Nikon disqualifies Small World winner after AI was found in video

Nikon disqualified the video because it did not comply with the competition’s rules on generative AI, which say AI-generated videos are not allowed. The company reached that decision after re-examining the video and supporting materials and consulting members of its judging panel.

Nikon didn’t spell out exactly which part of the rules was broken. It stressed that the ruling was based only on whether the video was eligible, and that it should not be read as a judgment of the entrant’s professional reputation, scientific work or intent.

The winning entry was submitted by Dr. Ning Xu, an imaging researcher who has been linked in reports to both Tsinghua University in China and the National University of Singapore. His short video claimed to show cilia, tiny hair-like structures that line the airways, beating abnormally in a child with primary ciliary dyskinesia (PCD). PCD is a rare genetic disorder that stops cilia from clearing mucus properly, leading to chronic lung, sinus and ear infections.

The trouble began after the winning video was posted on the competition’s LinkedIn page in late September. Scientists, including former judges of the contest, started telling Nikon that it didn’t look real.

• Things that vanish. Edward Phelps, an associate professor in Florida, pointed out that purple structures appeared and disappeared, and green cilia seemed to come from nowhere and didn’t match their known size.
• Wrong biology. Others said cell features appeared and disappeared in ways real cells don’t, and that the video had structural abnormalities.
• Hidden watermarks. Some researchers said they found what looked like AI watermarks in the source files.

Dr. Robert Hirst of the University of Leicester, lead scientist for the NHS center for PCD diagnosis there, was among the first to raise concerns. He said he has diagnosed PCD for 20 years by studying cilia, and that the cells in the video looked nothing like those from PCD patients. He added that the controversy had affected patients, doctors and support networks around the world.

Several scientists said the dispute could have been settled quickly if the original raw grayscale video, the unedited footage straight from the microscope, had been shared.

Xu admitted using AI, but said he believed he was within the rules and cooperated with Nikon’s review. He said AI helped separate and display features in reconstructed grayscale images, specifically in the area beneath the cilia.

He denied using AI to create the experimental footage, the cilia or their movement, and said he did not use it to rebuild the underlying cells. He said his aim was a video that would grab viewers as well as inform scientists, and that a fuller description of his optical system and analysis of the cilia’s motion is being prepared for publication.

That gap is at the heart of the case. Using AI to clean up or highlight an image is common in modern imaging. Using it to produce what you’re showing is not. Where one ends and the other begins is exactly what contests, journals and labs are now trying to define.

With the original winner removed, everyone else moved up a place. Nguyen Nam Nhat of Vietnam is the new first-place winner, for a video of a tiny roundworm and a single-celled organism called Dileptus.

• Second: Benedikt Pleyer of Bavaria, Germany, for jellyfish larvae suspended in water droplets.
• Third: Andrew Moore of the Howard Hughes Medical Institute in the US, for neighboring cells dividing in sync.
• Fourth: Patrick Hickey of Hypha Research in Edinburgh, Scotland, for mitochondria and chloroplasts moving inside leaf cells.
• Fifth: Igor Adameyko, for a small, colorful sea slug on the move.

Small World in Motion started in 2011 as a video spin-off of Nikon’s Small World Photomicrography Competition, which has run since 1974. Both celebrate what can be seen through a light microscope.

For anyone who enters photo contests, the message is simple: read the AI rules closely, keep your original files, and be ready to share them. Raw footage is the fastest way to prove an image is real.

For the rest of us, it’s a reminder that even expert judges can be fooled at first, and that the people who caught the problem were specialists who knew exactly what real cells look like. When you see a striking image of something you can’t check yourself, it’s worth asking where it came from and whether the original is available.

Nikon says the case showed how fast imaging and AI are changing, and that it needs to revisit the rules and evaluation procedures for both Small World contests. Expect clearer limits on AI processing, and more demands for raw files, when entries open for the next round.