Images of ChatGPT 2: Why OpenAI Developed a New Image Model After Killing Sora
A lot has changed in the AI industry in the four months since OpenAI released ChatGPT Images 1.5. We’ve seen a hot race to build agent tools, an unprecedented deal with the Pentagon and endless AI slop.
Now, OpenAI is back in the productive media game. The company announced on Tuesday that it will release ChatGPT Images 2, the next-generation image model.
ChatGPT Images 2 is intended for creating text-heavy designs, like in this matcha ad and fake magazine cover.
It may seem strange that OpenAI is releasing a new image model just a month after announcing the shutdown of its once-viral Sora AI video program to focus on building “core products” that are business-friendly. But it’s clear from how the new model is built that OpenAI isn’t backing down from that goal.
ChatGPT Images 2 is designed to produce text-heavy images, including infographics, scientific posters, study guides and marketing materials. The days of weird Sora videos and Studio Ghibli-inspired memes are over.
Now, the company is building an AI that can perform what it calls “economically important creative tasks.”
“The scope and use cases of virtual intelligence have recently expanded, and we believe that this is very important to ChatGPT’s idea of creating your own assistant, because your creative assistant is a big part of who you are,” Adele Li, who leads the product of ChatGPT Images, told reporters at a press conference.
(Disclosure: Ziff Davis, CNET’s parent company, in April 2025 filed a lawsuit against OpenAI, alleging that it infringes Ziff Davis’ copyrights in training and using its AI programs.)
From these examples, you can see how much better ChatGPT Images 2 is at rendering text readable.
OpenAI has been chasing the dream of a big app, a one-stop shop for all things AI, built on its Codex platform. ChatGPT Images 2 delivers a creative piece of that puzzle.
The new model naturally improves typing, iconography and composition to produce professional AI images. It can produce text in many languages. AI graphics models are notorious for creating legible, realistic text. ChatGPT Images 2 is the best OpenAI model for that yet. Google has previously improved its translation of text Nano Banana Probut even that “best” model had an accuracy problem.
ChatGPT Images 2 is available to all users now. Your generation limit depends on your system: The more you pay, the more AI images you can generate.
Developers using the model in the API can create images in 2K and 4K resolutions, although these higher resolutions are still in beta and may be wonky. Paying users can also create images using inference and reasoning models, which help them search the web for information, compile it into a readable design and double-check their work.
“Image model” doesn’t seem like the right term for ChatGPT Images 2, even if it’s technically correct. ChatGPT doesn’t capture the stunning surrealism of AI graphics like Midjourney, nor does it offer anywhere near the editing tools of Adobe Firefly.
But it’s aimed at a middle-class user group of Midjourney art lovers and professional Adobe creators: those who need to create engaging content.
Like Anthropic’s recently released by Claude DesignOpenAI’s ChatGPT Images 2 is aimed at working professionals. Teachers can use it to create study guides and illustrated lesson plans. Marketing managers can create social media posts and visual assets.
You can create up to eight images in a single notification, such as a three-page report, keeping visual consistency across them all.
You can make long reports with ChatGPT Images 2, all pages are the same.
This is the second part of the AI generated lime pie recipe. Note the visual consistency.
Another disadvantage is that if you want to modify the AI image, you will still need to reproduce it. For most text-heavy projects, that may be necessary, so you’ll use up your credits quickly. OpenAI said it is focused on maintaining its iterative, fast-backed workflows to keep it simple to use.
OpenAI’s security practices haven’t changed much since its last graphical model. It still includes metadata at the C2PA standard, to identify the origin of AI images. Abusive and illegal images are still prohibited in the OpenAI policies, an important clue for AI companies to use successfully, given the recent examples of AI-generated deepfakes and non-consensual intimate images.
