When AI Paints the Picture: The Microscopy Image Controversy Shaking Science

Posted on 2 October, 2026 by Mehrdad Fathi

A prize-winning video of lung tissue has ignited a sharp debate about where AI-assisted visualization ends and data misrepresentation begins.

The winning entry

On September 15, Ning Xu, an optical engineer at the National University of Singapore, was announced as the winner of Nikon’s prestigious Small World in Motion competition — an annual celebration of light microscopy videography. His entry captured the movement of cilia lining the airways of a child with primary ciliary dyskinesia, a rare genetic disorder affecting the lungs, sinuses, and ears.

The video was visually striking. Below the moving cilia sat vivid red, purple, and blue structures — arresting to look at, and immediately suspicious to experts.

The controversy

Within days of the announcement, microscopy researchers began raising questions on social media. Edward Phelps, a bioengineering researcher at the University of Florida, wrote on LinkedIn that the purple structures resembled mitochondria — but that “sub-epithelial structures composed of extracellular mitochondria of the same size as cell nuclei do not occur in biology.” The blue structures looked like nuclei in form, he noted, but did not behave like them. The identity of the red stain remained unclear.

On September 22, Nikon updated the competition blog post to disclose that Xu had used an “unsupervised” AI model for “post-processing.” Xu later clarified on LinkedIn that AI was not used to generate the cilia or their motion, but was applied after imaging “to distinguish and color structures with similar morphology” in the underlying grayscale data. The goal, he wrote, was to make the final video “visually engaging as well as scientifically interesting.”

Xu did not respond to requests from Nature about the biological accuracy of the structures, nor to suggestions that he share his unprocessed source video.

Why this matters

The controversy cuts to something fundamental about scientific images. “Scientific images are not just illustrations,” says Melanie White, a developmental biologist at the University of Queensland. “They are data, and we need to be able to trust that what we are seeing is grounded in the underlying measurement.”

Markus Sauer, who studies super-resolution microscopy at the University of Würzburg, agrees that AI-assisted visualization is not inherently problematic — the issue arises when it misrepresents or exaggerates experimental findings. AI-generated images, he notes, tend to simplify complex biology, potentially creating the impression that a mechanism is fully understood even when the underlying data remain ambiguous.

White argues that the path forward is transparency: Xu should publish the original, unprocessed video alongside a detailed account of how AI was applied. “If the underlying data demonstrate that the final movie faithfully represents the biological information acquired, that should address many of the concerns.”

Nikon, for its part, said it was re-evaluating its vetting process and that Xu had provided detailed technical documentation. The outcome of that review has not been publicly disclosed.

A broader challenge for science

This is not a story about one video. It is a story about a tension that will only intensify as AI tools become more capable and more integrated into scientific workflows.

Researchers are increasingly using AI to extract signal from complex, noisy data sets — a legitimate and often powerful approach. But the line between enhancing visibility and fabricating biology is not always obvious, especially to competition judges, peer reviewers, or the public. As White puts it, the field needs clear standards for what must be disclosed when AI is used in image processing, particularly when the output is presented as representative of real biological structures.

The Nikon Small World competition was designed to celebrate the beauty of the microscopic world. The irony is that beauty, in science, can only be trusted when it is honest about what it is.

Source: Nature (2026). doi: https://doi.org/10.1038/d41586-026-03086-z