It's a diffusion model. If the initial random noise it started from had a dark blob there, and there's lots of masked Kamala pics in the training data, then adding a mask would make sense.
Diffusion Models Training Data Bias Analysis Explained
By
–
By
–
It's a diffusion model. If the initial random noise it started from had a dark blob there, and there's lots of masked Kamala pics in the training data, then adding a mask would make sense.