With ChatGPT Images 2.0, OpenAI fixes the biggest flaw of its generator

Last year, OpenAI made a strong move by unveiling ChatGPT Images, an image generator based on generative AI. Relying on GPT-4o instead of Dall-E, the tool enabled many users to generate creative, successful, and generally high-quality images. The generator encountered monstrous success, to the extent that OpenAI was temporarily forced to curb its usage.

More than a year later, the startup has just announced a major update to its image generator. Indeed, on Tuesday, April 22, OpenAI unveiled ChatGPT Images 2.0, which it presents as a new version of the image generation engine integrated into ChatGPT. Based on the model gpt-image-2, this new iteration aims to be more accurate and includes a range of new features. For CEO Sam Altman, the update is « a considerable leap forward ».

Also read : Here is GPT-5.4-Cyber, the ChatGPT programmed to hunt security flaws

ChatGPT Images finally writes correctly

Firstly, the update tackles one of the main flaws of image generators : text. As we realized last year, ChatGPT Images had difficulty writing correctly. Words appeared distorted, unreadable, and sentences sometimes had no meaning at all. Too often, we ended up with absurd and incomprehensible text fragments.

The new model is capable of restoring long, readable texts and placing them correctly within an image. Where previous models struggled with three words, Images 2.0 handles entire paragraphs integrated into elaborate layouts. On this point, it's a true leap forward.

Eight images with a single request

Beyond text, OpenAI has deeply rethought how AI generates images. Now, it is possible to generate up to eight images from a single request. Previously, it was necessary to generate each image separately, with a dedicated request, risking the result being a collection of content lacking cohesion.

With ChatGPT Images 2.0, OpenAI aims to attract professionals

Without surprise, Sam Altman's startup is targeting professional users, its new core audience. It mainly targets marketing teams or content creators. The AI model is thus capable of handling a wide range of formats, from 3:1 to 1:3, which includes web banners, presentation slides, large format posters or visuals optimized for smartphone screens.

Still aiming to attract professionals, OpenAI adds that the model is now capable of analyzing a complex request before producing anything. When a user submits a request, the AI can launch a web search to enrich its response, then generate several distinct visuals tailored to the request. According to OpenAI, it « moves from a quick interpretation to something that resembles a deliberate construction »

In this way, the tool integrates even better into a user's workflow on ChatGPT. Among the use cases cited by OpenAI, there are indeed a lot of purely professional situations, such as creating technical diagrams or designing interfaces.

An accessible professional tool for all internet users

ChatGPT Images 2.0 is available today to all ChatGPT users, without additional fees. As expected, some advanced features, such as the ability for AI to think before generating an image, are reserved for paid subscribers of the Plus, Pro, Business, and Enterprise offers. Moreover, the model underlying the tool, gpt-image-2, is available via OpenAI's API. Developers can thus integrate it directly into their own applications.

The release of ChatGPT Images 2.0 comes on an increasingly competitive market for AI-generated images. Google, Adobe, Midjourney, and Stability AI are currently competing in the AI-assisted creation market. On this booming market, will ChatGPT manage to come out on top and attract professionals?

👉🏻 Follow tech news in real time: add 01net to your Google sources, and subscribe to our WhatsApp channel.