Every image 1st tag was - "Raelsa", so use it in your prompt. Other keywords can be found in the dataset provided with the model. Add "from behind"/"side view"/"front view" to positive and "braid"/((braid)) into negative if the generation is not on point - generating braid instead of ponytail - this is what I found while testing, maybe an edge case. Test for your self and post results below!

*tests preformed on illustorius and pony based models - propably better to stick to illustrius since model is based on it and performs better

2 Outfits - "green uniform" and "work uniform" (maid like one) //at least I tried - The green one is consistent, not 1:1 to the manhwa but it's ok. The working one is bit worse. You can think about those uniform tags as test tags. You gotta test them for your self. In the end model is as strong as the data put in and output image is as good as promt is. Illustarius is consistent on outfits, Pony... we don't talk about it.

Images in the showcase info: TEST Images for my model | Second set of images - (9 is a set) is direct comparison in generating -> illustorius based model first 9, Pony based model second 9 images

This is my first attempt to make LORA on CIVITAI. Feedback on how to improve eventual future models is appreciated.

This model is ok, but could be better, if anyone thinks the same, make better one please XD

//and tell me what you did to improve it