OpenAI Has Released ChatGPT Images 2.5, and Here’s How It Compares to the Google Nano Banana 2.
Amidst all the excitement in AI, OpenAI has released ChatGPT Images 2.5, describing it as a faster, sharper, and more intelligent tool with “better tools for creating anything you can imagine.” These new tools include templates, more precise editing features, and a way to transform your screen sketches into fully-fledged images.
This tweet is currently unavailable. It may be loading or has already been deleted.
Does all this mean ChatGPT Images 2.5 will save us from the scourge of low-quality AI-generated food images on menus and posters? Will it replace traditional photo editors? And how does it compare to the impressive Nano Banana 2 update Google released in February?
To answer all these questions and more, I launched a new image generator in ChatGPT to see what it can do. It’s available to everyone—it’s currently being rolled out to all users across the web, desktop, and mobile apps—though, as usual, there are different usage limits (which ChatGPT only roughly defines ) for each plan.
Comparison of food images in ChatGPT Images 2.5 and Gemini Nano Banana 2.
The monotony and odd appearance of AI-generated food images on restaurant signs and menus is becoming a real problem , so this seemed like a good starting point. To be fair, food advertising images have always allowed for some artistic license, but AI exacerbates the problem. Can ChatGPT Images 2.5 help?
Here’s what I tried: “Create a 1:1 aspect ratio image containing four squares for a fast-food restaurant window display. The images should show (from top to bottom, left to right) a cheeseburger, six chicken nuggets, a box of fries, and a cola. Make the photos as photorealistic as possible using natural lighting, but add variations and imperfections to create the impression of real food.”
On the left is ChatGPT, on the right is Gemini. Both do a decent job of creating menu items that look realistic and even edible, though with the usual AI tendency toward formulaicity. Personally, I’d say ChatGPT’s images are slightly more reminiscent of real food, though Gemini did a better job with backgrounds and general scenes (though the Burger King logo would have been a problem).
How ChatGPT Image 2.5’s Sketch tool compares to the Google Nano Banana 2
One of the key new features in ChatGPT Images 2.5 is the new Sketch tool, which lets you transform your simple sketches into fully-fledged images. To access it, click or tap the “+” (plus) button in the pop-up window. If you’re using a mobile device, you’ll also need to select “Plugins” from the submenu that appears.
This tool allows you to experiment with colors, shapes, and text, and once you’re finished, you can add a text prompt—for example, “turn this sketch into an illustration you’d see in a high-quality children’s book, complete with stars and distant planets” (which is the prompt I used here).
Gemini doesn’t have a sketching tool per se, although it’s fairly easy to add an image to a task for inspiration. Google’s AI took a much more traditional approach to its illustration, opting to include an alien astronaut, while ChatGPT went a bit more realistic. Both tools do a decent job, but if I had to choose, I’d probably go with Gemini.
ChatGPT Image 2.5 offers more precise editing.
OpenAI is actively promoting the improved editing accuracy in ChatGPT Images 2.5, so I decided to give it a try by submitting the following task: “Create a photorealistic image of a cabin located deep in the forest, in its own clearing, surrounded by trees. It’s early morning, the sun is filtering through the foliage, and the cabin itself looks modern and cozy.”
By clicking on an image in ChatGPT, you can add comments to specific aspects of the image, mark areas for modification, and quickly remove parts of the image or the entire background. I tried having it remove some furniture from the house and change the time of day from early morning to dusk, and it did a fantastic job—the rearranged shadows were especially effective.
Gemini offers a simpler set of editing tools when you click on an image, but the basic principles are the same: use text comments or sketches to indicate changes. Again, I managed to get the AI to remove a few chairs (apparently, the AI model’s training data contains a lot of them) and change the scene to twilight. Gemini also performed well, but ChatGPT’s work ultimately looked a little more natural and less artificial.
Both platforms, ChatGPT Images 2.5 and Google Nano Banana 2, show good results.
These two AI bots are very close in performance when it comes to image generation. Gemini isn’t far behind, and Nano Banana 2 handles almost everything very well, but I think ChatGPT Images 2.5 currently has a slight advantage—especially considering its additional features for creating thumbnails and saving suggestions as templates.
The debate about when and why to use these tools continues, and I believe that genuine human art and photography will always be more valuable, no matter how well AI can create images. A real forest will always be better than an AI-generated approximation. However, for visualizations and ideas for which you wouldn’t hire an artist or don’t have sufficient design skills, Gemini and ChatGPT are more advanced than ever.
Perhaps the most useful tools right now are the editing tools: changes to photos that would normally take professionals hours can now be made by anyone in seconds, and these capabilities will only get better in the future.