Misfocused Photo’s Quick Fix with ChatGPT and Nano Banana

When on an assigment I often take with me a second camera with a vintage lens. This is a way to test old glasses along modern ones, without risking failure. Every now and then, therefore, I switch from the work gear to the personal one to give it a try. Sometimes I get good results, some other times I don’t, like in this shot taken during the 2026 Italian Skating Championship. As it is apparent, I misfocused the skater trying to get up, so the photo turned out as an egregious miss.

As they say, however, I exploited the mistake to turn a problem into an opportunity. So I decided to see if an AI could —and if so, to what extent— fix the photo with a very simple prompt: As a professional photoretouch expert and photoshop power user edit this photo so that the skater is in perfect focus.

This is how ChatGPT executed the prompt:

and this is how Nano Banana edited the photo:

Overall, ChatGPT’s editing is clearly superior, particularly with regard to refining the focus. It aggressively blurred the background  and sharpened the athlete in the foreground and saturated colours, as the right hand purple bracelet shows. However, the recurring issue of rendering hands accurately remains, as evidenced by the misrepresentation of the right hand’s fingers. Nano Banana, by contrast, was more conservative: it did not alter the foreground/background separation and did not try to fill the blanks by adding what was missed and supposed to be there.

Then, I tried a slightly more detailed prompt, by adding the specs of the camera and the lens: As a professional photoretouch expert and photoshop power user edit this photo so that the skater is in perfect focus, considering that the photo was taken with a Pentax K-1 II and a vintage Carl Zeiss Ultron 50/1,8 at ISO100, respect the depth of field and the hyperfocal of the lens.

ChatGPT’s rendering was essentially the same: the background is more blurred, but the right hand issue persists.

The Nano Banana-generated image is not so different than the previous one and did not improve the original photo:

In abstract terms, Nano Banana’s minimal editing ensured the image remained faithful to the original, whereas ChatGPT produced an entirely different result. One might, therefore, object that what ChatGPT did was actually a transformation rather than a sophisticated editing. However correct in abstract terms, this use case has nothing ado, I think, with the ongoing discussion about the impact of AI on photography as a business and a means of personal expression.

This experiment was not meant to explore in detail the potential of AI-assisted post production, but rather to see how much of a quick fix AI-editing could provide. Time factor considered – need to sent the photo back to the editorial team as fast as possible – ChatGPT was fast and good enough at turning a missed photo into one that could be used, at least in an online context; on the other hand, NanoBanana was not as useful at all.