Question about Lora and prompting

#7
by Arbadel - opened

First, thanks a lot for the model, is really great and easy to use.

I want to try and train my own lora models, ideally following good practices so the model are of a very good quality for the community. Could you please list a very simple good practices guideline to follow in the process? Or maybe guide me in the right direction?

And a question about a particular character. I was testing the Princess Connect characters, and I cant get a good Misogi using her default uniform, not sure why. I even tried using a prompt guiding her outfit, but the final result is different.

Again thanks! Have a good day!

Cagliostro Research Lab org

Thank you for your enthusiasm ^^

for the first question
Perhaps this response might not be as helpful as you'd hoped, but I'll strive to make it as straightforward as possible. To create a good LoRA, a good dataset is also necessary; typically, I use at least 30 images to develop a LoRA character.

Maybe I could provide some references for you to further explore LoRA
https://stable-diffusion-art.com/train-lora/#Train_a_Stable_Diffuson_XL_LoRA
https://civitai.com/articles/3105/essential-to-advanced-guide-to-training-a-lora

for the second question
animagineXL 3.0 have different tag ordering, and for even better results, try to follow according to the Danbooru tags.

1girl/1boy, character name, from what series, everything else in any order.

for misogi try this

1girl, misogi_(princess_connect!), \(princess_connect!\) orange_eyes, side_ponytail, upper_body, (masterpiece), (best quality), (ultra-detailed), illustration, perfect composition, moist skin, intricate details,

xl_output_00031_.png

and happy generating^^

Thanks! I really want to learn about SDXL, so the lecture is very much appreciated.
Wow, thats a great Misogi, thanks for the tag ordering, I will give it a try! =)

Cagliostro Research Lab org
edited Jan 17, 2024

You're welcome, I'm glad we could be of assistance. If you have any more questions or would like to discuss further, please feel free to bring them up ^^.

Asahina2K changed discussion status to closed

Sign up or log in to comment