<h2 id="wardrobe-slider">Wardrobe Slider</h2>A concept coverage slider for Anima. Slide one way to cover a character up, the other to strip them down — it changes how much clothing they're wearing while keeping their identity, outfit theme, pose, and art style intact.Negative weight → more covered / modest. Positive weight → more revealing. 0 = no change.⚠️ Trained for female characters only. Male subjects may work but don't bet on it.<h3 id="how-to-use">How to use</h3><ul><li>There is no single "correct" weight. Different characters react more or less harshly, and the usable span depends heavily on the base model — across different checkpoints I've seen useful ranges anywhere from roughly ±2 up to ±8. Start around ±1–2, then sweep up or down until you get the look you want.</li><li>Works across an enormous range of subjects — humans, anime characters, sci-fi/fantasy armor, androids, even anthro/monster designs — and across styles (anime, 2.5D, 3D, comic, painterly).</li></ul><h3 id="how-it-was-made-inference-only-(no-dataset)">How it was made — inference-only (no dataset)</h3>This slider was trained purely by inference — no images, no dataset. Instead of collecting example pictures, the LoRA learns a concept direction directly from the model's own predictions (the difference between a "covered" and a "revealing" version of the same scene). It's a guidance/concept-slider method, not image fine-tuning.Why that's good for you:<ul><li>It only changes coverage — nothing else. Because no dataset was used, it doesn't inject some training set's faces, bodies, or art style into your generations. Your character stays your character, and the base model's style stays intact.</li><li>It generalizes. It learned a direction, not memorized examples, so it works on characters, outfits, and styles it was never shown.</li><li>No overfitting artifacts from a small dataset, and it stays lightweight and composable with other LoRAs.</li></ul><h3 id="base-model-and-calibration">Base model &amp; calibration</h3><ul><li>Behavior is base-model dependent. On lighter/base Anima it's smoothest with a moderate range; on heavier fine-tunes and merges the response can be much wider (up to ±8) or more stepped. Always do a quick weight sweep on whatever checkpoint you're using.</li></ul><h3 id="tips">Tips</h3><ul><li>Stubborn iconic characters with very strong canonical outfits (e.g. Princess Peach, Makima) can resist the slider even at high weight. Add the tag <code>adapted costume</code>, soften the franchise token, and push the weight higher.</li><li>Keep a nudity negative if you want the high end to stop at skimpy rather than bare.</li></ul>

Files with SHA256: deecee296c5af52d216e3fec240295462756ee1cf048a3a46a8befade0c2375b