alexgenovese / train-sdxl-kohya

Kohya Training for XL models

  • Public
  • 116 runs
  • GitHub
  • License

Run alexgenovese/train-sdxl-kohya with an API

Use one of our client libraries to get started quickly. Clicking on a library will take you to the Playground tab where you can tweak different inputs, see the results, and copy the corresponding code to use in your own project.

Input schema

The fields you can use to run this model with an API. If you don't give a value for a field its default value will be used.

Field Type Default value Description
pretrained_model_name_or_path
string
stabilityai/stable-diffusion-xl-base-1.0
base model name or path
train_data_zip
string
Upload image dataset in zip format using this naming convention: XX_token className.zip
network_weights
string
Pretrained LoRA weights
output_name
string
new_model_name
Model name
save_model_as
string (enum)
safetensors

Options:

ckpt, pt, safetensors

model save extension | ckpt, pt, safetensors
resolution
string
1024
image resolution must be 'size' or 'width,height'.
batch_size
integer
1

Min: 1

batch size
max_train_epoches
integer
20

Min: 1

max train epoches
save_every_n_epochs
integer
5

Min: 1

save every n epochs
train_unet_only
boolean
False
train U-Net only
train_text_encoder_only
boolean
False
train Text Encoder only
seed
integer
98796

Min: 1

reproducable seed
noise_offset
number
0

Max: 1

noise offset
keep_tokens
integer
0
keep heading N tokens when shuffling caption tokens
learning_rate
number
4

Min: 1

Max: 9

Learning rate. It means 0.0001 or 0.0009
unet_lr
number
1

Min: 1

Max: 9

UNet learning rate. It means 0.0001 or 0.0009
text_encoder_lr
number
1

Min: 1

Max: 9

Text Encoder learning rate. It means 0.0001 or 0.0009
lr_scheduler
string (enum)
cosine

Options:

linear, cosine, cosine_with_restarts, polynomial, constant, constant_with_warmup

"linear", "cosine", "cosine_with_restarts", "polynomial", "constant", "constant_with_warmup
lr_warmup_steps
integer
0
warmup steps
lr_scheduler_num_cycles
integer
1

Min: 1

cosine_with_restarts restart cycles
min_bucket_reso
integer
256

Min: 1

arb min resolution
max_bucket_reso
integer
1024

Min: 1

arb max resolution
persistent_data_loader_workers
boolean
True
makes workers persistent, further reduces/eliminates the lag in between epochs. however it may increase memory usage
clip_skip
integer
1
clip skip
optimizer_type
string (enum)
Lion

Options:

adaFactor, AdamW, AdamW8bit, Lion, SGDNesterov, SGDNesterov8bit, DAdaptation, Prodigy

adaFactor","AdamW","AdamW8bit","Lion","SGDNesterov","SGDNesterov8bit","DAdaptation", "Lion", "Prodigy
network_module
string (enum)
networks.lora

Options:

networks.lora, networks.dylora, lycoris.kohya

Network module
network_dim
integer
32

Min: 1

network dimension
network_alpha
integer
16

Min: 1

network alpha

Output schema

The shape of the response you’ll get when you run this model with an API.

Schema
{
  "type": "string",
  "title": "Output",
  "format": "uri"
}