DALL-E 3: Its 2023 Launch, ChatGPT Workflow and API Retirement

Updated October 15, 2026. This article was originally published on September 21, 2023 and retains its launch-era focus. DALL·E 3 is now a historical model: OpenAI removed it from the API on May 12, 2026. The official DALL·E GPT in ChatGPT has also been retired; these are separate product changes. This revision reviews official documentation; no image was generated or tested for this article.

What DALL·E 3 introduced

DALL·E 3’s distinctive promise was a more conversational route from an idea to an image. OpenAI described ChatGPT helping develop and refine image prompts, rather than requiring a user to formulate every visual detail in a single initial instruction. A person could describe a scene, review the result and request a change through the conversation.

That workflow matters because an image brief often starts incomplete. A request for “a poster about a community garden” leaves questions about composition, lettering, mood, aspect ratio and intended audience. A conversational assistant can help articulate those choices. It does not eliminate the need to inspect the final picture or guarantee that every instruction will be followed.

Announcement and rollout

OpenAI’s official DALL·E 3 introduction video was published on September 20, 2023, the day before this site’s original article. OpenAI’s later October 19, 2023 release post announced DALL·E 3 availability in ChatGPT Plus and Enterprise. It described generating images and requesting revisions within a chat.

Those historical plan names explain the rollout; they are not a statement about what an account can access now. A landing page or old tutorial may retain launch-era availability text after a product has changed. For current use, consult today’s product help and, for developers, the model lifecycle documentation.

Why better image descriptions mattered

OpenAI’s DALL·E 3 research paper examines training with improved image captions. The underlying idea is that more descriptive training captions can help a text-to-image model connect instructions with details in the image.

This is distinct from the user-facing ChatGPT workflow. One concerns how training examples are described; the other helps a user express a request at generation time. Neither should be summarized as proof that any brief will reliably produce the exact intended result.

The earlier version of this article cited a glass-heart scene as proof of the system’s capabilities and included interview-style quotations. This revision does not reproduce those quotations or present a vendor demonstration as our own test. A selected example can illustrate a feature, but it does not establish a success rate, consistency across prompts or suitability for a particular project.

Safety and creative controls

OpenAI described DALL·E 3 as designed to decline requests for images in the style of a living artist. That is more precise than the earlier article’s broad reference to “renowned artists.” The launch material also described mitigations for requests naming public figures and an option for creators to exclude images from future image-model training.

These were described design goals and controls, not guarantees that harmful or misleading output was impossible. The October 2023 release discussed checks on prompts and outputs, together with red-team feedback. An opt-out mechanism for future training should not be read as a promise that previously trained models have forgotten an image.

Before using generated imagery, consider the source material you supplied, recognizable people or brands, and the requirements of the intended publication. This article does not determine legal rights for a particular image or provide a blanket clearance for commercial use.

Current API and ChatGPT status

OpenAI’s API deprecation record lists May 12, 2026 as the shutdown date for DALL·E 2 and DALL·E 3. The API changelog confirms their removal. An integration that still requests those retired API models needs a supported replacement; an old tutorial is not evidence that the endpoint remains available.

Separately, the current ChatGPT Images FAQ confirms that the official DALL·E GPT has been retired. It points readers to ChatGPT Images. This is not the same event as API shutdown, and it does not imply that ChatGPT image creation as a whole has ended.

Choose a replacement using the current image-generation documentation and the requirements of your application. Check supported inputs, output options, cost and relevant restrictions rather than assuming a model change is a drop-in substitution. We have not migrated or tested an integration for this revision.

A practical brief and review checklist

For a low-risk illustration exercise, use a fictional community-garden open day. Specify the intended format, a simple scene with plants and tools, room for a headline, and the desired visual mood. Keep the date and location as separate editable text in the final layout if precise lettering is essential.

• Composition: is there enough clear space for the headline and event details?
• Details: are objects, hands and repeated elements visually coherent?
• Text: check every letter and number rather than trusting a thumbnail.
• Meaning: does the image suggest a real venue, endorsement or event that you did not intend?
• Delivery: check the actual dimensions, crop and final file in the medium where it will appear.

This is a proposed review method, not a report of generated results. For another distinction between a visual style and an actual file capability, see our guide to 3D-style figure images: an illustration that looks three-dimensional is not automatically a printable model.

DALL·E 3 remains useful to understand as a step in conversational image creation. Its historical significance should be kept separate from current access, model selection and the quality checks required for a finished image.