Aesthetic gradients

[ghost]

The bug and feature request section is overfilled, and I think the devs are not able to go through every single one of them. I found this today: https://github.com/vicgalle/stable-diffusion-aesthetic-gradients

Apparently it works like a textual inversion / hypernetwork but the results are even better. This only applies to art styles, it remains to be seen if this can be useful for faces or persons.

I wouldnt know how to implement this so I just leave that here

[TingTingin]

you can check pr before to make sure its not already being worked on #2498

[bmaltais]

Here are some results from a custom embedding:

Sample of style image I used to create the embedding:

Image with no aesthetic gradients applied:

Same prompt with aesthetic gradients 5 steps:

Same prompt with aesthetic gradients 10 steps:

Same prompt with aesthetic gradients 20 steps:

Same prompt with aesthetic gradients 40 steps:

[Lewington-pitsos]

damn this is good, how many images?