Use @N0a5tyl3 as the trigger word to apply the style.<ul><li>recommended to use euler or er_sde sampler with beta scheduler</li><li>weight range is 1 - 1.15</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, @N0a5tyl3, {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 @N0a5tyl3 style trigger, shuffled descriptive tags, and short local-LLM descriptions to preserve the stylewhile keeping character, pose, outfit, and composition editable.

Files with SHA256: 262771b5a608630a0cac40f1fcbc90d34bb77a3f4b8a5690f87e3590451d143e