https://gelbooru.com/index.php?page=post&s=list&tags=sasha_khmel+

Anima already know this artist. But as any artist trained/baked in, it also has every flaw in a dataset not manually curated. Watermarks, signature, patreon link baked in, some strong opinionated contrast etc...

A lora can "fix" that. Here the dataset was cleaned of most text (speech bubbles, artist name, ...) and the contrast changed a bit to be more neutral.

Using :

@sasha khmel 

Is a possibility still, that can reinforce the style. But will so inevitably bring back a bit of the "flaws"

But the lora work without trigger word too.

The purpose is to be have flexibility with the style. Not a 1:1 reproduction. Merging, mixing should be the main usage.