Lora trained on musubi for 1500 steps(also included 2000, 2750 steps. its kinda inflexible with styles but still good if the 1500 step doesnt capture likeliness) Was really pondering on the steps since it is my first character lora. But here you go. Suggestions to improve training would be nice. Works with all artstyles and is flexible with everything from outfits. This is her default outfit or something similar(Dataset has multple outfits) She wears a light grey long-sleeved blouse with white floral lace embroidery across the chest and shoulders, a white high collar, and a thin dark blue ribbon tied at the neck. A structured black corset-style waist with metallic gold and silver straps is worn over the blouse. Her long dark navy-blue paneled skirt has fine vertical gold stripes, layered black overskirt panels, gold accents, and a white triangular ruffled hem. astesiachar from Arknights, with long dark navy-blue hair, cyan-blue eyes, and blue-white feather ornaments. Just astesia char works , but it is not perfect likeliness. I dont know why but adding with long dark navy-blue hair, cyan-blue eyes, and blue-white feather ornament gives perfect likeliness even though i didnt include it in the dataset captions. probably just the model itself recognizing <a target="_blank" rel="ugc" href="https://huggingface.co/Hiccup1234/Astesia_lora_krea2/tree/main">https://huggingface.co/Hiccup1234/Astesia_lora_krea2/tree/main</a> Hugginface for all checkpoints V2: Definitely better than V1, particularly for non plastic skin. more versatile as well. use 1500 steps or 3000 depending on need

Astesia (Krea2)

Files with SHA256: 6a33dc31733ba1ea5abc8cd0c18116eea288f0618cd4d112f71d0d3c4914a9a0