Photo realistic mix (majority Asian girls). To use with my loras.
Recommended sampler Euler, 2M Karras, & Restart.
v3.5: Cleaner background and model
v2.4: Ease of use & realism somewhere in the middle. Better background and generic model.
v2.2: Harder to use, more photographic look, less perfect, more realism
v1.4: Easier to use, more details. Recommended for beginners
More idols and newer version Loras also available at https://ko-fi.com/makina69
Training Tips with v3.5
30 images = 35 repeats.
40 = 30 reps
50 = 25 reps
mixed_precision = "fp16"
clip_skip = 2
max_train_epochs = 6 (Epoch 4, 5 or 6 should be the best)
[general_args.dataset_args]
resolution = 768
batch_size = 2
[network_args.args]
network_dim = 128
network_alpha = 64.0
min_timestep = 0
max_timestep = 1000
[optimizer_args.args]
optimizer_type = "AdamW"
lr_scheduler = "cosine"
learning_rate = 0.0001
unet_lr = 0.0001
text_encoder_lr = 5e-5
min_snr_gamma = 5
1. The rights to reposted models belong to original creators.
2. Original creators should contact SeaArt.AI staff through official channels to claim their models. We are committed to protecting every creator's rights.