LoRA Tests

  • All 3 tests are run on stable-diffusion-2-1-base - 512 x 512
  • The resemblance is not yet there, its probably the fault of the training params
Test 1 
MODEL_NAME="stable-diffusion-2-1-base";
LR = "constant"
LEARNING_RATE="1.e-4"
LEARNING_RATE_TEXT="5e-5"
TRAINING_STEPS="1000"
Test 2 
export MODEL_NAME="stable-diffusion-2-1-base";
export LEARNING_RATE="3.71e-5"
LR = "constant"
export LEARNING_RATE_TEXT="1.9e-5"
export TRAINING_STEPS="2700"
Test 3 
MODEL_NAME="stable-diffusion-2-1-base";
LR = "Polynomial"
LEARNING_RATE="6.71e-5"
LEARNING_RATE_TEXT="5e-5"
TRAINING_STEPS="2700"