DALL·E and CLIP: text and images finally talk

OpenAI ships two models on the same day: DALL·E, which paints images from a text description, and CLIP, which figures out which caption best fits an image.

DALL·E isn't public yet, but the demos (an avocado armchair, a daikon walking a dog) spread everywhere. CLIP gets open-sourced and immediately becomes a building block of half the generative research that follows.

This is the first time a computer "sees" and "writes" using the same logic. From here on, every generative image model uses CLIP variants to understand prompts.