<h2 id="heading-3274">Usage</h2><ul><li>Activation tag: <code>yonememei</code></li><li>Recommended extra tags: <code>red hair, blue eyes, single hair bun</code></li><li>Recommended ratio: 0.75</li></ul><h2 id="heading-3275">Training</h2><ul><li>Base model: AnyLoRA</li><li>Training data <a target="_blank" rel="ugc" href="https://huggingface.co/datasets/ashley-ng/lovelive-train-dataset/tree/main">here</a>120 anime screenshots, augmented using <a target="_blank" rel="ugc" href="https://github.com/deepghs/waifuc">waifuc</a> ThreeStageSplitter to get around 150 total training images</li><li>Optimizer: AdamW8bit, cosine_with_restart, 3 restarts</li><li>2 repeat x 10 epochs</li><li>Unet lr 5e-4, TE lr 1e-5</li><li>LoRA dim 32/16</li></ul>Different from previous LoRAs which used fully automatic tagging, for this I use a hybrid approach: auto tag and then half-manually curate the tags.<ul><li>Filter tags with frequency lower than 10</li><li>Ensure there's no tags of other characters (ex. the auto tagger might tag "blonde hair" if parts of Sumire is visible in the image)</li><li>Use more specific tag: single hair bun instead of hair bun, pinafore dress instead of dress (for the uniform), ...</li><li>Manually go through the image and add other tags where relevant, such as from side, close-up, ...</li></ul>This one performs better than my (unpublished) Natsumi LoRA, which used over 200 training images but the tags were not prepared as carefully as this one.

[LoRA] Yoneme Mei - LoveLive! Superstar!!

mei-10.safetensors

v1.0

Usage<ul><li>Activation tag: yonememei</li><li>Recommended extra tags: red hair, blue eyes, single hair bun</li><li>Recommended ratio: 0.75</li></ul>

Files with SHA256: b8417427197d2fd4d1b3e440c0511b800b6771032ab579336679d58463cff842