I have been making an attempt out the picture enhancing capabilities in Google’s newest generative AI mannequin, and you’ll, too. Named Gemini 2.5 Flash Picture, it upgrades the flexibility to edit your pictures natively in Gemini.
AI lovers have referred to it because the “nano bananas” mannequin, spurred on by a sequence of teasers from Google execs. Google has been investing closely in its generative media fashions this yr, dropping up to date variations of its picture and video generator fashions at its annual I/O builders convention. Google’s AI video generator Veo 3 shocked with synchronized audio, a primary among the many AI giants. And creators have made greater than 100 million AI movies with Google’s AI filmmaker software, Movement.
I’ve spent a number of time testing AI inventive software program, and I used to be excited to see Google’s newest drop amid all of the banana fanfare. However my testing of two.5 Flash Picture confirmed that simply because one thing has a flashy entrance does not imply it will all the time lives as much as its hype. Here is how my expertise with Gemini nano bananas went: the nice, the dangerous and the irritating.
What labored
The Gemini bananas mannequin is spookily good at including components to current photographs, mixing AI-generated components nicely into any image you snapped. It additionally maintains a decently secure degree of character consistency — that means the folks in my pictures weren’t too distorted or wonky after going by means of the AI processing. These are each vital distinctions for AI picture packages, and one thing Google mentioned it had labored to enhance.
You’ll be able to see each of those traits on this image of my sister and me. Our basic appearances are unchanged within the edited model (proper), exhibiting off that character consistency. I requested Gemini so as to add a 3rd sister who appeared much like the 2 of us, which it did scarily nicely by including a 3rd lady in between the 2 of us.
Welcome to the household, scarily accurate-looking AI sister.
I used to be additionally fairly impressed with how shortly Gemini may spit out accomplished photographs. Wherever beneath a minute will get a gold star from me, and Gemini was usually dealing with requests in beneath 15 seconds. I additionally appreciated the way it added a watermark to all the photographs it created and edited — even when I do not love how tech firms have corrupted the sparkles emoji for AI, it is extraordinarily vital to have some markers of AI-generated content material. Google’s SynthID and behind-the-scenes work additionally assist differentiate AI content material from human-created imagery.
Gemini is sweet at wholesale AI picture creation, too, however I like to recommend utilizing its Imagen 4 or one other AI picture generator as a substitute — they’ve extra hands-on controls and settings that get you nearer to what you need with much less work.
What actually did not work
There are severe limitations to Gemini bananas. It routinely generated sq. photographs, and follow-up prompts asking for photographs to be tailored into different dimensions have been ignored or failed.
I additionally seen that Gemini decreased the decision of lots of my pictures. I primarily take pictures with my iPhone 16, which has stellar cameras, however after going by means of the Gemini bananas mannequin, these wonderful particulars have been typically blurred. That is annoying and will not win over any photographers.
Regardless of my request to brighten up the goal’s rings, the unique picture (left) has extra richness and texture than the AI-edited one (proper).
I attempted repeatedly to get Gemini to deal with picture edits that might’ve been tough for me to do manually. That is one space in picture enhancing the place AI is meant to excel — automating mundane however detail-intensive edits. Sadly, Gemini actually struggled with immediate adherence right here, that means it did not do what I requested.
I attempted many instances to get Gemini to take away reflections from a snap of a Freakier Friday film poster, however they stubbornly remained. And the extra I attempted to get it to take away the reflections, the poorer the standard of the picture turned with each immediate. As soon as-clear textual content was in the end illegible after I lastly gave up, to not point out the unintentional, scary-looking harm accomplished to the faces of Lindsey Lohan and Jamie Lee Curtis.
Gemini nano bananas struggled to generate photographs in numerous dimensions. Resizing and cropping photographs is a core picture enhancing course of, however Gemini did not — or could not — deal with easy sizing tips in my prompts.
All I needed was for Gemini to take away the reflection within the upper-right nook of the poster, however as a substitute it warped the film tag line and faces.
I reached out to Google concerning the decision and dimension points and a spokesperson mentioned the tech firm is “conscious and actively engaged on each points. It has been an enormous replace from our earlier mannequin however we’ll proceed to enhance on the mannequin.”
Total, Gemini nano bananas proved to me that Google is severe about persevering with to dominate in generative media. But it surely has vital pitfalls, with too huge a deal with producing new components moderately than utilizing AI to enhance and tweak widespread picture points. For now, the nano bananas mannequin is greatest fitted to Gemini followers who wish to make huge edits shortly. For these of us in search of extra exact instruments, we’ll have to attend for Google’s subsequent huge replace or discover one other program.
Gemini nano bananas availability, pricing and privateness
You needn’t do something to entry the brand new mannequin; it is routinely added to the bottom Gemini 2.5 Flash mannequin. Gemini is offered without spending a dime, with extra fashions and better utilization caps obtainable in Google’s AI plans beginning at $20 per 30 days.
In case you’re a paying subscriber, you might also have the ability to entry the mannequin by means of Google AI Studio. From there, all you need to do is add a picture and sort out your immediate. Every immediate makes use of wherever from one to 2 thousand tokens, relying on the extent of element required. Adobe Categorical and Firefly customers may entry the brand new mannequin now.
Google’s Gemini privateness coverage says it may possibly use the knowledge you add for bettering its AI merchandise, which is why the corporate recommends avoiding importing delicate or non-public info. The corporate’s AI prohibitive use coverage additionally outlaws the creation of unlawful or abusive materials.
For extra, try the greatest AI picture mills and all the things introduced on the Made by Google Pixel 10 occasion.
