there are more sample images this time around.do check them out.hope it is sufficient for your imaginations.but seems there is a lot of confusion. so to help those people out:<pre><code>very handsome </code></pre><pre><code>malay </code></pre><pre><code>burly </code></pre><pre><code>beefy </code></pre><pre><code>grandfather </code></pre><pre><code>just1n</code></pre><pre><code>high quality realistic photo</code></pre><pre><code>general </code></pre><pre><code>army </code></pre><pre><code>military</code></pre><pre><code>full body</code></pre><pre><code>walking </code></pre><pre><code>shopping </code></pre><pre><code>mall</code></pre><pre><code>confident </code></pre><pre><code>fierce </code></pre><pre><code>muscular thighs</code></pre><pre><code>middle aged </code></pre><pre><code>equinecock </code></pre><pre><code>realistic photo</code></pre><pre><code>hairy</code></pre>try to use the above in short 2 -3 words sentences and you should be good.<pre><code>[[datasets]]

[[datasets.subsets]]

num_repeats = 1



[general]

resolution = 1024

shuffle_caption = true

keep_tokens = 3

flip_aug = true

caption_extension = ".txt"

enable_bucket = true

bucket_reso_steps = 64

bucket_no_upscale = true

min_bucket_reso = 256

max_bucket_reso = 2048</code></pre>

anime malay 01

below are the prompts used for the sample images. this is the first lora that i get to use the early access feature. cool~!<pre><code>[ 
{"prompt": "malay father with thick hyper puffy pectorals, nipple piercings, prince albert piercings, and chain inserted into urethra,"}, 

{"prompt": "malay father with hyper erect equinecock, hairy thick pectorals thick muscular thighs"}, 

{"prompt": "realistic photograph of group of married malay men, beefy bodybuilders, hyper erections, thick PA and chain insertions"} 
]</code></pre><pre><code>[additional_network_arguments]

unet_lr = 0.0005

text_encoder_lr = 5e-5

network_dim = 128

network_alpha = 16

network_module = "networks.lora"

[optimizer_arguments]

learning_rate = 0.0005

lr_scheduler = "cosine_with_restarts"

lr_scheduler_num_cycles = 3

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 = 73

sample_sampler = "euler_a"

train_batch_size = 4

noise_offset = 0.1

clip_skip = 1

weighted_captions = false

max_token_length = 225

lowram = false

max_data_loader_n_workers = 8

persistent_data_loader_workers = true

save_precision = "bf16"

mixed_precision = "bf16"

save_state = false

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

[model_arguments]

v2 = false

[dreambooth_arguments]

prior_loss_weight = 1.0

[dataset_arguments]

cache_latents = true</code></pre>

Files with SHA256: 1ee5b794a604b3719fe57a1fd040820986a0add8a7c566d2a60f9ae5077ddfc8