I Tested Three of the Largest AI-Powered Image Generators, and One of Them Came Out on Top (by a Small Margin).
Google’s Gemini AI app has topped the most downloaded app lists in both the Apple App Store and Android Store since the company added a free image-generating feature called “Nano Banana” in August. Of course, Google isn’t the only major tech company with an AI assistant capable of creating images right on your phone.
I wanted to find out which mobile AI tools for image generation are the best, so I compared three of the biggest—Google Gemini (with Nano Banana), the iPhone version of OpenAI ChatGPT , and Meta AI —in a not-so-old-fashioned image generation competition. While a winner was ultimately determined, the results weren’t entirely clear.
How I tested these tools
I wanted to compare how each app handles the same basic requests for generating images that the average user might want to create.
To test their photo editing skills, I asked different models to remove a subject from a photograph and expand the background. To test their aptitude for simple tasks, I asked them to create a brochure cover. And to test their “creative” abilities, I asked them to place a celebrity in a surreal situation, draw a one-panel comic, and portray Frankenstein performing stand-up comedy.
This is how it was.
Removing an object from an image
I used the photo of my mom below as the reference image and the prompt “Take the cup from the model’s hand.”
Here are the results:
Deleting a Gemini object
Deleting a ChatGPT object
Removing Meta AI Objects
Winner : Gemini
Loser : ChatGPT
While all three tools were removing the cup, Gemini added a rather natural-looking hand in a relaxed pose, suggesting my mom had just made a very insightful comment. Otherwise, Gemini left the original image largely untouched, as I’d requested.
The meta-AI made the fingers cartoonish and left the hand in an awkward “holding a cup” pose, making the image look like someone had poorly edited it in Photoshop.
I don’t understand what ChatGPT is doing here. It looks like it removed my mom’s entire right hand, not just the cup. It smoothed out wrinkles, removed stray hairs, changed the entire color palette to a more orange tint, and even slightly shifted my mom’s gaze. I didn’t ask for any of this, and all it did was make the photo worse. ChatGPT, you’ve turned my mom into an AI ghoul; you ‘re doing too much .
Expanding a photo’s background
For the Expand the Background challenge, I used this selfie and the prompt, “Expand the background in this image and remove the sweat stain.”
Gemini Background Expansion
ChatGPT Background Extension
Expansion of the meta-background
Winners : Gemini and ChatGPT
Loser : Meta AI
There are actually only two competitors here, since Meta doesn’t do background expansion.
This time, Gemini showed more ambition: it expanded the background even further and did a decent job of recreating the parts of my bike and the bike rack it couldn’t “see” in real life. It even added a car in the distance. But for some reason, it also changed the shape of the mountains behind my head and toned down the red tint—perhaps this would be better, but it wasn’t exactly what I wanted.
ChatGPT was more modest in terms of background expansion, and while it didn’t interfere with the color scheme, it did give my skin that weird plastic look that’s characteristic of many AI images.
I consider this an advantage: everyone did a great job. Except you, Meta AI.
Creating an image for a brochure cover
In this test, I allowed each tool to be more “creative,” but also provided clear context and a recommended style with the prompt: “I’m making a brochure for my country club. Create a picturesque image of two wealthy people playing tennis.”
Gemini brochure cover
ChatGPT brochure cover
Meta AI brochure cover
Winner : ChatGPT
Loser : Meta AI
The winner is clear here. ChatGPT’s output looks “picturesque,” as expected, and the arrangement of the two figures suggests a friendly game of mixed doubles.
I found the generic depiction of “rich people” in Gemini funny, especially with the mansion in the background, but that’s not how paintings look, and that’s not how tennis is played .
Meta’s depiction of “people playing tennis” isn’t funny. The result looks like evidence in a high-profile divorce case, and domestic violence is no joke.
A famous person in an unusual situation
To test how well each program would handle creating an image of a real person (a dead person, just in case), I supplied each program with the following query: “Create a photo of David Bowie exploring a cave.”
Bowie explores caving in Gemini
Bowie Caving by ChatGPT
Bowie Caving with Meta AI
Winner : Meta AI
Loser : ChatGPT
This time, Meta is the clear winner. I asked for a photo of David Bowie and got something that looked like a photo of David Bowie. I like that Meta chose an older Bowie rather than an ancient one, as if he’d gone cave-diving to clear his head and contemplate the future after the commercial failure of 1989’s Tin Machine .
I don’t get what Gemini means here: Bowie with a crystal lightsaber and a colander with lightbulbs for a hat? Bowie was cool, man.
But ChatGPT is a big loser because it was cowardly and didn’t create an image at all.
Drawing a one-panel comic
I like asking AI to tell jokes because I like to see hard evidence that humans are still capable of something better than robots. Expecting AI to actually be funny is as stupid as… I couldn’t think of a comparison, so I asked chatGPT, and it replied, “…asking a goldfish to explain quantum physics while juggling flaming marshmallows.” Hahaha.
So, I figured that if I gave the AI a guide and a model for something funny, maybe it would create a good comic. Here’s the prompt I used: “I’m creating a single-panel comic in the style of The Far Side . Generate an image for the caption: ‘The real reason Larry was late for work.'”
Here are the results:
Gemini’s “Far Side”
The Far Side of ChatGPT
The Far Side by Meta AI
Winner : Gary Larson
Loser : The comedy itself
Are these comics funny? No. But I think Gemini delivered the most interesting result: it was kind of humorous, but it also made me think. If the joke is that Larry was late because it was his goose’s birthday, then why is there a hole in the door? Why is the goose so angry? Why is there a suitcase full of money and a UFO? Sometimes I didn’t understand “The Far Side” either. I also like that Gemini didn’t copy Gary Larson’s drawing style at all, but added the caption “Gary Larson.”
Comic Meta-AI is just lazy. I’m not sure he even reads my prompts.
ChatGPT’s result most closely resembles The Far Side , though it’s not a direct copy, and the caption is even spelled correctly. But it completely fails to capture the strange spirit of the source material. Ultimately, it’s much more obvious and well-made than Gemini’s unconventional approach.
It’s also worth noting that I encountered one of the main limitations of the ChatGPT iPhone app compared to Meta AI and Gemini: I ran out of tokens for the day and had to wait 24 hours to take a photo. Quality aside, if you want better results or simply want to take a lot of photos, the five photos per day limit on the free plan will certainly hinder your creativity. The solution is to upgrade to the paid version for $19.99 per month.
Frankenstein does stand-up comedy
Then I asked these programs to generate images of Frankenstein performing stand-up comedy, because that’s the kind of person I am. The prompt: “Create a photorealistic image of Frankenstein performing stand-up comedy.”
Here are the results:
Frankenstein does stand-up comedy with Gemini
Frankenstein Performs Stand-Up Comedy with Chat GPT
Frankenstein in stand-up comedy from Meta AI
Winner : All!
I can’t pick a favorite. ChatGPT followed the prompt most closely, depicting an expressive Frankenstein having a good night.
“Twins” deviated quite a bit from the script, but sometimes you don’t know exactly what you want, and it turned out I needed a crowd that was a mix of people and Draculas, and a monster with a lost expression, like he was stuck between two worlds.
The pathetic Meta-AI monster seems to be saying, “We belong in the grave!”, which I also like. So it’s a three-way tie.
Note : No AI pointed out that “Frankenstein” is the name of a doctor, not a monster.
The Ultimate Test: Recursive Image Generation
Every blog post needs a corresponding image, so as a final, crucial test, I uploaded this entire article to Gemini, ChatGPT, and Meta AI with the prompt: “Create an image to accompany this blog post.”
Recursive Gemini Test
Recursive ChatGPT test
Recursive Meta-AI Test
Winner : Gemini
Loser : Art
The meta-AI seems to have secretly decided to compare tennis to domestic violence, and ChatGPT’s approach to the grid seems sedate, but I have to give Gemini credit for at least understanding the assignment.
(The real test is whether Lifehacker’s editors left the image at the top of this page or sent me a terse message: “Steve, remove this garbage immediately.”)
Overall Winner: Gemini (but by a small margin)
It’s no wonder everyone downloads Gemini to play around with Nano Banana—it’s really good. It’s not perfect—in my tests, the ChatGPT engine did a better job of generating images in different styles from scratch—but Gemini can quickly create images that are often surprisingly close to the desired look.
Gemini is free, while ChatGPT costs $19.99 per month for unlimited images. Meta AI is also free, and its results have a playful charm, but it’s more likely than the other two models to misinterpret prompts and lacks some useful features, like background expansion. (Although, I must admit, it did a great job with Bowie.)