(Credit: OpenAI)
OpenAI rolled out an update to ChatGPT’s image generation model on Tuesday that can reference web information, create multiple outputs at once, and deliver images containing non-Latin text. Images 2.0 is also ChatGPT’s first image model with "thinking" capabilities.
This is the company’s third image model, following the launch of Images last March and Images 1.5 in December. OpenAI says the new Images 2.0 is “a step change” over its predecessor in terms of instruction following, object placement, text rendering, and handling aspect ratios.
Images 2.0 is now open to all ChatGPT and Codex users. However, advanced thinking outputs are restricted to ChatGPT Plus, Pro, and Business users. I gave the new model a quick try, and the results didn’t disappoint.
When you select a thinking or pro model, the model can reference web information before delivering outputs. Its knowledge cut-off is currently December 2025.
The thinking model can also help generate a group of images with a single prompt. For example, ask ChatGPT to generate multiple pages of a comic book, and it will maintain consistency across the pages for characters, font, color palette, and overall mood.
You can generate up to eight pages at once. The feature can also be helpful for social media teams, as they can ask the model to generate media assets in various aspect ratios at once.
Recommended by Our Editors
Support for non-Latin text is another bonus. You can now ask ChatGPT to generate posters, flyers, or instructions in non-English languages, such as Japanese, Korean, Chinese, Hindi, Bengali, and others.
One of OpenAI’s samples had a book cover in the Indian language of Gujarati. I tried getting Images 2.0 to create a set of instructions in that language, and unlike its predecessor, the output had clear text, grammatical accuracy, and natural phrasing.
Another noticeable upgrade is photorealism. Images 2.0 can generate realistic humans with accurate skin tone, features, and dressing style, while also adapting the background to match the time period in the prompt. I asked it to create an image of a man enjoying a burger in a crowded McDonald's in the 1990s, and you can check out the result below.
(Credit: Jibin Joseph/OpenAI)The model has a few more use cases, and you can check them out in OpenAI’s blog post. Images 2.0 competes directly with Google’s Nano Banana Pro and Nano Banana 2. The last time we compared flagship image generators from the two, Nano Banana Pro was the winner.
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.