Disco Diffusion v5.1 : A mini model study

Hello everyone. Hope you are doing fine.
For those of you using the Disco Diffusion notebook for generating ai art, here's a small model study I did; some of you might find it useful.

Method:
I did not generate images with single secondary models. Instead I used combos and tried 5 of them including the default. I used the same prompt, same seed, same resolution and settings for all runs except modifying the selection of secondary models to be used.

Results:

"ViTB32"
"RN50"
"RN50x4"

TimeToDisco(87)_0.png

"ViTL14"
"RN101"
"RN50x4"

TimeToDisco(88)_0.png

"ViTB32"
"RN101"
"RN50x16"

TimeToDisco(89)_0 (1).png

"ViTL14"
"RN50"
"RN50x16"

TimeToDisco(90)_0.png

The default setting:
"ViTB32"
"ViTB16"
"RN50"

TimeToDisco(92)_0.png

I liked the first and the last(default setting) combos most. Here are some more images generated with the same combo of secondary models as my first combo :

TimeToDisco(85)_0.png

TimeToDisco(86)_0.png

Kaliyuga has a much more comprehensive study here. But this is for the turbo version of the notebook I think.

Thanks for reading, see you in the next one!