Use @rakupote as the trigger word to apply the style.<span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">For captioning I followed the recommendations from Anima <a target="_blank" rel="ugc" href="https://huggingface.co/circlestone-labs/Anima#prompting">prompting</a><span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> guide.For example:<pre><code>masterpiece, 1girl, character name, series name, @rakupote, {tags}, {natural language description}</code></pre><span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">For this particular LoRA I had to use an edit model to preprocess the dataset. Then I generated stable tags using a recent model with good performance. Finally, during training I used compact structured captions with a stable style trigger, shuffled descriptive tags, and short natural-language descriptions made using a local LLM, helping preserve the artist style while keeping details like character, pose, outfit, and composition editable.

Files with SHA256: d1139d28197a57a74cb7e8854d7b04830e7f184e8f20cd54608c4a1c2ecf48fc