OpenAI introduces DALL·E 2 in a limited research preview
OpenAI introduces DALL·E 2, built around CLIP image embeddings and a diffusion decoder, and initially opens it only to a small group of invited users.
What happened
Event details
DALL·E 2 uses a two-stage text-conditional image architecture: a prior maps text to a CLIP image embedding, and a diffusion decoder generates an image from that embedding. The same representation supports semantically related variations, image editing and language-guided modifications. OpenAI reported four times the resolution of the first DALL·E; evaluators preferred DALL·E 2 for caption matching 71.7% of the time and for photorealism 88.8% of the time. The initial release was not open registration. OpenAI began with a research preview for roughly 200 artists, researchers and trusted users, then expanded access gradually. Early safeguards restricted violent, hateful, adult and political content and photorealistic faces of real people. Waitlist removal and public API access came in later, separate events.
Assessment
Why it matters
DALL·E 2 combined diffusion generation, CLIP alignment, variations and editing in a staged product deployment, helping move text-to-image systems into mass creative use while making access controls and safety filters part of the release story.
Availability
Access notes
The first release was an invitation-only research preview; model weights were not published.