LoRA model fine-tunes Power from Chainsaw Man, optimized for cosplayer datasets with eye color variations. It balances efficiency, stability, and accuracy, maintaining Power’s key traits: long blonde hair, red horns, orange eyes with symbol-shaped pupils, and sharp shark-like teeth. Trained on Flux.1 (Dev-fp8) with AdamW8bit, it ensures compact, high-quality outputs while preventing overfitting.