How AI image generators make a picture from your words
Cross-attention: how your prompt steers every part of an AI picture
The image generator reads your prompt only once. But it builds the picture in dozens of denoising steps. How do your words reach every step and every part of the picture?
-
1
At every step, the diffusion model gets the square as it is now, and with it your embeddings: the same ones, every time.
-
2
But no embedding says which part of the square becomes fur and which becomes snow. So every small patch of the square has to work that out for itself.
The rest of this episode is for subscribers
Subscribe to read every premium episode. You can cancel when you want.
Paid yearly: $99 instead of $119.88