I ran a test on a new optimizer using some really old training data, and it looks like it works pretty well.

This LoRA was trained using screenshots taken from the Blu-ray version.

Using the prompt "ONY12 \(style\)" makes the generated style closer to that of the anime version.

If you want to use prompts for other art styles, try including "ONY12 \(style\)" in the negative prompts with an appropriate weight to get a better style response.

The prompt "anime screencap" can help fix body and hand structures to some extent, so use it as needed.

For details on how to use it, check out the prompt words in the preview image.

Weight: 0.8-1

ONY12 \(style\)

TH_ONY, grey hair, short hair, long bangs, short twintails, green ribbon, hair ribbon, grey eyes, hair over one eye, eyes visible through hair, white overcoat, green short kimono, green detached sleeves, puffy sleeves, front slit, green trim, green wrist guards, pentagram print wrist guards, blue sash, long sash, white tabi, zouri,

brown shoulder strap,small brown messenger bag,