ct-neonight anime

Traning Settings ct-neonight FLUX.1_dev-fp8 Repeat 4 Epoch 2 Batch size 1 Mixed precision bf16 Sample Image resolution 768 * 768 Seed 1000000001 Sampler Euler Prompts (ultra realistic,32k, masterpiece:1.2),(high detailed skin:1.1),( high quality:1.1), (masterpiece, best quality), best quality, masterpiece, photorealistic, ultrarealistic, professional photograph shot on Canon EOS R6, More detail,) 1girl, solo, long hair, looking at viewer, bangs, skirt, black hair, red eyes, jewelry, standing, full body, weapon, pleated skirt, earrings, japanese clothes, sword, blunt bangs, nail polish, thigh strap, blue background, sandals, pale skin, black nails, red nails, geta, platform footwear, platform heels Negative Prompts 请输入反向提示词 Saving Settings Save every N epochs 1 Save precision bf16 Learning Rate & Optimizer Learning Rate 分开设置 unet 和 text encoder 学习率后 这个值失效 1e-4 Unet lr 0.0001 Text encoder lr 0.00001 Lr warm up 0 Lr scheduler cosine_with_restarts Optimizer AdamW8bit Lr scheduler num cycles 1 Train unet only Train text encoder only Min snr gamma 0 Network enable block weights 仅适用network.lora Network Rank Dim Suggestion4~128 32 Network Alpha Suggestion <= Dim 32 Tagging Shuffle caption weighted captions keep n tokens 1 Max token length 225 Noise Settins Noise offset（0~1） Disabled when setting Multires noise 0.1 Multires noise iterations 0 Multires noise discount 0 Others Seed -1 clip skip 1