Diffusion Models

Question 1

3 parts of stable diffusion model

Answer 1

A language model, a diffusion model, and a decoder

Question 2

At a high level what does the language model in stable diffusion do?

Answer 2

transforms the text prompt you enter to a representation that can be fed to the diffusion model

Question 3

SD’s diffusion model is what?

Answer 3

Basically a time conditional U-Net

Question 4

What does SD’s diffusion model take as input?

Answer 4

some Gaussian noise and the representation of the text prompt

Question 5

What does SD’s diffusion model do with its inputs ?

Answer 5

Denoise (for several times) the Gaussian noise to get closer to the representation of your text prompt [IIRC Gaussian noise is one of the inputs]