Just a couple of years ago, telling apart human-created and AI-generated images was relatively straightforward. Back then, asking an image model to design a menu for a Mexican restaurant would result in bizarre, invented dishes like “enchuita,” “churiros,” “burrto,” and “margartas.”

Today, when I request a Mexican food menu from the latest ChatGPT Images 2.0 model, it produces something that could be used in a real restaurant immediately, with customers unlikely to spot anything amiss. (Although, a ceviche priced at $13.50 might raise some doubts about the seafood quality.)

For reference, this is what DALL-E 3 generated for the same request two years ago, back when ChatGPT itself did not create images:

Historically, AI image generators have had significant difficulty with spelling. This is largely because they typically relied on diffusion models, which reconstruct images from visual noise. As one explanation puts it, “We can assume writings on an image are a very, very tiny part, so the image generator learns the patterns that cover more of these pixels.”

Researchers have since investigated alternative methods for image generation, such as autoregressive models. These function more like large language models (LLMs), making sequential predictions about an image's appearance. Notably, OpenAI chose not to specify during a recent press briefing which model type powers ChatGPT Images 2.0.

The company did, however, detail that the new model possesses “thinking capabilities.” This allows it to perform web searches, create multiple images from a single prompt, and review its own outputs. These features enable Images 2.0 to produce marketing materials in various dimensions and craft multi-panel comic strips. OpenAI also states the model has improved its ability to render non-Latin text in languages including Japanese, Korean, Hindi, and Bengali.

It's important to note the model's knowledge is current only through December 2025, which may affect the accuracy of its outputs for prompts involving more recent events.

In a press release, OpenAI highlighted the model's advancements: “Images 2.0 brings an unprecedented level of specificity and fidelity to image creation. It can not only conceptualize more sophisticated images, but it actually brings that vision to life effectively, able to follow instructions, preserve requested details, and render the fine-grained elements that often break image models: small text, iconography, UI elements, dense compositions, and subtle stylistic constraints, all at up to 2K resolution.”

These enhanced capabilities mean generating an image isn't as instantaneous as asking ChatGPT a text question. However, creating something complex, like a multi-panel comic, still takes just a few minutes.

All users of ChatGPT and Codex will gain access to Images 2.0 starting Tuesday, with paid subscribers able to generate more advanced outputs. The company will also release a gpt-image-2 API, with pricing structured according to the desired output quality and resolution.