DALL·E 3

DALL·E 3 came out of OpenAI's research group, and the credits on openai.com name the people behind it: core research and execution by Gabriel Goh, James Betker, Li Jing and Aditya Ramesh, with dozens more across inference optimization, product, safety and policy. The reason they built it is stated plainly on the launch page: "Modern text-to-image systems have a tendency to ignore words or descriptions, forcing users to learn prompt engineering." The team's goal was a model that sticks to the text you give it, so you don't have to memorize keyword tricks or stack style modifiers to get what you asked for. OpenAI says it makes notable improvements over DALL·E 2 even when given the same prompt.

The most interesting design choice is that DALL·E 3 is built natively on ChatGPT. You don't write the final prompt yourself. You describe what you want, in anything from a single sentence to a full paragraph, and ChatGPT writes a detailed, tailored prompt for the image model. If an image is close but not quite right, you ask for tweaks in a few words and ChatGPT handles the rewrite. It's available to ChatGPT users and to developers through the OpenAI API, which bills on a pay-as-you-go basis.

Here's the honest part: the DALL·E 3 page now describes itself as the post that introduced the model and points you to ChatGPT Images 2.5 for the current experience. So DALL·E 3 is the previous generation inside ChatGPT, and you should know that before you build a workflow around it. What still stands out is the rights policy. OpenAI says the images are yours to reprint, sell or merchandise without asking permission. The guardrails are strict, though: it declines requests that name public figures and requests for the style of a living artist. If you want a conversational image tool with clear commercial terms, it's a sensible pick. If you need celebrity likenesses or artist-style mimicry, it won't do it.