Making models can be expensive. Do you like what I do? Consider supporting me on Patreon 🅿️ or feel free to buy me a coffee ☕
Requested on Discord, this was trained on my hand selected and tagged dataset of anime screencaps. Trigger word is shinkai makoto
and contains substyles that are each triggered with kimi no na wa.
(mind the dot), tenki no ko
, and kotonoha no niwa
respectively. See the example images and final comparison. Keep in mind this is not a character lora. While it obviously overfitted on the main cast, it won't get characters accurately 100% of the time without a character specific LoRA/TI.
It works most accurately on AnyLoRA and I've offset it so that it works at 1 weight. Should actually work on most anime models and on NeverEnding Dream.
Clip skip 2 should be better, but careful as some of the examples are using Clip skip 1 because I forgot the setting from another test I was making. Pic 3 and 10 are generated with clip skip 1 and upscaled with clip skip 2.
How to use LoRA's in auto1111:
Update webui (use git pull
like here or redownload it)
Copy the file to stable-diffusion-webui/models/lora
Select your LoRA like in this video
Make sure to change the weight according to the instructions (by default it's :1
)