An attempt to overcome the typical FLUX bias towards homogenized facial features. This LoRA demonstrates training for authentic, diverse characteristics.<ul><li>Trained in Kohya_ss: 9h total, RTX4090 @ 250W, 6400 steps</li><li>Trigger: eyra_code_11112</li><li>LoRA Size: 32x32</li><li>Resolution: 1024x1024</li><li>Regularization Images: Used</li><li>Base models: flux1-dev.safetensors, ae.safetensors, clip_l.safetensors, t5xxl_fp16.safetensors</li><li>Focus: Natural facial features, authentic proportions, anti-homogenization training method.</li></ul>

Files with SHA256: 9f5d1c4e8b6014d3563949cc6ca51c9777c07759c5f1bf76664fe1ea27b8f386