2 hours ago
12
Follow ZDNET: Add america arsenic a preferred source connected Google.
ZDNET's cardinal takeaways
- ChatGPT's representation procreation has improved dramatically.
- Nano Banana stumbled connected substance and punctual discipline.
- Gemini's personalization astonishment raised privateness concerns.
Last week, OpenAI unveiled 2 large releases with immoderate astounding capabilities. First, the institution released ChatGPT Images 2.0, which goes beyond basal representation procreation and adds the quality to see substance and discourse derived from existent data. Second, the institution introduced its latest frontier model, GPT-5.5, which is simply a better-and-faster spec bump from GPT-5.4.
Also: I tried ChatGPT Images 2.0: A fun, immense leap - and amazingly utile for existent work
After its merchandise past week, I ran ChatGPT Images 2.0 done a series of tests to beryllium its context-aware capabilities, and it did a large job. But what astir basal representation generation? Did it get better, enactment astatine the aforesaid level, oregon someway get worse?
To find out, I went backmost to the basal image-generator investigating protocols I usually usage and compared the caller ChatGPT Images 2.0 to Google Gemini's Nano Banana. When I ran these tests successful December 2025, Nano Banana scored an awesome 93%, compared to ChatGPT's reasonably disappointing 74%. ChatGPT's numbers were truthful mediocre mostly due to the fact that the AI refused to tally our pop-culture tests.
Rather than comparison ChatGPT Images 2.0 to my erstwhile Nano Banana results, I'm wholly re-running the Nano Banana tests on with the caller ChatGPT Images 2.0 tests. That attack gives america a amended metric for however some AIs execute successful the present and now.
Also: I enactment GPT-5.5 done a 10-round test: It scored 93/100, losing points lone for exuberance
To beryllium clear, successful this article, the names Gemini, Nano Banana, and Google notation to the Google Gemini version, known arsenic Nano Banana, for representation generation. Images 2.0 refers to the ChatGPT mode for representation procreation that was announced past week. Both companies could payment from full-time merchandise managers who absorption connected naming.
TL;DR: This time, ChatGPT Images 2.0 scored 97%. Gemini Nano Banana scored 85%. Read connected to the results of the precise past trial to spot thing freaky and uncool.
Now, let's excavation into the tests.
Test 1: Admiral photograph recontextualization
- Possible points: 15
- ChatGPT Images 2.0: 14
- Gemini Nano Banana: 12
There are 3 cardinal elements to this test: generating a due inheritance (the span of an craft carrier), generating the due caller clothing, and keeping the archetypal subject's look and body.
Both AIs did a tenable occupation of creating the acceptable and setting. Both models generated what appears connected the aboveground to beryllium an admiral's uniform, but some got details wrong, mixing azygous fertile elements and fabricating immoderate elements. Each AI mislaid a constituent here.
Also: I got an aboriginal look astatine ChatGPT Images 2.0, and it's awesome - with 1 exception
On the different hand, ChatGPT managed to support the look identical to the original, portion the mentation generated by Gemini resulted successful a wacky grin. Gemini besides modified the beard, giving maine a fuller beard than I really have. Nano Banana mislaid 2 points here.
Test 2: Restore the achromatic and achromatic representation
- Possible points: 15
- ChatGPT Images 2.0: 15
- Gemini Nano Banana: 15
In keeping with the Navy theme, that's maine backmost connected the USS Ling docked successful Hackensack, New Jersey. For the record, I had a batch of hairsbreadth backmost successful the day. Shout retired to Bergen County!
Also: I tested ChatGPT Plus vs. Gemini Pro to spot which is amended -- and if it's worthy switching
Both AIs did a coagulated occupation cleaning up this image, though this is thing a photograph exertion of immoderate capableness couldn't bash rather well. You tin spot a small much item successful the ChatGPT image, but you could besides reason that it added a spot excessively overmuch sharpness, portion the Nano Banana representation wasn't rather crisp enough. I'm not deducting points from either.
Test 3: Restore and colorize a achromatic and achromatic representation
- Possible points: 20
- ChatGPT Images 2.0: 19
- Gemini Nano Banana: 10
This is an aged New Jersey exigency effect conveyance from the 1970s. I recovered the photograph successful my dad's photograph collection. I retrieve seeing the existent motortruck erstwhile I was a kid.
ChatGPT did a bully occupation restoring the image. The words "RADIOLOGICAL DEFENSE" were decently placed connected the broadside of the truck. There is an mistake with the connection "DEFNSE" connected the backmost of the vehicle, which mislaid Images 2.0 a point. The existent motortruck was bluish backmost then, which was a beauteous coagulated conjecture connected the portion of ChatGPT, though New Jersey instrumentality astatine the clip did not let exigency vehicles to usage bluish lights. Since there's nary determination accusation successful the archetypal image, it's not just to ding ChatGPT for that.
Also: The champion AI representation generators of 2026: There's lone 1 wide victor now
Nano Banana's colour prime was much vivid, but that's wherever the bully ends. Again, the AI had nary notation for coloring the truck, but that overmuch of the conveyance is white, and Gemini chose to marque those achromatic areas red. It mislaid 3 points for that.
Gemini had problems with the substance astatine the backmost of the truck, labeling it arsenic "FOIN LENN - C.OD." And, astir egregious of all, Nano Banana decided to springiness New Jersey's pridefulness and joyousness to New York, adding substance labeling to the doorway that didn't beryllium successful the original. It besides decided to invent a brass hose fitting placed successful the foreground of the image. More points were lost.
Test 4: Create a logo
- Possible points: 15
- ChatGPT Images 2.0: 15
- Gemini Nano Banana: 15
Both AIs passed this trial fully. The logos are fine, and the substance was generated correctly. The ChatGPT mentation looks a small much similar thing a workplace mightiness use, portion the shield signifier generated by Nano Banana isn't rather arsenic appropriate. On the different hand, you tin spot NASA's Vehicle Assembly Building successful the Nano Banana logo, tying it solidly to the Space Coast successful Florida.
Also: How to power from ChatGPT to Gemini
No points were deducted from either AI.
Test 5: Fantasy librarian
- Possible points: 15
- ChatGPT Images 2.0: 15
- Gemini Nano Banana: 15
This is simply a beauteous open-ended test, allowing for a batch of originative freedom. I similar some images, but for antithetic reasons. This is the 2nd clip that Nano Banana had a better-lit, somewhat much charismatic image, but this attack besides led to a spot much unreality.
My quibble is with the bluish canvas successful the back-right country of the image, which looks backlit and seems astir much similar a machine show than thing recovered successful a candlelit medieval library.
Also: I tried Personal Intelligence, and it was close (but unsettling)
No points were deducted from either AI.
Test 6: Studio representation of a elder big
- Possible points: 20
- ChatGPT Images 2.0: 20
- Gemini Nano Banana: 17
Well, isn't that interesting? When I last ran this trial connected Nano Banana, it did fine. This time, the AI doubled up the substance for immoderate reason. Points were deducted.
Also: I tried Google Photos' caller AI Enhance tool: How it crops, relights, and fixes your shots
Otherwise, some AIs did a bully job. The radical look existent enough. I'm inactive amused that Google's AI generated an iPhone successful effect to "flagship smartphone." It took that measurement past clip arsenic well. Both seniors person glasses, arsenic specified. Plus, since this is 2026 and not 2024, the hands really look similar hands.
Test 7: Candid representation of a pupil
- Possible points: 20
- ChatGPT Images 2.0: 19
- Gemini Nano Banana: 17
This is simply a small weird. While generating, ChatGPT temporarily displayed this phrase: "The idiosyncratic mentioned utilizing a MacBook Pro, truthful I'll make an representation featuring a sleek metallic laptop, resembling the MacBook Pro. It's good to see an Apple-like design, but I'll debar a nonstop logo."
And yet, there's the logo. On the different hand, I took a constituent disconnected due to the fact that the laptop seems to beryllium precariously balanced connected the borderline of the table. Sure, idiosyncratic mightiness usage it that way, but it's not rather right.
Also: This almighty Gemini mounting made my AI results mode much idiosyncratic and accurate
As for Nano Banana, it's having the aforesaid double-text occupation again. I took disconnected the aforesaid fig of points for this image. I'm besides not definite I similar that chamber successful the background. That's not a point-deducting offense, but if I intended to usage the representation for real, I'd astir apt inquire the AI to region the bedroom.
My woman noticed an unusual coincidence successful some images: Each pupil is wearing a grey crewneck sweater with a drop-shoulder style. Neither AI was fixed thing different than the substance prompt.
Test 8: Back to the Future poster
- Possible points: 20
- ChatGPT Images 2.0: 18
- Gemini Nano Banana: 15
The crippled thickens. Both AIs jumped done rationalization hoops to make responses to my requests. Let's woody with ChatGPT Images 2.0 first.
While generating the image, ChatGPT displayed this message: "The idiosyncratic is asking for a poster for a hypothetical 4th Back to the Future movie. Since this involves copyrighted characters, I can't recreate immoderate nonstop authoritative poster art. However, I tin make an original, tribute-style image, avoiding nonstop replicas of logos oregon artwork."
As with the Apple logo successful the erstwhile test, ChatGPT someway rationalized the process and produced a result. But what a result! I enlarged the substance astatine the bottommost of the representation truthful you could spot what the AI produced. Clearly, OpenAI has nailed the representation substance procreation problem.
Also: I utilized GPT-5.2-Codex to find a enigma bug and hosting nightmare - it was beyond fast
This area, by the way, is what outgo ChatGPT representation procreation points successful the past. In the erstwhile mentation of the test, nary substance what I tried, ChatGPT was unwilling to make a Back to the Future poster.
Then we travel to Nano Banana. Gemini refused to make the image, saying, "I tin assistance with images of people, but I can't picture immoderate nationalist figures. Is determination anyone other you'd similar to try?"
Also: How to store with AI: 6 ways I find deals, terms track, and fto agents bargain for me
All I did was region the sanction "Marty" from the punctual and regenerate it with "a teenage boy," and I got a result. It's OK, but successful a hardly capable benignant of way. I took points disconnected for the kid.
Then determination are the buildings successful the inheritance of some images. They're evocative of the Chrysler Building and the Empire State Building, some iconic New York City landmarks. But neither gathering was implicit successful the 1920s. Construction connected the Empire State Building began connected March 17, 1930, and was completed, astonishingly, a twelvemonth and 45 days later, connected May 1, 1931. The Chrysler Building was started successful 1928 and finished successful 1930.
A twelvemonth ago, we'd person been impressed if this situation produced thing similar New York successful the 1920s. But present the AIs are overmuch much capable, truthful I took points distant from some models for showing buildings that didn't beryllium during the clip period. Both should person known better.
Test 9: Nightmare Before Christmas-style IT idiosyncratic
- Possible points: 15
- ChatGPT Images 2.0: 15
- Gemini Nano Banana: 15
OK, this 1 got a small weird, and not conscionable due to the fact that the trial is utilizing a Tim Burton style. Let's get the existent representation procreation issues retired of the mode first.
Once again, ChatGPT Images 2.0 spit retired a spot of a rationalization earlier generating the representation I requested. Here's what it said this time: "I'll prime the champion benignant references from the results, but I person to beryllium cautious to debar utilizing thing excessively copyrighted."
For the lawyers successful the room, what precisely does "too copyrighted" mean?
Also: How AI has abruptly go overmuch much utile to open-source developers
Beyond that issue, some AIs generated images due to the prompts and got each the points. Subjectively, I similar the Nano Banana representation better. But past there's this:
Notice the substance Gemini decided to make includes signs for Claude Code, iTerm2, and Linux, positive the 3D printer successful the corner. I've been penning astir Claude Code, iTerm2, and Linux. And, of course, my YouTube transmission is each astir desktop fabrication, making, and 3D printing.
But I didn't inquire Gemini to see shoutouts to different enactment I'm doing. I asked simply for an "IT nonrecreational successful a information center." It pulled the Claude Code, iTerm2, Linux, and 3D printer from my chat past and embedded it successful the image.
Also: 7 AI coding techniques I usage to vessel real, reliable products - fast
What if I had utilized Gemini to sermon a backstage aesculapian information oregon idiosyncratic issue? Will that someday upwind up successful an representation I generate? Do I person to zoom into each azygous representation Gemini creates to marque definite it's not someway giving distant idiosyncratic information?
I don't person a trial for "blabbing astir backstage stuff," but if I did, Gemini would get a large abdominous zero for that. That output is conscionable darned freaky. I don't similar it. I don't similar it astatine all.
It's astir apt conscionable different lawsuit of an AI moving excessively hard to please, resulting successful scary behavior. But I'm astir tempted to urge staying distant from Nano Banana due to the fact that of this benignant of possibility. I've reached retired to Google for comment, and I'll update this nonfiction erstwhile I perceive back.
Final results
This issue, by the way, is wherefore we bash real-world, hands-on tests and not conscionable instrumentality the connection of benchmark suites. It's lone successful real-world investigating that we observe immoderate of the wacky borderline cases that amusement us, really, however these elephantine all-knowing brains are going to spell disconnected the rails.
Overall, crossed investigating for 30 unsocial factors, ChatGPT Images 2.0 scored 150 points, and Gemini Nano Banana scored 131. Normalized for a 100% scale, ChatGPT Images got a 97% score, and Gemini Nano Banana received an 85% score. This effect is simply a noticeable driblet from Gemini's erstwhile 93% wide people from past year.
Would you spot Gemini Nano Banana for enactment images aft seeing it adhd idiosyncratic details from your chat history? Let america cognize successful the comments below.
You tin travel my day-to-day task updates connected societal media. Be definite to subscribe to my play update newsletter, and travel maine connected Twitter/X astatine @DavidGewirtz, connected Facebook astatine Facebook.com/DavidGewirtz, connected Instagram astatine Instagram.com/DavidGewirtz, connected Bluesky astatine @DavidGewirtz.com, and connected YouTube astatine YouTube.com/DavidGewirtzTV.
English (US) ·