tested to create the following test images posted originally:

1. prompt:

   • "looking after our family pet in the farm"

2. prompt:

   • "man with equinecock, male focus,"

3. prompt:

   • "malay men working at the farm, male focus,"

[additional_network_arguments]

unet_lr = 0.0005

text_encoder_lr = 5e-5

network_dim = 128

network_alpha = 16

the above results in large files. more dim less alpha = bigger filesize.

[optimizer_arguments]
learning_rate = 0.0005
lr_scheduler = "cosine"
lr_warmup_steps = 0
optimizer_type = "Adafactor"
optimizer_args = [ "scale_parameter=False", "relative_step=False", "warmup_init=False",]

[training_arguments]
max_train_steps = 0
max_train_epochs = 68

sample_sampler = "euler_a"
train_batch_size = 5
noise_offset = 0.03
clip_skip = 1
weighted_captions = false
max_token_length = 225

save_precision = "bf16"
mixed_precision = "bf16"

xformers = true
sdpa = true
no_half_vae = true
gradient_checkpointing = true
gradient_accumulation_steps = 1

[advanced_training_config]
multires_noise_iterations = 6
multires_noise_discount = 0.3
min_snr_gamma = 5.0