
This Flux LyCoris was trained locally using SimpleTuner as an initial training test to see how to train poses. The dataset is comprised of only 26 images mainly with "gorilla press" being the only tag used. It took a few tries and there were some merges done with various epochs along the way, but I found 1e-4 to 7e-4 learning rates to be the most effective. It's not the most robust or versatile, but it gets the pose most of the time.
The training config for the final run:
{
"--resume_from_checkpoint": "latest",
"--data_backend_config": "config/multidatabackend.json",
"--aspect_bucket_rounding": 1,
"--seed": 42,
"--minimum_image_size": 0,
"--disable_benchmark": false,
"--output_dir": "output/models",
"--lora_type": "lycoris",
"--lycoris_config": "config/lycoris_config.json",
"--max_train_steps": 10000,
"--num_train_epochs": 0,
"--checkpointing_steps": 250,
"--checkpoints_total_limit": 20,
"--model_type": "lora",
"--pretrained_model_name_or_path": "black-forest-labs/FLUX.1-dev",
"--model_family": "flux",
"--train_batch_size": 1,
"--gradient_checkpointing": "true",
"--caption_strategy": "textfile",
"--caption_dropout_probability": 0.0,
"--resolution_type": "pixel_area",
"--resolution": 1024,
"--validation_seed": 42,
"--validation_steps": 250,
"--validation_resolution": "1024x1024",
"--validation_guidance": 3.5,
"--validation_guidance_rescale": "0.0",
"--validation_num_inference_steps": "20",
"--validation_prompt": "gorilla press",
"--mixed_precision": "bf16",
"--optimizer": "adamw_bf16",
"--snr_gamma": "2",
"--input_perturbation": ".1",
"--learning_rate": "7e-4",
"--lr_scheduler": "constant",
"--lr_warmup_steps": 100,
"--base_model_precision": "int8-quanto",
"--validation_torch_compile": "false"
}
描述:
This Flux LyCoris was trained locally using SimpleTuner as an initial training test to see how to train poses. The dataset is comprised of only 26 images mainly with "gorilla press" being the only tag used. It took a few tries and there were some merges done with various epochs along the way, but I found 1e-4 to 7e-4 learning rates to be the most effective. It's not the most robust or versatile, but it gets the pose most of the time.
训练词语: Gorilla Press
名称: gorilla_press_dataset.zip
大小 (KB): 40647
类型: Training Data
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
名称: Gorilla_Press_Flux.safetensors
大小 (KB): 106968
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success