DALL-E is a series of Text-to-image generation generation models developed by OpenAI, named as a portmanteau of the artist Salvador Dali and the Pixar robot character WALL-E. The original DALL-E was announced in January 2021 as a 12-billion-parameter Transformer (architecture) model, based on the same architecture family as GPT-3, trained to generate images from text descriptions by treating image generation as a sequence prediction problem over discrete image tokens, rather than using the diffusion techniques that would later dominate the field.
Evolution across versions
DALL-E 2, released in April 2022, replaced the original discrete-token approach with a diffusion-based architecture guided by CLIP, OpenAI's joint image-text Embedding model, producing substantially higher-resolution and more photorealistic images and introducing an "inpainting" feature for editing parts of existing images from a text description. DALL-E 2's public beta, followed by general availability later in 2022, drew significant mainstream media attention and is widely credited, alongside contemporaneous releases such as Midjourney and Stable Diffusion, with bringing AI-generated imagery to mainstream public awareness for the first time. DALL-E 3, released in 2023 and integrated directly into ChatGPT for subscribers, improved prompt adherence and text rendering within images, and used ChatGPT itself to expand and refine short user prompts before passing them to the image model.
Reception and impact
DALL-E's releases sparked substantial public debate about the future of visual creative work, generating both enthusiasm from users experimenting with AI-assisted art and design and concern from professional illustrators and photographers about displacement and about the use of copyrighted images in training data without consent, a dispute that fed into broader AI and copyright litigation across the generative AI industry. The system also drew scrutiny over its potential for generating disinformation and non-consensual imagery, prompting OpenAI to implement content filters and, for a period, restrictions on generating recognizable faces of public figures.
Legacy
DALL-E is generally credited as one of the models, alongside earlier academic work on GANs and later open competitors, that established text-to-image generation as a mainstream consumer and commercial technology rather than a research curiosity. Its success contributed directly to OpenAI's broader multimodal strategy, informing the image-understanding and image-generation capabilities later integrated into GPT-4 and subsequent models, and helped popularize prompt-writing as a skill relevant to visual, not just textual, generative AI systems.