I let Gemini edit my photos, and what the AI is good at surprised me

• Gemini vs. photo editing software
• Gemini isn’t always the best with simple edits
• Gemini fairs much better with bigger, more creative edits
• Gemini is not a straightforward replacement for Photoshop

Summary

• Gemini is pitched as a productivity tool, but Google is trying to make it a better tool for editing images, too.
• The company’s new image model lets you make edits just by typing them into Gemini’s prompt box.
• Gemini seems to excel at big, creative edits — convincing background changes and object removal.
• The AI sometimes falls short when it tries to make precise tweaks.

Google pitches Gemini as an all-in-one productivity tool, one capable of helping with multiple aspects of the average person’s personal, professional, and creative life. And if it wasn’t clear the company viewed its AI assistant and models that way, the fact it inserts Gemini all over Google Workspace, is hopefully proof. The company’s belief isn’t all smoke without any fire, though. Google has started to demonstrate that Gemini can do things like edit your calendar or work inside apps in the right setting. Now, though, the company’s also interested in making Gemini a better tool for editing photos with its new “Nano Banana” image model.

The promise of Al, and this updated version, is that you don’t need experience or knowledge of a specific piece of software to get the final image that you want, though.

Natural language photo editing — where you just tell Gemini how you want a photo to change — was part of the company’s pitch for the Pixel 10, but that feature is available in all the places you can access Google’s models now. While I remain skeptical that talking or typing your edits is better than physically manipulating with a mouse or stylus, after trying out Gemini’s new skills, I was impressed by just how much Gemini can do.

Gemini vs. photo editing software

Why would you let AI edit your photos?

So far, Google’s Gemini models have proven themselves adept at generating text and sorting through large quantities of data. As long as Google has considered Gemini “multimodal” it’s been able to understand and manipulate images, but the simple act of editing photos was still faster in Photoshop, Photomator, or Lightroom.

The promise of Al, and this updated version, is that you don’t need experience or knowledge of a specific piece of software to get the final image that you want, though. All you have to do is clearly ask for what you want and Gemini is supposed to be able to do the rest. I tried to experiment with Gemini’s improved photo skills with that in mind. Not necessarily being precise with the edits I wanted to see, but instead prompting the model with my gut feelings about what seemed off about each photo.

Gemini isn’t always the best with simple edits

The image model struggles with small tweaks

Using a collection of sample photos I uploaded to the Gemini app for iOS, I was able to adjust settings like color and white balance with ease, simply by asking. Sometimes the changes were subtler than I imagined, like in my photo wearing the Humane Ai Pin, but it always seemed like Gemini was at least trying to do something. Things got more complicated (and frustrating) when I asked for something more involved, like changing the orientation of an object in a photo, like asking for the Ai Pin to be straightened so it wouldn’t lean to the left. Gemini just wasn’t able to do it.

The AI assistant was fairly competent at zooming and cropping around a specific part of an image, but in the case of a photo of dogs herding goats I uploaded, the cropped image does have some of that tell-tale smoothness I associate with Al imagery. I think the image is still serviceable, but the details Gemini generates to fill-in for information your smartphone just didn’t capture aren’t always going to be of equal quality.

Based on my tests, describing what seemed wrong about an image and then asking Gemini to fix it produced better results, than trying to get granular with tweaks. You’ll still likely need follow-up prompts to get exactly what you want out of Google’s image model. In the editing software I’m familiar with, I’d probably get similar results faster, though, and some software’s automatic correction features might even work better than Gemini.

Gemini fairs much better with bigger, more creative edits

The wilder the idea, the better the image model is at selling it

Rather than little adjustments, what Google’s updated image model seems to really excel at is making big stylistic and creative changes. If you want to completely reinvent or alter an image, there’s a good chance Gemini can do it in a convincing way (which, as you can imagine, isn’t great for a shared notion of truth). I was able to remove a fence from a photo of emus without any additional prompting, and I think the final result looks very natural.

Asking Gemini to make a photo of a house in San Francisco look like it was taken on a rainy day was similarly successful, complete with lighting changes, background replacement to add clouds, and a faux rain effect. These images might not fool anyone looking closely (the Gemini watermark is also a dead giveaway), but if you’re scrolling past them on social media, they’re convincing. I think that because people expect a certain amount of creative license with these images, it’s also easier to overlook discrepancies.

Gemini is not a straightforward replacement for Photoshop

Don’t cancel that Creative Cloud subscription just yet

Based on these experiments, I don’t think I can confidently say Gemini is a perfect photo editing tool, particularly if you just want to make simple tweaks. You’ll still want normal software for that, and the built-in editing tools in your phone’s photo gallery app might be enough.

For more heavy-handed changes, though, I think there’s a compelling case for Google’s image model becoming the one-stop shop for wild edits. This new image model does seem pretty good at creating images that would be well out of reach of the average smartphone photographer, and if you find that interesting, it’s well worth a try.