How DALL-E 2 Actually Works
At the highest level, DALL-E 2's works very simply:
This was followed by three statements that made no sense and didn't sound simple at all.
The approach of "explain everything with different shaped boxes" is sometimes helpful, sometimes not.
Overall it did help me understand some though. The wow one for me was the GLIDE model - generate a new image that keeps salient features from the original image.