V1.1 z-image: Use weight 0.7–0.9, start prompt with 'gta_sa_style. Digital illustration of a' ... Best results with 0.8 and Euler<s>V1.0 z-image</s> It's was fine but worse then I expected at start.V1.1 Pony: This version, in my opinion, is a bit farther from the original arts than v1.0 Pony, but it works much more better, and it is less problematic. Use weight 1.0 and lower, 0.8 works well for most cases. '3D, photo, CGI, render' in negative like in v1.0 Pony isn't always necessary, but you could put it by default if you want to keep the image flat more or less. In mostly generation, characters will have parted lips/open mouth/smirk/with slightly smile mouths. So if you want to surely avoid it, use 'closed mouth'.V1.0 Pony: Use weight 0.9-1. And add 'gta_sa_style'. With additional character LORAs I recommend put '3D, photo, CGI, render' in negative. Using without 'gta_sa_style' is possible but generations less flat and different colors, closer to GTA5 loader screens. Adetailer is recommended. Rarely it made an angry face; You could avoid by putting another emotion or putting 'angry' in a negative.

GTA San Andreas'  posters/loadscreens style

I erased one problematic image, pretty much changed the setting of training, increased detailization of some images, and made some face expressions more calm so that they had less effect on the Lora.

Files with SHA256: 5b28adc8d0c093d6646266ceb6017fbec75ed794a25fa88570941054945db975