VisualNote Portrait Illustration Key changes in training process - All images manually cropped to fit character and resized to 1344px. -- This cropped out complex effects, backgrounds, people, animals, creatures. - Reduced max pixels From 1536 to 1344px (1.9 MP) images -- (higher than 1.9 MP causes issues for illust xl models). - More Various Styles and Character Portrait illustrations. - Removed low quality images and bad arts after cropping. - More white background or flat background illustrations. - 2x Increase LR rate from 0.0002. - 5x Increase Text LR rate from 0.00005. - network Dim/alpha from 64/32 to 128/64. - Optimizer AdamW8bit. Steps: 4230 Loss: 0.0889 Parameter Settings Network Module LoRA Use Base Model Nova Anime XL - IL v17.0 Trigger words VisualNote Image Processing Parameters Repeat 3 Epoch 10 Save Every N Epochs 1 Training Parameters Seed - Clip Skip 2 Text Encoder learning rate 0.00005 Unet/DiT learning rate 0.0002 LR Scheduler constant_with_warmup Optimizer AdamW8bit Network Dim 128 Network Alpha 64 Gradient Accumulation Steps - Label Parameters Shuffle caption - Keep n tokens - Advanced Parameters Noise offset 0.03 Multires noise discount 0.1 Multires noise iterations 10 conv_dim 32 conv_alpha 16 Batch Size 1 Sample Image Settings Prompt 1girl, boots, sfw, solo, short hair, guitar, photoshop (medium), instrument, pink hair, smile, yellow eyes Sampler euler