One word on Nikon's page is doing an unusual amount of work: rendition.
The first-place entry in the 2026 Small World in Motion competition was introduced as a video of abnormal airway cilia from a child with primary ciliary dyskinesia. The winner's page now calls it a rendition and adds, "AI-assisted in post-processing."[1]
That revision is accurate as far as it goes. It also demonstrates the problem. In a scientific image, "AI-assisted" tells us about as much as "computer-assisted." It names a category of tool while leaving the consequential part vague: what did the tool add, remove, infer, or merely make easier to see?
A real sample and a disputed picture
Nikon's account says Ning Xu and his collaborators recorded a real biological sample with a custom diffractive super-resolution microscope. The measurements were reconstructed into a grayscale sequence, then an unsupervised neural network was used to distinguish and visualize features in that data.[2] Xu has said the AI did not generate the experimental movie, the cilia, or their movement.[3]
Several microscopy researchers are not convinced that the colorful structures beneath the cilia remain faithful to the measurement. They have pointed to objects that appear biologically implausible, change scale, or pop in and out of existence. Xu's response is that the processed regions were intended to improve visual presentation, not make anatomical claims.[3]
Nikon reviewed technical material from Xu and initially said the use counted as enhancement rather than generation. The company also said it saw no rule violation, although it is re-evaluating the entry and its AI policy while the video remains listed in first place.[4][5]
That leaves an important uncertainty. I have not seen the raw grayscale sequence, the model, or the intermediate outputs, so I cannot tell whether the disputed forms were measured, inferred from measurements, or invented by the processing. Neither can most people looking at the finished video. That is precisely why the label is inadequate.
Scientific images have never been raw windows
A microscope image is already the result of choices: illumination, exposure, optics, reconstruction, contrast, color mapping, and often a stack of software. Processing is not automatically deception, and AI does not turn an otherwise untouched view into a synthetic one. There was no untouched view.
But scientific images carry a promise that illustrations do not. A viewer should be able to trace the visible features back to measurements. Beauty can be one purpose of the picture, but evidence is another, and the second purpose requires a stricter receipt.
The US Office of Research Integrity published useful guidance long before today's generative tools. It says digital scientific images should be treated as data, original files should be retained, and simple adjustments to an entire image are usually acceptable. It treats filters used to improve biological images and changes confined to one region with more suspicion.[6]
Those rules do not settle the Nikon dispute, but they ask better questions than "Was AI used?" Can the final feature be located in the original measurement? Was the same transformation applied consistently? Can another person reproduce it? Does the caption distinguish observation from interpretation?
Post-processing is not a method
The phrase can cover a brightness adjustment, a denoising model, a reconstruction algorithm, or a system that supplies plausible texture where the instrument recorded ambiguity. Those operations should not inherit the same harmless-sounding label.
A stronger presentation would show the grayscale measurement beside the processed video, identify which structures the model inferred, and describe enough of the workflow for specialists to challenge it. If the colorful layer is an artistic visualization, call it one. If it is evidence, preserve the chain back to the instrument.
The argument is not really about whether science may be beautiful. Nikon built an entire competition around the obvious fact that it can. The harder question is whether a beautiful scientific image still tells the viewer which parts were seen and which parts were suggested. Until that is clear, rendition is not a minor caption edit. It is a change in what kind of thing we are looking at.
Sources
- Nikon Small World: 2026 first-place video
- Nikon Small World: Ning Xu, Studying Disease through Motion
- Nature: This award-winning microscopy image used AI
- The Scientist: Allegations of AI use in Nikon competition spark controversy
- BBC: Tiny image sparks big backlash in Nikon photo contest
- US Office of Research Integrity: Guidelines for Best Practices in Image Processing