LoRA trained on the style of the artist Masshiro Kachi. Trained on 139 images scraped from the internet (latest works first).

It is recommended to put the following string

(signature:2.0), text, watermark

in the negative prompt, due a large number of signatures, watermarks, and adversarial noise (Glaze/Nightshade) in parts of the dataset.

It seems to work well in a range of strength values, all showcased images are at strength 1.0. Adetailer is highly recommended for faces/eyes.