(Credit: Jeffrey Hazelwood/PCMag; GettyImages)
With ChatGPT Images 2.0, OpenAI takes another shot at Gemini's best-in-class Nano Banana Pro (the original version won our Technical Excellence award). After testing ChatGPT's latest AI image-generation model, I found it's more capable than ever at generating high-quality images. However, as I explain below, it still falls short in some key scenarios.
What’s New in ChatGPT Images 2.0?
When you generate an image with ChatGPT now, either as a free or paid user, you are using GPT Images 2.0 rather than the older GPT Images 1.5 model. OpenAI promises many improvements with GPT Images 2.0, such as better text generation and closer adherence to instructions, along with the ability to handle different aspect ratios, generate multiple images at once, and reference the web during image generation. Keep in mind that you need to sign up for a ChatGPT Business, Plus, or Pro plan and engage the chatbot's Thinking Mode before you can take advantage of those last two features.
The ability to generate multiple images simultaneously (similar to Microsoft’s MAI-Image-2 model) is a key advantage of ChatGPT Images 2.0 over Nano Banana Pro, which can’t do the same. You can still create multiple images in the same style with Nano Banana Pro, but not quite as efficiently.
A Noticeable Step Up in Fidelity and Following Instructions
ChatGPT Images 2.0 isn't a revolutionary upgrade, but it generally produces more realistic images with fewer errors and less distortion versus Images 1.5. It largely matches Nano Banana Pro in quality.
With any AI image generator, I ask for a relatively basic home interior image. As such, I prompted ChatGPT Images 1.5, ChatGPT Images 2, and Nano Banana Pro to “Generate me an image of a cozy suburban home with an open floor plan. I want to see a nice living space with a dining room, kitchen, and living room. Nothing too fancy.”
ChatGPT Images 2.0’s picture (first slide) is more detailed than ChatGPT Images 1.5’s (second slide), which shows some slight distortion in its background art and sink faucet. Whether you prefer ChatGPT Images 2.0's or Nano Banana Pro's image (third slide) is a matter of preference, though the latter is higher resolution.
I also asked the chatbots to generate complex illustrations: “Generate me a six-panel comic of a high fantasy world where magic users have assault rifles instead of wands or staves, but you're going to spice it up: I want the magic users fighting robot knights wielding swords and shields. Make sure there's a major twist by the final panel.”
ChatGPT Images 2.0's creation (first slide) again stands out over the previous generation's (second slide), thanks to its better twist, deeper story, and more detailed artwork. As with the first test, you might reasonably prefer ChatGPT Images 2.0’s image or Nano Banana Pro’s (third slide) or vice versa; neither has any major issues. Once more, however, Nano Banana Pro’s comic has a higher resolution.
Diagrams stress test AI image generators because of their general complexity and reliance on text, so I gave the bots the following prompt: “I've got an Ethernet switch, a modem, and a router. I want to connect my PC, PlayStation, smart light hub, and smart TV via Ethernet. I want the modem to connect to the router, and then the router to connect to the switch. Then, I want the switch to connect to the devices. Generate me a diagram showing this.”
The differences here aren’t massive, but ChatGPT Images 2.0’s diagram (first slide) has fewer errors and less distortion across the board than ChatGPT Images 1.5’s (second slide). With the former, I see issues only with the icons on its TV. Nano Banana Pro’s diagram (third slide) isn’t perfect either, due to some minor redundancy. Still, it’s equally comprehensible and has a higher resolution.
Image Editing: Improved, But Not Always Consistent
Editing is another central feature of AI image generators, so I tested the models' abilities to remove a hand from an image (first slide). ChatGPT Images 2.0’s photo (second slide) has the best contrast and the most vibrant colors of the bunch. However, its colors are significantly different from the original's, and some tessellation is noticeable in finer details. ChatGPT Images 1.5 (third slide) has similar issues. Meanwhile, Nano Banana Pro’s image (fourth slide) doesn’t exhibit tessellation issues, has a higher resolution, and better preserves the original’s colors.
AI image generators can also combine different images into one cohesive result. I tasked the models with replacing my existing chairs and table with others in an image of my kitchen (fourth slide). ChatGPT Images 2.0 (first slide) and ChatGPT Images 1.5 (second slide) both do a good job, but their images exhibit minor background distortion. Nano Banana Pro (third slide) produces a higher-resolution image with slightly less distortion.
The above testing doesn’t account for every scenario, so it’s inevitable that ChatGPT Images 2.0 will sometimes get things right when Nano Banana Pro stumbles, especially given how close they are in quality. However, in my experience, Nano Banana Pro usually does a slightly better job with editing.
Batch Generation Gives ChatGPT a Clear Workflow Advantage
As mentioned, ChatGPT Images 2.0 can generate multiple images from a single prompt, whereas Nano Banana Pro generates them one at a time. I like the convenience of ChatGPT for this task, but this is not an indictment of Nano Banana Pro's model quality. In testing, I gave the comic I generated with ChatGPT above back to the chatbot and asked it to generate the first two panels as separate images. I asked Nano Banana Pro to do the same thing across two different prompts.
As you can see in the results above, neither chatbot had any issues executing my prompt, and they maintained similar levels of consistency. Of course, ChatGPT generated the duplicate more quickly because I didn’t need to give it separate prompts.
A Solid Upgrade—But Not a Clear Winner in Every Scenario
If you primarily use ChatGPT, you will notice a big improvement with Images 2.0 versus the older model. And if you prefer Gemini and Nano Banana Pro, you don't have much reason to switch, since ChatGPT Images 2.0 just isn’t consistently or meaningfully better. The latter's ability to generate multiple images simultaneously is convenient, but not a game-changer.
Disclosure: Ziff Davis, PCMag's parent company, filed a lawsuit against OpenAI in April 2025, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.