DALL-E

basicwidespread

DALL-E is OpenAI's family of artificial intelligence image generation models, which creates new images—drawings, paintings, or photo-realistic illustrations—from text descriptions in a short time.

In plain terms

DALL-E is like a very fast illustrator to whom you only need to describe in words what you want to see. You type "a vizsla in a spacesuit on the shore of Lake Balaton, watercolor style", and you get an image that did not exist before.

It does not search among existing images like an image search engine; instead, it generates a new one for every request. This is why you can ask it for scenes that no one has ever photographed or drawn.

DALL-E was one of the first tools to show the general public that machines can create images from text. Many encountered it through ChatGPT or Microsoft's image generation services, often without realizing it was working in the background.

By now it is more a name that marks an era than a fresh product: OpenAI's newer image-generation models have taken its place, but news coverage and everyday speech still often use the name for the "text-to-image" genre.

How it works

Training: the model sees countless images along with their descriptions and learns which forms, colors, and styles correspond to which words.
Request interpretation: the entered text (the prompt) is converted by the system into numbers carrying the meaning of your request. In the case of DALL-E 3, ChatGPT first expands your input into a more detailed description.
Image generation: newer versions start from random image noise and step-by-step 'clean' it until an image matching the description emerges. This is called the diffusion process.
Filtering: the system checks both the request and the finished image, and refuses content such as violent or sexual material or depictions that misleadingly portray real people.
Refinement: if the result is unsatisfactory, you refine the description or request a modification, and the model creates a new version.

An everyday example

You are creating an invitation for your child's birthday and type into ChatGPT: "draw a cheerful dinosaur with a cake, storybook style". Half a minute later, the image is ready for you to put on the invitation. Between autumn 2023 and spring 2025, this image was generated in the background by DALL-E 3; today it is created by OpenAI's newer image generation model.

Who is affected

marketerYou can create illustrations for campaign ideas, social media posts, and drafts in minutes without a graphic designer or stock photos. → Marketing

parent, student, teacherFor teaching materials, short presentations, or stories, you can create your own image, and meanwhile it is worth discussing how the machine-made image differs from a real photo. → Education & Family

entrepreneur, executiveYou can get branding ideas and illustrative images cheaply and quickly, but you must pay attention to terms of use and copyright issues.

everyoneMore and more images on the internet are generated by machines, so you must know that a realistic-looking image is not necessarily real.

A common misconception

MythDALL-E cuts together a collage from existing images, or looks up a similar one on the internet.

In realityThe model does not select from an image database or stitch images together. Based on learned patterns, it creates something new every time. The style of the training images may still be recognizably reflected, and occasionally a close copy of a known image may appear; this is precisely what the copyright debate is about.

Limits and risks

Mistakes can occur in the details: botched hands, nonsensical captions, incorrect item counts, or physically impossible arrangements, especially in older versions.
It is a legally uncertain area who owns an image created by a machine and whether the creators of works used to train the model are entitled to anything.
It adopts biases from the training data, so stereotypes may appear in images regarding professions or gender roles.
Realistic but false images can be created with it, so the provider's filters reject many requests. Under the EU AI Act, from August 2026, so-called deepfake images that misleadingly imitate real people or events must be labeled.

Similar, but not the same

ChatGPT – ChatGPT is a conversational application, while DALL-E is the image generation model that created images in ChatGPT between 2023 and 2025.
Midjourney – Midjourney is an independent image generation service from another company, with a different style, interface, and business model.
Stable Diffusion – Stable Diffusion is a downloadable model that you can run and modify on your own machine if you comply with the license terms. DALL-E, however, is closed and can only be used through OpenAI and its partners' services.

Related concepts

Broader

generative AI, image generation, text-to-image model

Narrower

DALL-E 2, DALL-E 3

On the edge

OpenAI, ChatGPT, Midjourney, Stable Diffusion, diffusion model

A brief history

OpenAI introduced the first DALL-E in January 2021. The name combines the painter Salvador Dalí and the robot from the animated film WALL-E. DALL-E 2, far more lifelike, arrived in April 2022, followed by DALL-E 3 integrated into ChatGPT in autumn 2023. In March 2025, GPT-4o's native image generation in ChatGPT replaced it.

Where things stand

Text-to-image technology became a common tool by the end of 2026, but the name DALL-E is being phased out. Since March 2025, newer image-generating models from OpenAI have been working within ChatGPT, and the company is removing DALL-E models from its offerings.

Status: 2026 November

Also known as

DALL-E · DALL·E · Dall-E · DALLE · DALL-E 2 · DALL-E 3 · DALL·E 3

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.