OpenAI has launched the new version of its image-generation AI, ChatGPT Images 2.0, in a not-desperate-but-almost attempt to reclaim the crown in a market where it had long since been overtaken by Gemini.
And yes, it absolutely has.
I do not like using an exaggerated tone when I write, but the word “crushes” in the headline is accurate. Literally.
In fact, just a few hours after launch, this new model was already leading the LLM Arena, the most widely recognized popular benchmark for evaluating all kinds of AIs, in the image-generation category, with a major (enormous) lead over Nano Banana 2.
Let us take a look at what allowed it to pull that off.
As soon as you open ChatGPT’s new image-generation interface for the first time, it is clear that Sam Altman’s team has taken heavy inspiration from the success of Gemini and its Nano Banana 2. Just as in that case, Images 2.0 boasts several features that are immediately appealing to users, such as its commitment to adding advanced layers of reasoning, which help it better understand the underlying intent of the user’s request and return a prompt that is more closely aligned with what they expect.
That reasoning is visible throughout the entire image-generation process, which may take several seconds, using the typical text messages along the lines of “analyzing request,” “preparing sketch,” “generating environment,” or “one final touch.”
On the other hand, and to make it even more practical, ChatGPT has placed several predesigned styles within easy reach of the user, with a kitten (of course) as the common thread, allowing users to choose from looks such as “anime,” “dramatic portrait,” “coloring book”.
According to OpenAI’s documentation, ChatGPT Images 2.0 claims to have made a qualitative leap in following detailed instructions and in the precise placement of objects. As a result, it can render dense text, iconography, user interface elements, and complex compositions with a high degree of fidelity.
“Use the first part of the following text to create a realistic image in which the first part of the article appears on the pages of a 1970s paper newspaper, in black and white. The newspaper is folded on a worktable with a cup of coffee on it.” I then attached the full text of the Linkedin Study post
In addition, the model offers greater flexibility in terms of formats, supporting aspect ratios ranging from panoramic 3:1 to vertical 1:3. This versatility makes it possible to generate, in a single session, everything from website banners to social media graphics or posters, while consistently maintaining continuity of characters and objects.
“Now let us make a version of this image in which the newspaper appears in the hands of a 40-year-old director in her office. The image looks like an anime”
One of its biggest strengths? It can create up to eight images in a single batch, while maintaining continuity across characters and objects.
This attention to consistency even makes it possible to evolve them into more complex formats, such as the following manga-style page. This was the prompt used: “Now let us generate a manga comic page with several panels developing a story in which the director dramatically places the newspaper on the table and says to an employee, dressed in a suit, ‘Let us enter the Ecommerce Awards!’ The employee runs off in an epic style and heads to his computer. The computer screen is not visible, only the keyboard. He grabs the computer mouse with his hand. And clicks. The final panel is an onomatopoeia that says ‘!!!click!!!’”
Another fundamental pillar of this update is stronger real-world knowledge, since the model has knowledge updated through December 2025, enabling it to create accurate and clear infographics, visual summaries, and educational materials. In addition, the AI can search the web for information in real time to ensure that content is contextually accurate, reason about the structure of the image before generating it, and carry out internal checks of its own results.
OpenAI is also highlighting the tool’s improved multilingual understanding, which makes it possible to use it to create more accurate text in multiple languages and writing systems.
As a result, ChatGPT Images 2.0 no longer just generates strong images with text using the Latin alphabet, but can also deliver excellent results in languages such as Japanese, Chinese, Korean, Hindi, and Bengali. In this way, OpenAI is opening up new market opportunities for its own tool… and enabling brands to create localized campaigns where language is an integral and fluid part of the design, not just a translated layer placed on top.
Of course, not everything is quite so wonderful, and OpenAI itself acknowledges clear limitations. It still struggles with tasks that require fully coherent physical understanding, such as complex constructions like origami or puzzles, as well as the representation of extremely dense details or partially obscured or inverted surfaces.
Similarly, elements such as diagrams with highly precise labels or complex structures may require review.
ChatGPT Images 2.0 is now available to all users… although access to its most powerful features is segmented according to the type of plan. At present, anyone with a ChatGPT account can start using the model, but the “advanced reasoning” capabilities, which allow the AI to search for information in real time, generate multiple coherent variations at once, and verify the accuracy of its own results, are reserved exclusively for subscribers to the Plus, Pro, and Business plans. In this article, we explain the key details of all ChatGPT plans.
For development professionals and companies seeking deeper integration, the tool is also available through the OpenAI API and within Codex. In fact, the company wants image generation within Codex to be understood not as an isolated feature, but as part of a broader workspace that makes it possible to ideate, prototype, and develop products, presentations, or visual materials within the same environment.
In these cases, the pricing scheme is not a flat fee, but rather based on a pay-as-you-go model. The final cost of each generated image will depend directly on the configuration selected by the user: factors such as rendering quality (standard or high definition) and the chosen resolution (which now supports formats up to 2K in beta) will determine the final price of each visual asset created.
Image: ChatGPT, based on the instruction to search for information about Marketing4eCommerce.net and display it on a blackboard, presented by a friendly android wearing a bow tie
Your email address will not be published. Required fields are marked *
Δ