See version info for tags

Fed some manga to the trainer; not sure how capable pony is at it but IL seems to work a little bit better, I imaigne there is a lot of bias towards fully colored images in the lora/checkpoint data you might have to work around

I recommend putting "Greyscale, Monochrome" as your first tags (after pony's preamble?) then removing all color-specific tags from your prompt as it may attempt to spot color i.e. "green eyes". Adding "3d, realistic" to the negative tags seems to help sometimes but may not be necessary. In the same vein most images were tagged with "Comic" and "text/English Text" which you may need to play around with