Model Training: FLUX A short explanation of the settings I am using currently to train on my models.
Base Model: 4.5 (Experimental) - Real_WoW (Flux+XL)Network Module: LoRANetwork Dim: 64, Network Alpha: 32Epochs: 10aim for ~ 150 tokens costResolution: 1024x1024Clip Skip: 2Text Encoder LR: 0.00001UNet LR: 0.001464LR Scheduler: cosine_with_restarts (4 cycles, 0 warmup steps)Optimizer: AdamW8bitGradient Accumulation Steps: 16Shuffle Caption: On, Keep n tokens: 7Noise Offset: 0.03Multires Noise Discount: 0.3Multires Noise Iterations: 12conv_dim: 64, conv_alpha: 64
