DALL·E generates images from text

OpenAI shows a transformer that composes novel images from natural-language descriptions.

What happened and why it matters

DALL·E extended the GPT approach to image tokens and could combine unrelated concepts ("an armchair in the shape of an avocado"), an early sign that the same architecture generalises across modalities.

Sources

• DALL·E: Creating images from text ↗ OpenAI · 5 Jan 2021