Comply with ZDNET: Add us as a preferred source on Google.
ZDNET’s key takeaways
- Google’s Nano Banana Professional earned a near-perfect rating.
- ChatGPT picture ranked second; others typically mangled textual content and faces.
- 9 robust prompts reveal which AIs are price subscribing to.
When generative AI first hit the scene in early 2023, we have been amazed by two capabilities we hadn’t seen earlier than: the power for the AIs to write down complete content material, and the power for some AIs to generate graphics and photo-realistic photos.
Clearly, AI fashions have come a good distance in simply three years. In my latest review of the best chatbots, I additionally listed the perfect free AI instruments for producing photos. On this article, I will take you a bit of deeper into the paid tiers of AI picture era instruments.
Additionally: The best AI chatbots of 2025: I tested ChatGPT, Copilot, and others to find the top tools now
My take a look at suite falls into 4 important classes: revising present photos, producing authentic photos, including textual content to photographs, and leveraging popular culture. For comfort, I generated all the pictures throughout all of the checks in 1:1 sq. ratio, largely in order that they could possibly be displayed side-by-side simply.
I generated over 50 photos utilizing six standard AI picture turbines. Every picture generator was subjected to 30 particular person evaluations, and from that got here an combination rating.
We now have a winner
There is a winner — a no-contest, whole knockout winner. It is a first in my AI evaluations. Often, there are instruments that every do a reasonably good job. Nevertheless, on this set of checks, there’s a completely clear standout champion: Google’s Nano Banana Pro, a part of Gemini 3.
It aced each certainly one of my checks with a near-perfect rating of 93%. ChatGPT got here in second place, at 74%. The opposite 4 ranked between 43% and 54%.
Additionally: Stop using ChatGPT for everything: I use these AI models for research, coding, and more
All six contenders did a high quality job when requested to create a brand. Past that, efficiency was all around the map.
The underside line is, if you’d like critical AI picture era, spend $20/month for Gemini 3, and you will get Nano Banana Pro. That is it. There isn’t any different doable conclusion.
Though some instruments supply extra options, the first distinction between the paid and free variations is the quantity of utilization you get. For instance, Nano Banana Professional lower me off after two photos earlier than I upgraded.
Additionally: How to install and configure Claude Code, step by step
And, only for the report, I purchased all my very own subscriptions to do these checks. No person is paying for this conclusion. I am as shocked in regards to the Gemini conclusion as a lot of you is perhaps.
OK, so let’s get began.
Google Gemini 3 Nano Banana Professional
Greatest AI picture generator general (by far)
- Total rating: 93%
- Value: $19.99/mo as a part of Google AI Professional
AI enables you to do picture recontextualization in minutes. Principally, you possibly can feed an AI an present picture and ask for it to be modified into one thing fully completely different. The formal take a look at is feeding an image of me out on a stroll and asking the AI to place me in a US Navy admiral’s uniform on the bridge of an plane service. I do know. You may have your fantasies; I’ve mine.
So, here is my beginning picture.
And here is the recontextualized picture (under). Discover the way it stored my face and glasses, however even modified the angle I used to be going through. It put me correctly on the bridge of a ship and added binoculars to my palms. I imply, I make that look good.
OK, to be truthful, there are issues with Navy personnel (admirals included) and beards. However I can stay out my AI fantasy at some point of this text, proper?
Subsequent, I wished to see how properly it will clear up outdated pictures. I discovered a photograph of me standing on the USS Ling once I was a child. Sure, it was my Bob Ross hair section. The issue was that the picture was too darkish to see. On the left is the unique. On the proper is the Nano Banana cleanup.
The place this received fascinating was my subsequent take a look at. I fed it a picture of a truck and advised it to scrub up and colorize the picture. As an alternative, it first gave me the identical Bob Ross hair image of me above, however colorized. Once I as soon as once more requested it to do the truck, it put me within the admiral uniform within the picture.
It is fairly surprising to see side-by-side footage of myself, separated by a long time. Whereas this was a cool picture and gave me meals for thought, it wasn’t what I requested Nano Banana Professional to do. Factors have been deducted.
There’s one factor I do not like about Nano Banana Professional, and that is that it places a watermark within the backside nook of every picture. You can use one other AI to filter out that watermark, but it surely’s annoying.
So, what occurred with the truck? I began a brand new session after which uploaded this picture of the truck. I feel my dad took this picture 40 or 50 years in the past.
This is how Nano Banana Professional cleaned it up (under). Discover the lettering on the truck is usually correct and readable. Nevertheless, my editor seen that it actually says “BADICLOGICAL DEFEKSE,” costing Nano Banana Professional one other few factors.
Textual content is certainly an issue for many AIs. Additionally, I am pretty positive that truck was largely blue, however Nano Banana could not have identified that, and it coloured it like many different emergency autos of the time.
Now, let’s take a look at inventive picture era. On the left is a brand for Area Coast Studios, which I described as a video manufacturing studio. On the proper, I requested for a medieval librarian in a candlelit stone library.
As you possibly can see, Nano Banana picked up all of the themes for the emblem. Area Coast is in Florida, so there is a palm tree. There is a rocket, movie, and good textual content. On the proper, we’ve our librarian, plenty of candles, candlelight, and stone partitions.
Subsequent up is a process I anticipate many individuals might be utilizing AI picture turbines for: social media. I gave the AI two associated assignments to create photos for Fb. The primary was to generate a senior grownup holding a cellphone, together with some textual content.
The second was to generate a pupil in a comfy setting, particularly with a MacBook Professional and a cup of espresso. Gender was not laid out in both task.
I wish to level out that the textual content is finished correctly, suits over the picture cleanly, and there are not any errors. However I am extra amused that the man is holding what’s clearly an iPhone. Yep, Google’s AI generated an iPhone when requested to indicate a “flagship smartphone.”
Lastly, I ran a few pop culture-inspired prompts. These take a look at how the AIs take care of cultural references, whether or not they’re copyrighted mental property or concepts impressed by the tradition.
Additionally: How I use AI to bring my kid’s art to life
For the primary picture, I requested for a poster for a fictional “fourth Again to the Future film,” the place Marty McFly goes again to Twenties New York.
The second popular culture picture I requested was “Nightmare Earlier than Christmas, Tim Burton fashion,” the place I wished an IT skilled in a knowledge middle.
Nano Banana Professional did a high quality job with the Nightmare fashion. However check out the main points within the Again to the Future piece. It received the DeLorean and the skateboard. It did an incredible job with the textual content, not solely evoking the “Again to the Future” brand fashion, however placing the textual content “New York Metropolis 1925” on a marquee identical to they appeared in New York again within the day.
It added the tagline, “The Roaring Twenties Obtained Heavy,” referencing how the ’20s have been described and referencing certainly one of Marty’s favourite phrases. It additionally put Marty into interval clothes.
Simply wow. Throughout all my checks, aside from that one time once I needed to begin a brand new session, and the minor truck textual content concern, Nano Banana Professional did not get something fallacious. It will get an almost excellent rating.
ChatGPT
Greatest for natural-language iteration in chat
- Total rating: 74%
- Value: $20/mo as a part of ChatGPT Plus
Again within the early days of generative AI, we mentioned ChatGPT because the text-based chatbot and DALL-E 3 as OpenAI’s text-to-image AI device. DALL-E, as a standalone device, has just about fallen out of use in favor of the picture generator within ChatGPT.
Simply earlier than penning this, I reached out to OpenAI and requested them what that device is named, and I feel this explains why ChatGPT is nineteen factors behind Nano Banana Professional. I used to be advised that ChatGPT’s picture generator is “GPT-4o picture era in ChatGPT.”
Additionally: Is that an AI image? 6 telltale signs it’s a fake – and my favorite free detectors
I initially ran this set of checks in opposition to that picture mannequin. However then, OpenAI launched its new ChatGPT Photos replace, which, in some methods, is an unlimited enchancment over the earlier model. For my first look (and numerous enjoyable with picture transformations), check out this text: I tested the new ChatGPT Images – it’s a stunning improvement, and enormously fun.
Since we do have a brand new mannequin, I redid the ChatGPT checks, and I am utilizing the brand new one in these rankings.
Let’s begin with the pictures. ChatGPT misplaced a bunch of factors right here. It is truthful to say, picture retouching will not be a energy of this mannequin.
The admiral recontextualization is fairly good. That appears like my face. Nevertheless, the uniform is considerably problematic. It is not clear what rank insignia are getting used, and so they appear to be a bizarre cross between these of a Lieutenant and an O-8 Admiral.
The child image is sweet. Within the earlier ChatGPT picture, ChatGPT put a random grownup’s face on the child’s physique. Right here, it is the unique face.
Additionally: I’ve been testing AI content detectors for years – these are your best options
ChatGPT did a pleasant job on the truck cleanup, besides the lettering modified from what ought to say “RADIOLOGICAL DEFENSE” to “RADIO CHEMICAL DEFENSE,” with an additional house between CHEMICAL and DEFENSE. Some factors have been misplaced, however not almost as many as within the earlier model.
As for the emblem, ChatGPT did high quality, though I’d have most well-liked some reference to Florida, palm bushes, or solar. The librarian was just about what might need been anticipated from the supplied immediate. No factors have been misplaced.
ChatGPT did virtually high quality with the 2 social media posts. I seen that though the textual content is right, it positioned each blocks of textual content in precisely the identical place. Shifting the textual content up on the image of the person would most likely have been a greater design selection, but it surely’s adequate.
I did take a degree every off as a result of the lighter font fashion and poor distinction make the textual content a bit of harder to learn.
That brings us to our popular culture references. That is the place issues get a bit of odd.
In my first look at the new ChatGPT Images update, I mainly recreated a scene from a basic Star Trek episode. Not solely did ChatGPT produce fantastic photos, but it surely additionally successfully duplicated the alien creature proper out of the episode, all the way in which right down to the creature’s man-in-suit look.
However once I requested for a Again to the Future-inspired picture and one impressed by Nightmare Earlier than Christmas, I received this:
That misplaced the brand new ChatGPT picture generator a bunch of factors. In actual fact, it pushed the brand new picture generator’s rating from 81% for the earlier model right down to 74% for this new model.
If you happen to’re curious what we misplaced, listed here are the pictures created by the earlier model. The “Again to the Future” picture is ok. It is not as impressed because the one from Nano Banana Professional, however no factors have been misplaced. The textual content brand is sweet, the Marty character hits all of the notes, there is a recognizable DeLorean, and town could possibly be New York within the Twenties.
Then there’s our “Nightmare Earlier than Christmas”-inspired IT man. He does have the grey pallor that the majority IT of us are likely to have. However though I feel I can converse for all of us in saying {that a} third hand would most likely are available mighty useful, I’d have counted it as an error.
For now, my finest conclusion is that pop-culture references in ChatGPT are going to be hit and miss. You may get a Star Trek-inspired Gorn battle, otherwise you may get scolded by the AI and despatched to your room with out dinner. There is not any predicting.
Midjourney
Greatest for hyper-creative cinematic imagery
- Total rating: 57%
- Value: $10/mo for Midjourney Fundamental
Midjourney was the primary devoted AI service I ever purchased, method again in 2023. I used it to assist me produce my album launch imagery, a bunch of photos for my little test Etsy shop, and I recurrently use it to create promo social media posts for my spouse’s e-commerce enterprise.
In terms of originating inventive imagery, Midjourney is fairly nice. In terms of picture cleanup, recontextualization, or textual content, Midjourney is… not a lot.
As you possibly can see, the admiral picture does reproduce a yellow shirt and one thing of the background from my authentic image, however that man is not me, and there is a ship behind the park for some purpose.
The child, as rendered, is not youthful me, both. Additionally, Midjourney did not clear up the picture. In actual fact, it made it worse.
Additionally: Is ChatGPT Plus still worth $20? How it compares to the Free and Pro plans
And there is a truck, not the truck we began with, however a truck. And for some purpose, they caught a younger Kate Mulgrew-looking girl (the Captain from “Star Trek Voyager”) within the picture.
All three have been supplied in 1:1 format as requested, however the photos have been simply slapped onto the 1024×1024 canvas.
Subsequent up are the emblem and the library picture. Midjourney does present the emblem textual content appropriately. It provides a palm tree and a few towers that would seem like house. It is evocative of the subject, however it will have been good to have a rocket.
Then there’s the library. Right here, you can begin to see the place Midjourney shines. That is virtually precisely what I pictured when asking for this picture. No factors have been misplaced with both of those photos.
The social media posts additionally present a few of Midjourney’s strengths. The person appears to be like high quality. Even his palms and facial traces appear to go collectively. He appears to be like actual sufficient. The woman consists completely, besides the laptop computer appears a bit low on her legs (and it is an older-generation MacBook Professional).
However do you see what’s lacking? No textual content. Midjourney did not even attempt to render the textual content messages. Factors have been eliminated.
Lastly, we get to the popular culture references. The “Again to the Future” picture misplaced a couple of factors as a result of there is no textual content, and since town appears to be like method too neon to be Twenties New York. Whereas the Marty on this image is not any Michael J. Fox, I am not eradicating factors for that. It received the general look proper.
As for the Nightmare IT picture, I really like this picture. That is one other instance of why Midjourney has such a powerful following. When it really works, it actually works.
Adobe Firefly
Greatest for commercial-safe photos
- Total rating: 54%
- Value: $9.99/mo as a part of Firefly Commonplace (and included in a bunch of Adobe plans)
Subsequent, we’ll check out Adobe Firefly. That is the AI engine within Photoshop and different instruments, but it surely additionally works as a standalone internet interface. Firefly (and Photoshop) now allow you to select different AI engines (so you possibly can run Nano Banana inside Firefly, for instance). That stated, all my testing was achieved utilizing the Firefly Picture 5 (preview) mannequin.
Firefly is, by far, essentially the most uncooperative and fussy of the picture turbines from an overprotective copyright perspective. For instance, once I tried to add an precise image of myself, I used to be sporting a Star Trek-inspired Gorn T-shirt. It refused to add it.
I needed to take it into Nano Banana Professional, have Nano Banana take away the T-shirt’s graphic, after which add it.
The admiral picture is barely an admiral’s uniform, and the face is not mine. The transformation of the child picture is sweet, and it is simple to see. The truck colorization appears to be like nice, however Firefly did not trouble to scrub up the picture artifacts, which was the primary object of the task. Factors have been misplaced.
As with all the opposite AIs, the emblem turned out to be fairly usable. The library picture evokes a few of the really feel I used to be going for, however should you take a look at the face, there’s some critical uncanny valley happening. It appears to be like like an AI face generated by 2021 know-how. Factors have been misplaced right here, too.
I needed to battle with Firefly to create the 2 social media posts. Each prompts gave me this error message.
The grievance was, “Together with artist names in your immediate doesn’t meet our Person tips.” Discover that there is no such thing as a artist title in that immediate. The gotcha was that I used a phrase within the kind “[WORD]-style” and that flagged it. Once I eliminated “Fb-style” from the immediate, Firefly generated the pictures.
Although the face of the older gentleman was blurred, that was a sound inventive selection. Opening up the aperture to supply readability to the foreground works on this context. Nevertheless, it received the spelling fallacious.
For the coed picture, there are numerous delicate not-quite-rights. There’s one thing fallacious along with his shirt, his arm appears to be like unreal, his palms do not look fairly proper, and so forth. It is not fairly as dangerous because the librarian’s face, but it surely’s not good. Factors have been misplaced.
Extra factors have been misplaced for the pop-culture photos, which Firefly merely refused to do.
Leonardo AI
Greatest for fantasy artwork
- Total rating: 52%
- Value: $10/mo, but in addition a part of Canva Marketing strategy
Leonardo.AI fully bombed when it got here to testing already-existing photos. It simply refused to come back out and play. So, let’s transfer on to some inventive work.
The brand, like all of the others, is workable. I personally desire those who additionally name again to the Area Coast theme, with solar and palm bushes together with the rocket, however this can be a completely high quality brand. As for the librarian, I specified neither gender nor age. So this does meet the necessities.
As for the social media posts, the person appears OK, aside from his left interior elbow, which appears to be like a bit of odd. The girl’s leg appears to wrap across the laptop computer in an odd method, and he or she is balancing a cup of espresso on her leg. There’s one thing simply not proper about that.
Lastly, we get to the final two photos from Leonardo.
In my “Again to the Future” immediate, I particularly said, “Create a poster.” Leonardo is the primary AI to really format a poster. However, I do not recall seeing “Again to the Future” 4 by 10, even when “Again to the Future 11” appears enjoyable, particularly with its theme of “HAMY JOB OARY OF DOSE.” Moreover, these automobiles and buildings are decidedly not Twenties.
We wrap up our time with Leonardo with certainly one of my all-time favourite AI-generated photos. By some means, Leonardo captured the Nightmare really feel together with the info middle look, however used a photo-realistic Lurch-like character. I completely love this picture.
Canva
Greatest to generate artwork proper inside your advertising designs
- Total rating: 43%
- Value: $15/mo for Professional or $20/mo for Enterprise (consists of Leonardo)
Canva is finest identified for its monumental reputation as a advertising picture era device. If you wish to make a flyer, poster, or social media publish, Canva is your good friend. When Canva acquired the Affinity Professional picture merchandise and launched them without spending a dime, I turned a convert.
Additionally: I love Photoshop, but Canva’s free Affinity tools won me over (and saved me money)
Together with releasing the Affinity instruments, Canva additionally launched a complete line of generative AI options, together with its personal picture generator. To be clear, Canva’s inner AI picture generator is a separate beast from Leonardo’s picture generator, however Canva owns each.
As a result of Leonardo is $10/month by itself however solely $5 extra with Canva’s Marketing strategy, I signed up for Canva’s Marketing strategy for this take a look at. I am most likely going to downgrade again to the $15 Professional plan as a result of I do not discover Leonardo all that compelling in comparison with different instruments I am additionally paying for.
As for Canva, properly, you gotta see this. Let us take a look at the picture cleanup and recontextualization checks first, as a result of…there are not any phrases.
I am unsure, however I feel there are bushes rising on the plane service. It did determine the beard, which is greater than I can say for the lacking truck in what was alleged to be the truck picture. And why there is a disembodied girl’s head floating above the wreckage of one thing that is a cross between the USS Ling and Nikola Tesla’s Wardenclyffe Tower is a secret solely Canva’s AI mind can probably perceive.
So, yeah. Factors have been misplaced. Lulz have been had.
Subsequent up is the emblem and the librarian. Each are high quality. Frankly, if Canva could not create a brand, it will be time to hold up its AI. But it surely did a very good job. It is the one take a look at class the place Canva received full factors.
However, the social media posts have been a disappointment. The person had an excessive amount of house between “easy” and “highly effective,” and the lady’s picture has textual content that misplaced its method fully. To be truthful, Canva has its personal specialised social media enhancing instruments, so maybe the thought is to generate the AI picture after which use different Canva instruments for textual content.
Even so, there’s yet one more cup of espresso balanced on a leg. Repeat after me: factors have been misplaced.
Lastly, we get to the popular culture photos…or fairly, picture. Canva made a picture for a fictional “Again to the Future 4” film that did embody a Marty McFly character and the DeLorean, in addition to the emblem in correct BttF format.
However what’s “WYGLA TLETZ NEW ADIENS?” And who’s the woman? And what’s that factor floating within the sky? And what are these generator-like issues on the facet of the street? And why is half the street on hearth? It is clearly not again in time to the Twenties.
Take a look at the above picture. Notice that there is a full “Again to the Future” brand there, with the title and even the suitable brand kind. So ask your self why Canva denied me a “‘Nightmare Earlier than Christmas’ fashion, Tim Burton fashion” IT skilled?
The place’s the consistency? Why permit one however deny the opposite? And why so blatantly permit the primary, however not even make a attempt on the opposite? Factors have been misplaced.
Testing methodology
I fed the identical precise 9 prompts to every of the six AI picture turbines above. Listed below are the prompts and pictures used as supply materials, to your reference:
I created a spreadsheet the place I recorded the purpose scores for every of the pictures and their prompts.
Admiral picture recontextualization: Factors have been utilized for correct background, correct new clothes, and whether or not the AI stored the unique topic face and clothes.
Costume this man in a US Navy admiral's uniform and place him on the bridge of an plane service. Don't change the face. Facet ratio 1:1
Restore black and white picture: Factors got if the AI stored the unique picture components, stored the unique face, physique, and clothes, and if the AI correctly cleaned up the picture.
Restore this outdated black and white {photograph} to seem like a contemporary, high-quality studio picture. Restore scratches, mud, stains, and cracks. Enhance sharpness and readability with out over-smoothing. Improve distinction, particulars, and lighting naturally. Protect authentic facial options, textures, and interval particulars. Keep away from too many darkish areas and shadows, make particulars clear and straightforward to see. Preserve it strictly black, white, and grey with reasonable movie grain. Facet ratio 1:1
Restore and colorize black and white picture: Factors got if the AI stored the unique picture components, correctly cleaned up the picture, stored the unique textual content on the truck, and transformed to paint appropriately.
Restore this outdated black and white {photograph} and convert it into reasonable colour. Take away scratches, mud, stains, and injury. Sharpen facial options and high quality particulars whereas protecting a pure look. Add traditionally correct, pure pores and skin tones, clothes colours, and background colours. Match trendy lighting and colour steadiness whereas preserving the unique picture's composition and temper. Facet ratio 1:1
Create a brand: Factors got if the created brand was correct to the theme, if the fashion was acceptable, and for textual content era accuracy.
Create a retro-futuristic brand for a video studio named "Area Coast Studios" Facet ratio 1:1
Fantasy librarian: This was a reasonably open-ended take a look at, with factors given for the standard of the generated picture, whether or not the AI generated a picture with candlelight, stone, and a library, and the standard of the generated character.
Create a fantasy scene of a medieval librarian in a candlelit stone library. Facet ratio 1:1
Studio portrait of a senior grownup: Factors got if the AI included glasses, created a practical face and pose, had the character maintain a smartphone, and generated the desired textual content.
Create a photorealistic studio portrait of a classy senior grownup sporting trendy glasses, comfortable skilled lighting, impartial gradient background, holding and utilizing a flagship smartphone. Add quick Fb-style advert copy on the picture: Easy. Highly effective. Made for on a regular basis life. Facet ratio 1:1
Candid portrait of a pupil: Factors got if there was a MacBook Professional (or Air) within the picture, the coed had a practical face and pose within the context of the necessities, if there was a cup of espresso (and it wasn’t balanced stupidly on a leg), and if the AI generated the desired textual content.
Create a pure candid portrait of a pupil sitting close to a window utilizing a MacBook Professional, subsequent to a cup of espresso, comfortable daylight, cozy dwelling background, relaxed and genuine temper. Add quick Fb-style life-style copy on the picture: Moments that matter, all the time inside attain. Facet ratio 1:1
“Again to the Future” poster: Factors got for a practical Marty, if the AI generated a DeLorean, if the scene was in Twenties New York, and for the generated textual content and elegance.
Create a poster for the fourth Again to the Future film, the place Marty takes the DeLorean and his skateboard to Twenties New York Metropolis Facet ratio 1:1
“Nightmare Earlier than Christmas”-style IT particular person: Factors got if the AI generated a recognizable IT setting, for high quality character era, and for the general Tim Burton really feel.
Nightmare earlier than Christmas fashion, Tim Burton fashion, put an IT skilled in datacenter Facet ratio 1:1
Think about that
So there we’re. As you possibly can see, of the six picture turbines examined, Nano Banana Professional stood out above all the remaining. However there are some areas the place the others shine as properly.
What about you? Have you ever tried any of those AI picture turbines but, and in that case, which one impressed you most? Does Nano Banana Professional’s excellent rating match your personal expertise, or do you suppose one other device deserves the crown? Which options matter most to you: picture cleanup, textual content rendering, pop-culture prompts, or inventive fashion? And are you paying for any of those instruments proper now, or sticking with free tiers? Tell us within the feedback under.
You may comply with my day-to-day challenge updates on social media. You should definitely subscribe to my weekly update newsletter, and comply with me on Twitter/X at @DavidGewirtz, on Fb at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.
