If you like my work, you're welcome to leave a review. I don't really care, I do it for fun.<hr /><h3 id="lora-information">Lora Information</h3>Came across this anime recently. Watched it, thought it was alright. I liked this character, decided to make a LoRA, even if there were existing ones. Trained on Pony Diffusion V6 XL. Trained on site (see model details). Dataset is ~67 images, full-screen ShareX captures. Lightly edited and autotagged (with minor adjustments). Dataset is public. I trained an Illustrious version but I didn't bother uploading. I still prefer Pony. If I ever improve this one, I'll do the same for IL and upload it. Preview uses Prefect Pony XL v5.0. Not sure what it looks like on v6.0. Prompt Usage The trigger word is LorraineVivi. The recommended weight is ~0.8 to 1. Here is a sample prompt (and generally the more important tags).<pre><code>lorrainevivi, 1girl, blonde hair, long hair, yellow eyes, bangs, breasts, large breasts, black choker, black dress, white coat, brown gloves, bare shoulders, standing, hand on own chest, looking at viewer, indoors, bookshelf</code></pre>I ran my dataset through civitai's autotagger, and made some edits. I prefer to do manual tagging but I didn't feel like it this time around. Due to images / tagging, her outfit isn't really modular. For the ponytail version, switch long hair with ponytail. There should be zero generations of Rennt. I cropped him out.

Lorraine Vivie (The Unwanted Undead Adventurer) | Pony XL

See description for trigger word/prompt guidance. Network Rank/Alpha: 32/16 Optimizer: Adafactor, cosine with restarts 3 Learning Rate: default, 0.0005 and 0.00005 Others: Noise Offset of 0.1, SNR Gamma of 5 Dataset: 67 images, 10 epochs, 6 repeats, 4 batch size, 1005 steps + Keep Tokens, No Flip Augmentation, Shuffle Captions, No Regularization

Files with SHA256: 32515ea2faca4b069a46a52f898704ba442872c725a8de024592ff24565380e6