OpenAI announces DALL·E 3, built into ChatGPT

Background

A text-to-image model takes a written prompt and returns a picture intended to match it. The user describes a scene in ordinary language, and the model produces the image instead of the user drawing or assembling it. OpenAI's image models carry the name DALL·E, and DALL·E 3 is the third generation in that line ^[1][2].

The announcement

On 20 September 2023 OpenAI announced DALL·E 3 ^[1][2]. Two things about the announcement shaped everything that came after it. First, the model was built to be used inside ChatGPT, not only as a standalone generator. Second, it was designed to refuse requests that asked it to imitate the style of a living artist ^[1]. Neither point is about image quality; both are about where the model sits and what it will decline to do.

How access rolled out

The ChatGPT integration arrived in October 2023 for Plus and Enterprise users ^[1][2][3]. On 19 October 2023 the model became available to ChatGPT Plus and Enterprise users through a direct integration with the ChatGPT app ^[3]. Subscribers on those two plans were prioritised: they could generate more images, and generate them faster, than other users ^[3]. Access through the API followed in early November 2023 ^[1].

So the release came in layers rather than all at once. A request for a picture could be typed into the same app that was already used for conversation, but only certain subscribers got that first, and only they got the larger and quicker allowance. Developers building on the API had to wait a few more weeks.

Why it matters

The rollout is a clear example of a model reaching people through a product they already use rather than through a separate tool they have to find. Placing image generation inside a chat app changes the everyday use case: a picture becomes one more thing to ask for in the middle of an ordinary exchange, alongside questions and text. It also means the image model inherits the app's audience and its subscription tiers, which is why the first wave of access was tied to Plus and Enterprise rather than being open to everyone at once.

The style restriction is the other consequential detail. Refusing to imitate the style of a living artist is a limit that is built into the model itself, not something the user negotiates in the prompt; an illustration of a particular artist's manner is treated as a request to be declined. That makes the launch a reference point for how image models handle the relationship between generated pictures and the work of people who make a living from a distinctive style.

Finally, the sequence — announcement in September, chat integration in October, API in November — shows how a single model release can be split into stages, with each stage reaching a different group of users with a different level of access.