<h1 id="how-to-use">How to use</h1><h3 id="version">Version</h3><ul><li><p>Use v1.0. Don't use v0.1</p></li></ul><h3 id="subject">Subject</h3><ul><li><p>This lora only works for closeup portraits</p></li><li><p>If you need wider framing, inpaint</p></li></ul><h3 id="prompting">Prompting</h3><ul><li><p>To get the effect, you must use the trigger, "<strong>ornamented face portrait</strong>"</p></li><li><p>Works great to simply pick two colors, e.g. "ornamented face portrait, <strong>purple and yellow</strong>"</p></li><li><p>But the model responds well to these other keywords matching the training images: <strong>butterflies, flowers, tassels, pearls, jewelry, ribbons, buttons, fur, stars, iridescent, glitter, fishnet, hearts</strong>.  And the model has learned to stick just about anything else in the prompt onto the face</p></li><li><p>You will get a Chinese woman unless you put "asian"in the negative</p></li></ul><h3 id="lora-weight">Lora weight</h3><ul><li><p>For SDXL base, try weights from 0.5 to 1.5</p></li><li><p>For PD/IL, try weight of 2.0</p></li><li><p>Go as high as you can until the face becomes unnatural. Highest weights will show better makeup on the faces and more ornamentation</p></li></ul><h3 id="adding-other-loras">Adding other loras</h3><ul><li><p>Makeup Master SDXL with "douyin makeup" trigger is a great compliment. Most of the training images had douyin style.  </p></li><li><p>I haven't tested with any character loras</p></li></ul><h3 id="training">Training</h3><ul><li><p>Trained on ~100 faces of Chinese models with this fashion style. The trigger phrase was the caption used during training</p></li></ul><p></p>

Ornameted Chinese Fairy SDXL

<p>v1.0 is properly trained.  Don't use the other version</p>

Files with SHA256: 5d547a8d50fce02b51ad44dc74204cc8ce504ff2239c25cb12f8fe7a479a9b60