Nikon Disqualifies Small World in Motion Winner After Scientists Flag AI-Rendered Cilia

TL;DR

On October 9, Nikon Small World disqualified the first-place video in its 2026 Small World in Motion competition, 23 days after announcing it. The clip, submitted by optical engineer Dr. Ning Xu, claimed to show airway cilia beating abnormally in a child with primary ciliary dyskinesia. Cilia researchers took one look and said the cells underneath the cilia could not exist, a PhD student reported a SynthID watermark in the file, Nikon's own write-up disclosed an "unsupervised neural-network method" in post-processing, and patient groups plus nine senior scientists drafted an open letter. Nikon's statement says the video "did not comply with the competition rules regarding generative AI," promotes the second-place roundworm video to first, and promises to rewrite the rules. The $3,000 question for anyone shipping an imaging pipeline: where does "post-processing" end and "generated" begin?

What Nikon announced on September 16

Nikon's press release for the 16th annual Small World in Motion contest gave first place to Dr. Ning Xu, listed as a researcher at Tsinghua University (later coverage places him at the National University of Singapore). The video was captured with diffractive super-resolution microscopy at 100X objective magnification: the team shaped the illumination with multiple light waves and a digital micromirror device, which let them film rapid cilia motion without fluorescent labels that can damage living cells.

The science hook was real. Primary ciliary dyskinesia (PCD) is a rare genetic disorder in which airway cilia beat abnormally, so the lungs cannot clear mucus. "To understand this disease, you need to watch the cilia through dozens of beat cycles, not just a still image," Xu said in the release. Second place went to Nguyen Nam Nhat of Vietnam for a roundworm meeting a single-celled Dileptus (differential interference contrast, 40X), and third to Benedikt Pleyer of Germany for jellyfish larvae in water droplets (darkfield, 25X). The judging panel was Joe Dragavon (University of Colorado Boulder), Quinten Geldhof (Museum of Science, Boston), science writer Corey S. Powell, and Meredith Sagolla (Genentech).

How it unravelled

Nikon reposted the clip on LinkedIn, and the people who stare at cilia for a living replied. Edward Phelps, a bioengineering researcher at the University of Florida, wrote that the purple structures under the cilia "resemble mitochondria" but sat outside cells and were as large as nuclei, a combination that "does not occur in biology." The blue nodules looked like nuclei but did not behave like nuclei, he could not identify the red structures at all, and, as Gizmodo and ABC News both reported, "purple structures pop in and out of existence" while "green cilia also appear from nowhere and do not match the known size."

Robert Hirst of the University of Leicester, who has diagnosed PCD for 20 years by examining cilia waveform, length, cell size and shape, told the BBC he "knew immediately that the video was fabricated" and that it looked "nothing like" footage from PCD patients. Ian Donovan, a PhD student at UT Southwestern Medical Center, posted that running the file through Google's tooling flagged an invisible SynthID watermark, the DeepMind marker that Gemini stamps on media it generates or edits. Nikon never cited that watermark; treat it as a corroborating claim, not the ruling.

Then Nikon's own competition blog quietly added a line that did more damage than any critic: "An unsupervised neural-network method was subsequently used as part of the post-processing and visualization workflow." By October 1, Nikon said it was "carefully re-reviewing" the submission and that Xu was "respectfully complying" with documentation requests covering equipment, imaging methods and processing.

Xu's defence: the cilia are real, the colour is AI

Xu did not deny using AI. In a LinkedIn response relayed by Nature and Gizmodo, he said AI was not used to generate the experimental movie, the cilia, or their motion. A network was applied afterwards to "reconstructed grayscale data to distinguish and color structures with similar morphology," and his team rendered the regions below the cilia "without making anatomical claims about what those rendered features represent." To ABC he put it more plainly: the rendering was there to "improve presentation," not to "provide an anatomical reconstruction."

That distinction is the whole case. The microscope measured a greyscale intensity field. An unsupervised network then decided which blobs were "similar" and painted them purple, blue and red. Think of an AI colouriser for old black-and-white photos: it will confidently decide grandma's dress was red, and it is painting a guess onto a measurement, not recovering a fact. 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."

The rule is one sentence long

The official Small World in Motion rules say: "AI generated videos are not permitted and Sponsor holds the right to request the original video for verification." The terms also void "Submissions generated by AI, script, macro or other automated means." Entries are judged on Creativity/Originality, Informational Content, Technical Proficiency, and Artistic and Visual Impact. First prize is $3,000.

Notice what the rule does not define: whether a network that invents colour regions over real footage makes a video "AI generated." Xu read it as no. The judges, on re-review, read it as yes. The rubric scores "Artistic and Visual Impact," and by every account the neural net was excellent at that part.

The open letter and the ruling

On October 8, cell biologist Pleasantine Mill published an open letter on FocalPlane on behalf of the PCD Foundation, PCD Support UK, the Ciliopathy Alliance, ERN-LUNG PCD, and researchers including Jeremy Reiter (UCSF), Paul Guichard (Geneva), Robert Hirst (Leicester), Simon Cleary (King's College London) and Amjad Horani (Washington University). It asked Nikon to correct the "in a child" caption while the review ran, release the original greyscale recording and full methods for independent examination, publish the outcome and withdraw the award if the recording could not support what was shown, and set clear AI and image-processing rules for future contests. The authors stressed they are "not against AI in imaging." Signatures were due at noon US Eastern on October 12, with delivery planned for October 13.

Nikon did not wait for the mail. Its October 9 statement, confirmed to PetaPixel and Gizmodo by a spokesperson, reads in full on the integrity point:

"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. Nikon has adjusted the competition rankings to reflect this decision. This decision is based solely on the submitted video's eligibility under the rules outlined for this competition. It should not be interpreted as a judgment of the entrant's professional reputation, scientific contributions, or intent."

Nikon added that "advances in imaging and artificial intelligence continue to evolve rapidly" and that the event "has shown the need for Nikon Small World and Nikon Small World in Motion to revisit rules and evaluation procedures for future competition entries." Nguyen Nam Nhat's roundworm now holds first place, the remaining winners move up one slot, and Xu's entry is gone from the gallery.

Why a builder should care

Three days ago we covered the Caine Prize, where a detector flagged a winning story and the humans followed. Here the order flipped. Nobody ran a classifier first; the people who diagnose PCD for a living saw mitochondria the size of nuclei floating outside cells and said so in public. The watermark claim came second and was never part of Nikon's reasoning. If you are building provenance or review tooling for images, video or scientific figures, the lesson is that detectors corroborate, and domain plausibility checks convict.

The second lesson is about the word "generated." Xu's pipeline matches a pattern that is everywhere now: real sensor data in, a model "enhances" the frame, a prettier artifact out. Denoisers, super-resolution upscalers, colourisers and segmentation overlays all live on that spectrum, and most contest rules, journal policies and client contracts have one sentence that does not say where the line is. Nikon is about to write a longer sentence. If you run a pipeline that touches measured data, write yours first: keep the raw greyscale, log every model that touched a frame, and label an output as rendered the moment a network adds structure that no sensor recorded.

The third is procedural. The open letter's most useful ask was "release the original greyscale recording," not "withdraw the award." Nikon's rules already let it demand the original, and the dispute would have ended in a day with the raw footage on the table.

Caveats

• Nikon has not said which element of the video breached the rule, only that it "did not comply with the competition rules regarding generative AI." Xu maintains the cilia and their motion are real microscope data.
• The SynthID watermark finding is an attributed claim by one PhD student on LinkedIn, relayed by CNN and Gizmodo. Nikon's statement does not mention it and no third party has published a reproduction.
• Xu's affiliation is reported inconsistently: Nikon's release says Tsinghua University, ABC says that was his institution at the time of filming, and PetaPixel and Gizmodo place him at the National University of Singapore.
• The open letter's signature count has not been published; delivery to Nikon was planned for October 13, after the ruling.

Key Takeaways

• Nikon disqualified the 2026 Small World in Motion winner on October 9, 23 days after announcing it, for breaching its generative AI rule; the second-place roundworm video now takes first and $3,000.
• The entrant admits an unsupervised neural net coloured structures over reconstructed greyscale footage and says it made no anatomical claims; cilia biologists say the coloured structures cannot exist in a living airway.
• Domain experts on LinkedIn broke the story; a SynthID watermark claim came later and Nikon never relied on it.
• The rule is one sentence, "AI generated videos are not permitted," and did not define whether AI-painted overlays on real data count; Nikon says it will revisit rules and evaluation procedures.
• Patient groups and nine senior scientists had drafted an open letter asking for the raw greyscale recording; keeping and releasing originals is the fix that would have settled this in a day.
• If your pipeline lets a model add structure to measured data, label the output as rendered and keep the raw frames, because the next reviewer may know the biology better than the judges did.

Sources: Nikon Small World in Motion Video Statement, Nikon Small World in Motion Official Rules, Nikon press release, Sept 16 (PR Newswire), FocalPlane open letter, PetaPixel, Gizmodo (Oct 9), Gizmodo (Oct 2), ABC News, CNN, Amateur Photographer