On my Gadget Picks channel, I reviewed the Kodak Charmera, a cheap keychain sized, 1.6-megapixel camera whose main appeal seems to be less about image quality and more about novelty. The camera is sold as a Labubu-style blind-box product, with different designs packaged randomly, and that scarcity has led some scalpers to charge far more than its original price. Amazon does have them in stock at the time of this writing (compensated affiliate link).
The image quality straight out of the camera is pretty bad—similar to what one might experience from an early consumer digital camera. But could Google’s new Nano Banana Pro AI model fix these images up and make them look modern? That’s what I explore in my latest video.
The Charmera produces images that are noisy, soft, and lacking in detail. On their own, they are barely usable. Using a prompt that Gemini itself helped generate, I fed in a selfie taken at my desk. The original file was a blur of digital noise, but the output that came back was far more detailed, with accurate colors and recognizable objects in the background. While there was some smoothing that made the image look slightly retouched, it largely preserved what was actually there.
That initial result led me to try a variety of other images. I photographed a small holiday decoration, a candle, my dog, and an outdoor scene, all using the Charmera. In each case, Gemini produced images that looked closer to what I might expect from a modern smartphone. Details that simply were not visible in the original files appeared in the processed versions, from textures on a figurine to fur and reflections. The framing and perspective stayed consistent, even when depth-of-field effects were introduced.
The experiment didn’t stop with new photos. I also revisited digital images from the late 1990s, taken with a Kodak DC120 camera. Many of those files I saved at very low resolutions, such as 320×240, which were the sharpest looking on my 1024×768 display at the time but look especially rough on today’s high-resolution displays. Running those decades-old images through Gemini produced mixed but often striking results. In some cases, textures and facial details appeared that made the photos feel contemporary, even though the originals had almost no usable information at the pixel level.
I also found Nano Banana to be a great compliment to another Kodak-licensed product, the Slide N Scan photo negative scanner. The scanner is inexpensive (comparatively) and can rapidly scan photo negatives and slides. But the output quality is nowhere near where it needs to be for professional use. But Gemini was able to dramatically transform a few of the images I fed through it from that scanner.
Not every result was faithful to the original. In some images, Gemini appeared to invent details when there wasn’t enough data to work with. A dog’s fur texture changed noticeably, and in one image of me running with my dog, my face was clearly not my own.
Scanned photos from books and yearbooks were generally handled well, including colorization, but there were occasional distortions in faces or text. Logos and lettering were sometimes incorrect or duplicated, especially when the source material was ambiguous or mirrored.
I also found that context mattered. When I scanned a 1994-era Polaroid of my Powerbook 180c and a Newton I had to give Gemini more specific hints about what was in the image. Gemini convincingly recreated the devices and dropped them in place. At first glance it looked amazing. But some elements—particularly text—were reconstructed inaccurately. In the below example you’ll see that Gemini replaced the “Macintosh” text on the computer with “Powerbook.”
Working through these examples made it clear that tools like Gemini are doing something very close to what modern smartphone cameras already do. Computational photography has shifted the process away from simply capturing light and toward interpreting data. In that sense, using Gemini on an extremely poor image from a toy camera is not all that different vs. what happens inside many smartphones today.
Used carefully, it can make old or low-quality images usable again. But it can very quickly cross the line from enhancement into fabrication. That balance is something worth keeping in mind as these tools become more accessible and more powerful.
