OpenAI launches ChatGPT Images 2.0 with API access
OpenAI has launched ChatGPT Images 2.0, making the image generation model available across ChatGPT, Codex and its API.
The update is intended to make image generation more practical for work tasks, rather than focused mainly on visual experimentation.
ChatGPT Images 2.0 introduces a new model, gpt-image-2, which developers can also access through the API to add image generation and editing to their own products and workflows. Access is rolling out to all ChatGPT and Codex users, while advanced outputs that use OpenAI's "Thinking" tools are reserved for Plus, Pro, Business and Enterprise users.
The latest version is designed to improve on earlier models in following detailed instructions, placing objects accurately and rendering dense text. It also supports a wider range of aspect ratios, from 3:1 to 1:3, to suit formats such as banners, slides, posters and mobile layouts.
Work uses
The launch reflects a broader push across the AI industry to move image generation beyond novelty and into marketing, publishing, design and product workflows. OpenAI is positioning the model for tasks such as creating recipe pages, magazine covers, storyboards, manga layouts and social media assets in different formats from a single request.
Users can request up to eight coherent outputs in one prompt, supporting projects that need multiple related images, such as poster series or multi-page visual sequences.
Another addition is "Thinking"-enabled image workflows in ChatGPT. When users select a thinking or pro model, the system can reason through a task, use web information where relevant, generate several distinct images from one prompt and check its own output. OpenAI says this is aimed at tasks where consistency, current information and close adherence to instructions matter.
Precision focus
OpenAI describes the update as a step change in instruction-following and in handling visual details that have often proved difficult for image models. These include small text, diagrams, iconography, user interface elements and dense layouts, where businesses and professional users have often found earlier tools limiting.
The model also works across more languages and can draw on broader visual and world knowledge to fill gaps in a prompt. In the API, images can be produced at up to 2K resolution.
The emphasis on text rendering and layout accuracy addresses one of the most persistent weaknesses in generative image systems. Such models have often been able to create striking visuals while struggling with legible text, structured page designs and scenes where several objects must appear in exact positions. OpenAI says Images 2.0 is built to handle those more demanding requirements more reliably.
That matters for commercial users because many practical image tasks depend less on artistic style than on precision. Marketing teams may need campaign visuals in several dimensions, product teams may want interface mock-ups, and publishers may need page layouts with readable headings and captions. By highlighting those uses, OpenAI is seeking to broaden the appeal of its image tools to customers who need outputs that require limited manual correction.
The release also extends OpenAI's effort to tie visual generation more closely to its wider product range. By making the model available in ChatGPT, Codex and the API at the same time, it is offering access to both end users and developers that want to build image tools into software and business processes.
As OpenAI put it: "Images are a language, not decoration."