Don't use it to post naughty pictures online please.

The two v3 models, they are trained exactly in the same way, the only difference is that one is trained on 768 by 768 images and the other is 640 by 640 images, in my opinion the 768 looks way better.

From now on i will just train on 768.

Recommendations

This can be used to generate images of 768/512, 512/512, 512/256 and anything in between, in all my testing no duplicate faces were generated.

Best CFG 7 - 9, i would not go any higher unless you wanna pump up the steps too.

Any checkpoint works but you may have to adjust the weight in order to get it to look right, the max usable weight is 1.3.

I recommend Restart/DPM++ 2M for the first pass and Restart/DPM++ 3M SDE for the second pass, this will give you the best results.

Don't go above 2.0 times the first pass resolution, is better to pass it to img2img or use some other tricks to obtain more detail.

Extra notes.

I am not able to produce higher resolutions right now, since i switch from Windows to Linux and auto1111 is all mess up, but the higher the resolution the better far away faces are going look.

In my experience 1024 res does not work, the faces start looking weird and ugly, just stick to the resolutions i recommend.