OpenAI Strikes Back at Google’s Nano Banana With ChatGPT Images
This year has been marked by rapid progress in AI image generation models, with Google’s Nano Banana Pro going viral last month. But OpenAI hopes to have the last laugh: On Tuesday, the company announced the latest version of ChatGPT Images, an AI image model with improvements that should help it follow instructions and create more detailed images.
You can enjoy the new model starting today. It is available to all ChatGPT users worldwide and available in the API under the name GPT-Image-1.5. You can use the template in a new dedicated space in ChatGPT for images in the sidebar, with predefined styles and trendy designs, designed specifically for creating images – a major upgrade from just chatting with ChatGPT to generate images.
“Many people’s first experience with ChatGPT is turning a text prompt into an image. It’s a magical way to see what this technology can do, but the chat interface wasn’t originally designed for that,” wrote Fidji Simo, CEO of applications at OpenAI, in a blog post. “Creating and editing images is a different type of task and deserves a space designed for visuals.”
(Disclosure: Ziff Davis, CNET’s parent company, filed a lawsuit in April against OpenAI, alleging that it violated Ziff Davis’ copyrights in the training and operation of its AI systems.)
OpenAI claims the new model is four times faster than the previous one and better able to follow instructions and handle precise editing. You’ll be able to use these editing tools to add specific objects, change the style of objects, try on specific clothing items, and remove items.
This image, generated with GPT-Image-1.5, shows improvements in generating text in images.
OpenAI claims that the new 1.5 model is also more efficient at generating readable text in AI images. This is a notoriously tricky task for generative media models, with even the first generation of OpenAI’s model failing to deliver similar promises. Test images taken with version 1.5 seem to highlight improvements, but we’ll have to wait to see how our hands-on tests go.
This infographic created with GPT-Image-1.5 contains fairly readable text.
Don’t miss any of our unbiased technical content and lab reviews. Add CNET as your preferred Google source on Chrome.
All of these updates would be massive upgrades over the original model. When I examined the first model, it was noticeably slower than its competitors and lacked essential editing tools for refining your images. The improvements in Model 1.5 aim to resolve these inconveniences.
This updated model comes a week after Disney and OpenAI reached an agreement to integrate over 200 iconic Disney characters into ChatGPT images and Sora AI videos. You won’t be able to use the new template to create Disney characters yet – that’s expected to happen in early 2026 – but it will surely make it easier to create those recognizable superheroes and villains.
New AI model, renewed concerns
OpenAI’s timing with this new model probably isn’t to help you create your greeting cards with AI. Google recently abandoned Nano Banana Prothe updated version of its AI viral image tool that comes with the new Gemini 3. Nano Banana Pro’s ability to create hyper realistic imageswith readable text, based on Gemini’s knowledge of the world, marked a huge step forward in generative media. It also reignited fears that it was harder than ever to identify AI-generated content.
OpenAI issued an internal code red after Google abandoned Gemini 3, highlighting the company’s desire to compete with the cutting-edge AI prowess displayed by Google. Last week’s release of GPT-5.2 from OpenAI was another effort to narrow the gap between it and Google.
The first generation of OpenAI’s image model sparked a trend among AI users to do “Studio Ghibli” versions of themselves with ChatGPT. It also reignited the debate on ethics and legality of using creative AI tools such as AI image generators, notably with Hayao Miyazaki, co-founder of Studio Ghibli, who declared that AI tools are “an insult to life itself”.
Since March, we’ve seen a number of creators — authors, writers and actors — talk about the danger of letting AI tools use human-made artistic works and human likenesses to create AI content. Sora, OpenAI’s popular video generator ignited these debates this fall.
Copyright owners have diverged in how they choose to handle AI, but many have chosen to take one of two paths: making deals or going to court. We saw both, Axel Springer and Future’s agreements with OpenAI and the prosecution of Reddit And British Encyclopedia filed this year. Disney is an example. One day before announcing its agreement with OpenAI, Disney sent Google a cease and desist letter for copyright infringement. Copyright has become one of the most important and controversial issues in the age of AI.
