<h2 id="overview">Overview</h2>This dataset is designed to train a female lifestyle / beach-oriented LoRA, with a strong emphasis on bikini outfits, leisure activities, and outdoor summer environments.The dataset primarily represents Asian female subjects in relaxed, casual, and vacation-like contexts. It uses Booru-style tag captions, compatible with SDXL and Pony training pipelines.Trigger token: <code>thm</code><hr /><h2 id="dataset-focus-(what-the-model-actually-learns)">Dataset focus (what the model actually learns)</h2>Based on the captions, the dataset strongly emphasizes:<ul><li>Bikini and swimwear</li><li>Beach / summer environments</li><li>Relaxed, playful lifestyle scenes</li><li>Casual poses and leisure activities</li></ul>Identity-related features are present, but they are not the dominant learning signal.This makes the dataset well suited for:<ul><li>bikini / beach lifestyle generations</li><li>casual summer scenes</li><li>playful, relaxed female poses</li></ul><hr /><h2 id="captioning-format">Captioning format</h2>Captions follow a comma-separated Booru-style tag format.Example structure:<pre><code>tbm, person, asian_woman, slim, long_hair, light_skin, smiling, bikini, standing, beach, summer
</code></pre>Common tag categories observed:<h3 id="identity-(secondary-signal)">Identity (secondary signal)</h3><ul><li><code>asian_woman</code></li><li><code>thai_woman</code> (inconsistently used)</li><li><code>slim</code></li><li><code>long_hair</code></li><li><code>light_skin</code></li></ul><hr /><h3 id="outfit-(primary-signal)">Outfit (primary signal)</h3>Outfit-related tags are very frequent and highly consistent, making them a dominant feature learned by the LoRA:<ul><li><code>bikini</code></li><li><code>onepiece</code></li><li><code>twopiece</code></li><li><code>bikini_top</code></li><li><code>swimsuit</code></li></ul>⚠️ Some captions include contradictory outfit tags (e.g. <code>onepiece</code> and <code>twopiece</code> together), which introduces noise but also reinforces “bikini” as a broad concept.<hr /><h3 id="pose-and-body-position">Pose and body position</h3>Common pose-related tags:<ul><li><code>standing</code></li><li><code>sitting</code></li><li><code>hands_up</code></li><li><code>arms_crossed</code></li><li><code>leaning</code></li><li><code>one_leg_raised</code></li></ul>These contribute to a casual, lifestyle-oriented pose distribution, rather than controlled studio poses.<hr /><h3 id="environment-and-context-(strong-signal)">Environment &amp; context (strong signal)</h3>A large portion of the tags describe environmental context, reinforcing the leisure theme:<ul><li><code>beach</code></li><li><code>outdoor</code></li><li><code>summer</code></li><li><code>sunlight</code></li><li><code>sunset</code></li><li><code>ocean</code></li><li><code>sand</code></li><li><code>palm_trees</code></li><li><code>vacation</code></li><li><code>resort</code></li></ul><hr /><h3 id="mood-and-lifestyle-descriptors">Mood &amp; lifestyle descriptors</h3>Captions frequently include mood or lifestyle-oriented tags:<ul><li><code>smiling</code></li><li><code>playful</code></li><li><code>relaxed</code></li><li><code>cheerful</code></li><li><code>fun</code></li><li><code>lifestyle</code></li><li><code>fashion</code></li><li><code>trendy</code></li><li><code>leisure</code></li></ul>These tags reinforce a positive, vacation-like aesthetic, but are subjective and vary significantly across samples.

Files with SHA256: 512afe876917578a66591e48a366afa099b1d739f3db07611abb9802265ebd14