in a child with primary ciliary dyskinesia, a rare genetic disorder known as PCD. Scientists said parts of it could not exist in a living body. Nikon re-examined the entry, consulted its judges and ruled that it broke the contest’s rules.

Read as a cheating scandal, the story ends there. It should not. The more uncomfortable reading is that the contest received the entry it was built to reward. Small World in Motion scores videos on originality, informational content, technical proficiency and visual impact, and the step that sank Xu was the one most likely to lift that last score. A microscope video is a measurement before it is a picture. Nikon judged the picture and took the measurement on trust, and that gap, more than any single entrant, is what it now has to close.

The stakes reach well beyond a niche prize. Microscopy is how most people ever see the inside of a cell, and images like these travel: into lectures, explainers and the patient forums where families try to understand a diagnosis. Robert Hirst, lead scientist at the NHS diagnostic centre for PCD at the University of Leicester, told the BBC that the row “has impacted a lot of patients, scientists, doctors and PCD support networks around the world.” When a striking image of a disease turns out to be partly rendered, the people who lose most are the ones who live with that disease.

What Nikon ruled, and what Ning Xu says he did

Nikon’s statement, given to PetaPixel, is careful. “After thoroughly re-evaluating the video and supporting materials and consulting with members of our judging panel, Nikon has determined that the winning video did not comply with the competition rules regarding generative AI,” a spokesperson said. The company added that the decision “should not be interpreted as a judgment of the entrant’s professional reputation, scientific contributions, or intent.” Nguyen Nam Nhat, originally second with a video of a tiny roundworm and a single-celled Dileptus, now holds first place, according to The Verge and PetaPixel.

Xu never denied using AI. In a LinkedIn comment quoted by The Verge, he wrote that “an unsupervised neural-network method was subsequently used for AI-assisted post-processing to distinguish and visualize features in the reconstructed grayscale images from the super-resolution optical imaging.” In another post cited by the BBC, he drew the line he believes he stayed behind: “AI was not used to generate the experimental movie, the cilia, or their motion.” Amateur Photographer reports that he described the processing as being “mainly to improve the visual presentation, rather than provide an anatomical reconstruction.”

The critics went further than style. Edward Phelps, a bioengineering researcher at the University of Florida, wrote on LinkedIn that “purple structures pop in and out of existence” and that “green cilia also appear from nowhere and do not match the known size,” as PetaPixel reported. Quoted by Gizmodo, he was blunter: a sub-epithelial structure made of extracellular mitochondria the size of cell nuclei “does not occur in biology.” Ian Donovan, an MD/PhD student at UT Southwestern Medical Center, said he ran the video through Google’s Gemini and it flagged a SynthID watermark, the invisible marker developed by Google DeepMind to tag AI-generated content, according to PetaPixel and to CNN as cited by Gizmodo. The BBC reported that two former competition judges also had doubts.

Remember where Nikon started. Early in the row, the company told the BBC: “At this time, we don’t see that any rules were violated.” The entry had already passed the contest’s vetting. Nikon’s first public note said it was “carefully re-reviewing the information provided during the initial vetting,” which is a polite way of saying the original check had not caught it.

A contest that scores visual impact invites the tool that supplies it

The detail that tells this story best is a sentence Nikon published itself. Before the disqualification, the company updated its blog post on the winning entry to explain, as Gizmodo quoted it, that “an unsupervised neural-network method was subsequently used as part of the post-processing and visualization workflow to distinguish and visualize features in the grayscale data, creating a more vivid and visually engaging video.”

Read that last clause again. Nikon did not present the AI step as a confession. It presented it as a feature, in the exact vocabulary of its own judging criteria: vivid, engaging, visual. The neural network was credited with making the science look better, and looking better is a scored category. Nobody in that sentence asks what the network added.

This is the structural problem. When he won, Xu told PetaPixel that Nikon Small World “has repeated one simple message: science is beautiful.” Entrants understand the assignment. Patrick Hickey, who finished fifth, told the BBC that microscopists rarely get to show the public their work and that “a lot of the images that scientists are making have some very abstract and artistic qualities to them.” A contest that celebrates the beauty of data will eventually attract the one technology designed to produce beauty without new data. Generative models are good at plausible detail. That is their job. In a gallery, plausible detail wins. In a lab, it is called an artefact.

Sarah Goetz put the emotional side plainly on Bluesky, in a post PetaPixel quoted: “If it’s AI, which is seeming likely, there is something profoundly cynical and sad about using synthetic images to cheat in a contest that is supposed to be about how beautiful and cool biology is.” Fair. But the cynicism, if that is what it was, met an institution that had already written “more vivid and visually engaging” into its own description of the entry.

The strongest case for Ning Xu

Xu’s defenders have a serious argument, and it deserves to be stated as they would state it. Microscopy has never been raw. Researchers stain samples, apply false colour, stitch tiles, deconvolve blur and reconstruct super-resolution frames from data no human eye could read directly. The BBC notes that artificially bright colour is common in microscopy and permitted in the contest. Gizmodo points out that one widely cited set of scientific image guidelines, written before the generative-AI era, accepts colourised and processed images as long as the changes are disclosed and described.

By that standard, the defence goes, Xu did what he was supposed to do. He disclosed the neural-network step. Nikon published it. The contest rule PetaPixel quotes says that “AI-generated videos are not permitted,” and Xu insists his video was not generated: the cilia and their motion came from his instrument, and the network only separated and coloured structures in data that already existed. If an algorithm that sorts pixels into colours counts as generation, then much of modern computational imaging is suspect. And disqualifying the entrant who disclosed his method teaches the next one to say nothing. Add the evidence that triggered the review, a pile-on in the LinkedIn comments and a watermark check run through a consumer chatbot, and the process looks less like peer review than like a mob that happened to be right.

Why the colour argument does not hold

The defence fails on its own terms, because the critics were never objecting to colour. False colour relabels pixels the microscope actually measured. Phelps’s complaint is about existence: structures that appear and vanish, cilia at the wrong scale, an arrangement of organelles that, in his words, “does not occur in biology.” Hirst, who told the BBC he has been diagnosing PCD for two decades by examining cilia waveform, length, cell size and shape, said the cells and cilia “look nothing like those from PCD patients.” Those are not judgements about palette. They are judgements about whether the thing on screen is there.

The whole case turns on one distinction. False colour paints what the microscope saw. A generative model paints what the microscope should have seen.

Xu’s own words, reported by Nature and quoted by Gizmodo, sharpen the problem rather than solve it. He said the structures beneath the cilia were processed “without making anatomical claims about what those rendered features represent.” But a scientific image cannot opt out of making claims. Every visible shape in a microscopy video reads as anatomy to the viewer, and no disclaimer travels with the pixels. Melanie White, a developmental biologist at the University of Queensland, told Nature that scientists “need to be able to trust that what we are seeing is grounded in the underlying measurement.” Andrew Moore, a former Small World in Motion judge, offered The Telegraph an analogy anyone can feel: “The high-res, colorised photo may look nice, but if it was your grandma who was face-swapped, it’s going to be unsettling.”

Disclosure is necessary. It is not sufficient. Describing a method in a paragraph does not tell anyone which pixels were measured and which were inferred. The one document that would have settled it was the one critics kept asking for, according to PetaPixel: the raw grayscale video. Nikon says Xu supplied detailed technical documentation during the review. The public never saw the raw footage, and so the argument was conducted, as these arguments now usually are, through watermarks and expert eyeballing.

Raw files, not AI detectors

Photography contests have been here before. Amateur Photographer recalls that the Creative Open winner of the Sony World Photography Awards in 2023 turned out to be an AI construct, and that a finalist’s entry in the Hasselblad Masters 2026 competition was removed this year for AI use. But a science contest is a different case. In art photography, AI raises a question of authorship. In microscopy, it raises a question of truth. A synthetic landscape is a disputed artwork. A synthetic cell is a false statement about the body.

Detection will not carry that weight. What exposed this entry was a SynthID watermark, which only exists because one company chose to embed it. Gizmodo notes that most AI detectors that rely on other signals perform poorly, and that without such watermarks it can be difficult or impossible to establish an image’s provenance with certainty. An entrant who used an unmarked model would not be caught the same way. Relying on watermarks means relying on the courtesy of the tool’s maker.

The durable fix is boring, and it comes from the lab, not the gallery: provenance. In MCM’s reading, any scientific imaging contest should require the raw acquisition files with every shortlisted entry, a processing log that names each step and its software, and a raw-versus-final comparison the judges actually review before visual impact enters the conversation. Nikon now says that “advances in imaging and artificial intelligence continue to evolve rapidly, creating new challenges for organizations across many fields,” and that the episode “has shown the need” to “revisit rules and evaluation procedures for future competition entries.” That is the right instinct. The rule it needs is not a longer list of forbidden tools. It is a standing request to see the data.

What is established, and what is still argued

Established: Small World in Motion announced its 2026 winners in mid-September, with Xu’s cilia video in first place. On October 9, 2026, Nikon disqualified the video for not complying with its rules on generative AI, said the decision was not a judgement on Xu’s reputation, scientific contributions or intent, and moved Nguyen Nam Nhat up to first, according to the BBC, The Verge and PetaPixel. Xu has acknowledged that an unsupervised neural network was used in post-processing. The BBC put the prize money at $3,000. Nikon says it will revisit its rules and evaluation procedures.

Still argued: whether the cilia and their motion are genuine footage, as Xu maintains, or partly rendered, as critics including Phelps and Hirst believe; whether the neural-network step counts as generation or as visualisation; and exactly which element Nikon found non-compliant, which its public statement does not specify. MCM has not examined the raw data, and the argument above about the contest’s design is interpretation, not a forensic finding.

Next year’s winner will still need to be beautiful. The question Nikon has to add is the one any lab would ask first: show us the raw file.

Tags: Generative AI, Nikon, Nikon Small World in Motion, Ning Xu, Photography contests, Primary ciliary dyskinesia