Add to your prompt one of the following (default / sweat + larger shapes / thinner lines):<ul><li>@mrayudante82_ </li><li>@mrayudante82_, 6m6zin</li><li>@mrayudante82_, s1mpl3</li></ul>For captioning I followed the recommendations from the <a target="_blank" rel="ugc" href="https://huggingface.co/circlestone-labs/Anima#prompting">Anima</a> prompting guide.For example:<pre><code>masterpiece, 1girl, character name, series name, @mrayudante82_, {tags}, {natural language description}</code></pre>I preprocessed and inpainted the dataset with an edit model, then generated stable tags with a community model. Training used compact structured captions, the @mrayudante82_ style trigger, shuffled descriptive tags, and short local-LLM descriptions.

Files with SHA256: bfc25a09ea24b4322db7271a186dc33b71fc9a822175ce8a2d2eb2620426b03a