LoCon
Use these keywords to activate "Kyoto Animation screencap" "Anime screencap" recommended weight: 0.5~0.7
Suitable Model: anything, anylora, dpepmkm, detailed project v4
HiRes Fix: Latent - AnimeSharp - ESRGAN
LoHA
Use this word to activate "kyoto style"
Recommended weights : 0.4~0.7(1 works too)
Hires Fix Upscaler: R-ESRGAN 0.7 Denoise or 0.4, i prefer 0.7 usually. You can use anime sharp or any upscaler of your own choice
Suitable Model: AnyLora
The finalized version is better than the experimental and the locon.
Description
Use these keywords to activate "Kyoto Animation screencap" "Anime screencap" recommended weight: 0.5~0.7 for the LoCon
FAQ
Comments (8)
what checkpoint model you are using?
Anything v4 or detailed project v4 by closertodeath on huggingface they do the best with this model, especially detailed project. Anything vae or klf8 vae doesn't help with saturation so to solve that you can use "image filters" and increase the color to 5 or 3
The metadata of images are already there tho, u can see what used and done
Hello, how many images were in the dataset and how many titles were screencapped?
431 images (my lowest screencap dataset)
For animes: Violet Evergarden, Chuunibyou, Hibike, Hyouka, and i forgot honestly.. it's been months
@KoRo_ ok thanks
@KoRo_ Does it include all Hibike girls! Euphonium
This is a list collected for all Hibike girls! Euphonium
https://www.reddit.com/r/anime/comments/5jmuxb/background_characters_of_hibike_euphonium/
@tamaokaima sorry no, it doesn't. I was gathering Dataset on phone and it was tough for me, you still can use LoRAs for characters