Leone Olfa from Eiyuu-ou, Bu wo Kiwameru Tame Tenseisu: Soshite, Sekai Saikyou no Minarai Kishi♀ (Reborn to Master the Blade: From Hero-King to Extraordinary Squire ♀) Update v1b (26th May 2023):<ul><li>Retrained model with adjusted parameters using D-Adaptation v3. The dataset is unchanged.</li><li>Block weights are not modified, unlike v1.</li></ul>Prompts:-<ul><li>Trigger Words: leone_olis, purple hair, purple eyes</li><li>School Uniform: twintails, hair ribbon, collared shirt, neck ribbon, cleavage cutout, shoulder cutout, detached sleeves, green skirt</li><li>School Uniform with Red Cape: twintails, hair ribbon, collared shirt, neck ribbon, (((red cape, epaulettes))), cleavage cutout, shoulder cutout, detached sleeves, green skirt</li><li>Outfit 3: twintails, hair ribbon, detached collar, (((blue dress))), black necktie, chest belt, belt buckle, elbow gloves, faulds</li><li>Outfit 4: double bun, black ribbon, off-shoulder dress, black belt, dress bow, red rose</li><li>Outfit 5: twintails, hair scrunchie, purple dress, short sleeves</li><li>Extra: 1girl, solo (please refer to my sample images in v1)</li></ul>Weight: 1 (you may experiment yourself)All my models were trained with 512x512 pixels. For optimal results, please upscale your generated outputs with your preferred methods. My fourth model.More experiments again...At first I tried kohya-ss's latest feature that allows changing the learning rate of each block, but it didn't quite meet my needs. I then realized that I could just use the SuperMerger extension to output specific blocks for my purpose. However, as I now need to adjust both block weights and LORA resize parameters, there are virtually unlimited possibilities. It is impossible for me to decide which resulting model is the best...Anyway, hopefully some of you will like it.

Files with SHA256: 48a08117bbe12fee7c9f294b22f729927f410a5095eccbc3afeb98b680e7c82a