Type | |
Stats | 423 |
Reviews | (92) |
Published | May 6, 2023 |
Base Model | |
Hash | AutoV2 59FE1BB4E7 |
Lora model of Touma Kazusa from anime/galgame White Album 2. The training images were screenshots from the anime, since I think the game's CG artstyle is a bit questionable.
I think someone trained a better LORA for touma a while ago, but since I couldn't seem to find it, I decided to train my own.
Here's some settings I used generating the sample images:
Note: I think you can just download the image (if civitai didn't compress it), change the extension to .txt and view most of these parameters that way
Model: Counterfeit-V2.5/Anything-V3.0/abyssorangemix2_Hard/DalcefoV3Anime etc.
Prompts: you can see by clicking on the images (it seems there is some issue with the anatomy, so negative prompts is definitely necessary, but too much seems to have a bad effect on the image generated, so you probably need to experiment yourself to find the right ones)
Sampler: DPM++ 2M Karras/Euler a
Steps: 20
Width: ~640
Height: ~920
CFG: 7 (some models may require higher CFG)
You can try with default settings (euler a + hires disabled), if it doesn't work well, please try enabling Hires.fix
settings: latent, 0 steps, 0.58 denoise, upscale by 1.5
ENSD: none