Luria Aensland from Tsukimichi: Moonlit fantasy
Keep in mind that settings may vary slightly between different versions. See below for more details.
When I tested this with different checkpoints and LoRAs, it seemed to work quite well.
Illustrious:
A LoRA weight of 0.9 works great, though you can adjust the strength as you prefer.
Prompts I usually used:
luria_aensland, blue eyes, aqua hair, medium hair, shoulder lenght hair, maid headdress, frills, aqua maid dress, long sleeves, red ribbon, cleavage cutout, green and white vest, white cuffs, aqua dress, long waist apron,Pony:
A LoRA weight of 0.8 works great, though you can adjust the strength as you prefer.
Prompts I usually used:
aqua hair, medium hair, blue eyes, aqua maid dress, maid headdress, frills, apron,SD 1.5:
A LoRA weight of 0.8 works great, though you can adjust the strength as you prefer.
Prompts I usually used:
aqua hair, medium hair, blue eyes, aqua maid dress, maid headdress, frills, apron,Description
FAQ
Comments (8)
Your arts are good, please, make one of the goddess, if possible, please.
There really isn't many images of her to train a LoRA, so I don't know if I can do it
@XPHG So it's not possible? I thought you were performing the Loras, for good reason, could you explain that part to me, please?
@Dragoin471 I'm training the LoRA's on images from the anime, or searching the internet for the images of the character,
I usually want to find around 20 images minimum of the character, and then I use a program to train it to generate the images.
@XPHG I didn't know that, what a shame, but if you are good at redrawing, or in editing, you could do it, using the Loras and images of other arts as a base, get closer to the character.
Well, I'll be keeping an eye on your channel and so on.
At Pixai.net, they are updating, I'll wait. I continue to support you, thanks for the arts and so on.
Could you share your process for generating images? Your pictures always look so good!
Ty!
I don't know if I'm doing anything special, but I can tell you as much as I know.
I use A1111 on my PC.
I generate a batch of images in txt2img at 896x504, until I get one that I like.
Then I use Hires.fix with R-ESRGAN 4x+ Anime6B upscaler, upscaled by 2, with denoising set to around 0.4 - 0.5, and also ADetailer to get better quality on the face.
After that I send it to img2img.
I use ADetailer for a second time then, for me it seems that I get better quality doing that, but I'm not sure.
Then I use Ultimate SD upscaler
There I use Target size type with Scale from image size with scale 2
Upscaler with 4x-UltraSharp with Seams fix enabled
No idea how understandable it is, but I tried to do my best :D , if you have any other questions and if I know how to answer I will do my best to explain that.
Thank you for your response, it's a great help!