@nochekaiser881 Thanks for your answer. I can understand that.

@redhome because most existing models are not as well trained in how to interpret or create those tags. Thats what a lora does. If a checkpoint was only trained for a specific art style, you can create a lora in that style to teach it posing, or specific characters. That is what @nochekaiser881 is so good at.

@OrricDrago The thing is most (if not all) Pony models understand that token even without the LoRA. Using LoRAs is always a trade-off between accurancy (of your initial prompt and the adherence to that prompt) and the quality of the output (concept bleeding). What that means is: using LoRAs will always introduce things from the LoRA that you don't want to happen, but happens anyway, because it was trained in a fashion the model was not. Let's say this LoRA was trained only on (xray, impregnation) pictures of women with big tiddies then the concept of "big tiddies" will bleed into your final image, even if you don't want that to happen (and you didn't prompt for that). The emphasis is put in front of the models own training. That's why I am asking: why introduce a LoRA when the model itself already knows the concept?