DALL·E 2
Era 5 · Generative AI, reasoning, and agents (2022–2026) › Images, video, and the physical world
- Lineage
- Neurons and networks
- Kind
- Implementation
Who
OpenAI
What happened
OpenAI presents DALL·E 2, which generates images from text with a diffusion model guided by CLIP's representations. Its images, far more realistic and varied than those of the first DALL·E, also make it possible to create variations of a photo and to edit it with words.
Why it mattered
Photorealistic images from text. Image generation with diffusion reaches the general public.
The full story
The first DALL·E, from January 2021, was a Transformer that generated the image piece by piece. A year later OpenAI changed methods and bet on diffusion, which since 2020 had been competing with the best adversarial networks.
DALL·E 2 worked in two steps. A first model, called the prior, took the written description and produced an image representation in the space of CLIP, the model that connects images and text. Then a diffusion decoder generated the image from that representation. According to the paper, going through that intermediate step increased the variety of the images with little loss of realism or of faithfulness to the description. Because CLIP's representation captures both content and style, the system could create variations of an existing image and modify it with written instructions.
OpenAI announced it on April 6, 2022, and the technical paper, by Aditya Ramesh and four colleagues, came out a week later.
That same year Google's Imagen and Stable Diffusion arrived, both also based on diffusion.
Builds on
DALL·E and CLIP (January 5, 2021)
Documented direct influence
DALL·E 2 first generates an image representation in CLIP's space and then generates the image from it (abstract)
DDPM (June 19, 2020)
Documented direct influence
The DALL·E 2 decoder is a diffusion model, and Stable Diffusion's latent diffusion applies diffusion models in a compressed space; both papers cite Ho et al. (2020)
Threads that pass through here
- From diffusion to world models (step 4 of 7)
- From noise to image (step 6 of 7)
Sources
Original source
- Hierarchical Text-Conditional Image Generation with CLIP Latents (arXiv:2204.06125) (opens in a new tab). Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen, 2022. open license: CC BY 4.0
How to cite
“DALL·E 2.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/dalle-2
This entry was last reviewed on November 13, 2026.