<a target="_blank" rel="ugc" href="https://gelbooru.com/index.php?page=post&amp;s=list&amp;tags=sasha_khmel+">https://gelbooru.com/index.php?page=post&amp;s=list&amp;tags=sasha_khmel+</a>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 :<pre><code>@sasha khmel </code></pre>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.

Files with SHA256: e39b03f765909fc05492ceaec0f2f9e73c14751b105a5566fda841e722b8f595