accelerate launch --num_cpu_threads_per_process 8 train_network.py --pretrained_model_name_or_path=/storage/breakdomainanime_A0440.safetensors --train_data_dir=/notebooks/LoRA/sd-scripts/learning --reg_data_dir=/notebooks/LoRA/reg --output_dir=/notebooks/LoRA/sd-scripts/outputDir --resolution=512,512 --train_batch_size=3 --learning_rate=1e-4 --dataset_repeats=10 --max_train_epochs=10 --save_every_n_epochs=1 --save_model_as=safetensors --clip_skip=2 --seed=42 --network_module=networks.lora --caption_extension=.txt --keep_tokens=7 --enable_bucket --cache_latents --gradient_checkpointing
Description
The base model is BreakDomainAnime. I used kohya and paperspace. I used 78 images, no regs. No trigger word is needed.
accelerate launch --num_cpu_threads_per_process 8 train_network.py --pretrained_model_name_or_path=/storage/breakdomainanime_A0440.safetensors --train_data_dir=/notebooks/LoRA/sd-scripts/learning --reg_data_dir=/notebooks/LoRA/reg --output_dir=/notebooks/LoRA/sd-scripts/outputDir --resolution=512,512 --train_batch_size=2 --learning_rate=1e-4 --dataset_repeats=20 --max_train_epochs=8 --save_every_n_epochs=1 --save_model_as=safetensors --clip_skip=2 --seed=42 --network_module=networks.lora --caption_extension=.txt --keep_tokens=7 --enable_bucket --cache_latents --gradient_checkpointing