<pre><code>sample images can be found in the link below.</code></pre><a target="_blank" rel="ugc" href="https://civarchive.com/posts/12266650">https://civarchive.com/posts/12266650</a>[[datasets]] [[datasets.subsets]]num_repeats = 4[general]resolution = 1024shuffle_caption = truekeep_tokens = 2flip_aug = truecaption_extension = ".txt"enable_bucket = truebucket_reso_steps = 64bucket_no_upscale = truemin_bucket_reso = 256max_bucket_reso = 2048another version of just1n but nerdier.

trying to get a nerdy character

tested to create the following test images posted originally:<ol><li>prompt:<ul><li>"looking after our family pet in the farm"</li></ul></li><li>prompt:<ul><li>"man with equinecock, male focus,"</li></ul></li><li>prompt:<ul><li>"malay men working at the farm, male focus,"</li></ul></li></ol><pre><code>[additional_network_arguments]

unet_lr = 0.0005

text_encoder_lr = 5e-5

network_dim = 128

network_alpha = 16</code></pre>the above results in large files. more dim less alpha = bigger filesize.<pre><code>[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",]</code></pre><pre><code>[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</code></pre>

Files with SHA256: e7d72ef081fe0c8d384b1a8155397d2c899f617eafde5e00ad2b95a09365522d