Talk:AI Art Generation Handbook/Dreambooth

Face Training Note Base Model - V1.5 Scheduler DDIM


 * High learning rates and too many training steps will lead to overfitting. It only generate images from your training data, no matter what prompt is used.
 * Low learning rates and too few training steps will lead to underfitting . It unable to generate images according to the concepts.

Recommended Learning Rate : 0.000002 Training Steps : 1200

Constant with warmup

Use LORA : TRUE

use 8bit ADAM : TRUE Mixed Precision : fp16 Memory Attentions: xformers Text Encoders Epoch: -1

Total Number of Class/Reg Images : -1 (Need to have prior preservations)

Result: 7hrs VRAM Average Usage : ~9.8 VRAM