Ran from Urusei Yatsura. Good at 0.7.
Negative promt affects the colors. Here's the negative promt that I use:
(worst quality, low quality:1.4), EasyNegative, bad-artist, low res, (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation
So.
A little overbaked because I wanted Ran to keep the shape and color of her unusual hairstyle at 0.7, a good mixing point between Lora and different models. But smaller steps did not show the desired result.
Trained only on anime screenshots, so don't expect a good match with realistic models. I used 0.5-0.6 and anything-v4.0.vae.pt to reduce the saturation level for realism, but use what you are more comfortable with.
Try the "gray bikini" or "bikini" token for clothes from episode 15. Alas, there were only few screenshots of these clothes.
Sometimes Ran likes to hold something in her hands, I don't understand why. Maybe it was a feature of the models I was testing Lora on. I'm sure I put the necessary tokens correctly.
Used Anything 3.0, Clip Skip 2.
Description
FAQ
Comments (5)
Que tipo de IA es el que usaste para Ran estilo dibujo? Quiero probarlo, por favor dime 🙏🏻
Sorry, I don't fully understand the question. If the question is how Lora was created, then the Kohya_ss GUI and screenshots from the anime were used. If how to replicate images from the gallery, you can either click (i) in the lower right corner of the image to see the generation information, or use "Copy Generation Data", which is much more accurate. The generation information also shows the model/checkpoint i used for the image. In this way you can easily recreate each image, and then make your own changes in promt.
@Kenfox Sorry, I forgot that I have the automatic translator, I said this:
"What kind of AI did you use for drawing-style Ran? I want to try it, please tell me 🙏🏻"
@LadyLum Well, still don't understand... Are you asking about the program on which the images above were generated? If so, it was Stable Diffusion running locally.
https://github.com/AUTOMATIC1111/stable-diffusion-webui
If the question is what models were used to generate, they are all listed in the generation parameters.
It depends on the model how much the style will change. Some lean towards anime, some towards realism. To get closer to the original style of the series, it is better to use models "anything 3.0" and "anything 4.0", because Lora was created based on these models. The best weight for Lora is 0.7.
Or is the question about the creation of Lora? Sorry, write more, I'll be happy to answer.
KenFox where are you?
I cant believe such a skilled creator of Stable Diffusion stopped to use it and disappear!
Please, come back!