Nikon Disqualifies Its Microscopy Video Winner After Judges Find Generative AI in the Footage

Nikon Instruments has disqualified the video that won first place in this year's Small World in Motion competition, saying that after a renewed review and consultation with its judging panel it determined the entry "did not comply with the competition rules regarding generative AI." The video has been removed from the contest's winners page, and the rankings have been updated.

The entry came from optical engineer Dr. Ning Xu and claimed to show tiny, hair-like cilia beating in the airway of a child with primary ciliary dyskinesia, a genetic condition affecting the structures that clear mucus from the lungs. Skepticism built online within days of the announcement: researchers flagged structures that do not exist in biology, and the clip carried a SynthID watermark, Google's marker for AI-generated content.

Xu acknowledged using AI in a LinkedIn comment, writing 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." Nikon initially treated that distinction as compliant — the company had said that using AI to enhance microscopy-generated source data, rather than to generate imagery from scratch, fell within the rules — before opening a formal review after the backlash.

The reversal landed harder because the subject was medical. Robert Hirst, who has diagnosed PCD at the University of Leicester for twenty years, told the BBC that whatever the video showed, it looked nothing like cilia from PCD patients, and that the episode had a negative impact on PCD patients, doctors and support networks worldwide.

First place now belongs to Nguyen Nam Nhat of Vietnam, whose video shows a roundworm and the single-celled organism Dileptus exploring their surroundings. Nikon was careful to separate the ruling from the person: the disqualification, the company said, "should not be interpreted as a judgment of the entrant's professional reputation, scientific contributions, or intent," and it plans to revisit its rules and evaluation procedures for future editions of the contest, which has run since 1975 and added a video category in 2011.

The episode is a clean illustration of where scientific imaging rules now stand. Disclosure norms built around color adjustment and contrast stretching predate generative models by decades, and they assume the underlying pixels are real. A generative model fabricates pixels, which makes the distinction one of kind rather than degree. The practical fix is procedural: competitions and journals increasingly need provenance metadata — capture settings, raw frames, watermark checks — rather than an honor system. Nikon's reversal suggests even organizers with five decades of history can get that wrong at least once.