<h3 id="heading-7"><a target="_blank" rel="ugc" href="https://forms.gle/h7JQGirRmJ11vprD8">I have opened up a request form</a></h3>Cless Gretchen from Lilitales. A request from the form. Works best around 1.0 weight. Training done on NAI.Trigger: "ClessGretchen, short hair, blonde hair, purple eyes" always needed, "large breasts" suggested for accuracy.<h3 id="heading-6">Suggestions/notes:</h3><ul><li>Limited data so expect some inconsistencies. </li><li>Note the hair may fail on some realistic leaning models so add "hair between eyes" as that part failed the most in my testing. </li><li>Main Outfit: "mitre, cross necklace, collar, capelet, tabard, white dress, detached sleeves, wide sleeves, belt, white thighhighs"<ul><li>Each part can be removed individually expect for the tabard as it tied to the other parts.</li><li>"mitre" for the hat.</li></ul></li><li>Very Flexible clothing wise due to regularization being used.<ul><li>Check samples above for ideas.</li></ul></li></ul>This lady was an adventure and was part of why I ended up taking another break due to the pain making this lora gave me. I was able to learn a lot about things I was doing wrong with my settings so in the end she came out really great even for how bad the data was. I hope you all enjoy.Feedback and reviews are always appreciated.Nerdy training numbers:Trained on D8Dreambooth trainerOptimizer: 16bitAdamWTraining resolution: 768Unet LR: 0.0001Tnec LR: 0.00001Unet weight decay: 0.08Tenc weight decay: 0.0950 Epochs - 2100 StepsTrained on 21 images with reg.

Cless Gretchen (Lilitales) | LoCon

Release version - works well considering the data and the reg gives nice style flexibility

Files with SHA256: a52bfe2dd8328bff45828702654a742e16a86e0d1fe436fab63355324b4f7a10