Lora trained on Chisato's one handed handstand scene from Lycoris Recoil
I would recommend using ~ 0.5 or 0.6 on the lora weight and generating at a 1:2 aspect ratio or else the legs will look squished
Trained using AOM2 and some random settings I stole from a youtube video.
I still have absolutely no idea what I am doing.
Description
FAQ
Comments (7)
Very good! Can you tell me what the training parameters are? I would like to study your results.
Thanks although I'm pretty sure I over trained this. Here they are anyway
I used 70 images in a folder called 100_chisatoHandstand (all scraped from danbooru with their original tags)
This is the saved json with the file paths removed
{
"pretrained_model_name_or_path": "",
"v2": false,
"v_parameterization": false,
"logging_dir": "",
"train_data_dir": "",
"reg_data_dir": "",
"output_dir": "",
"max_resolution": "768,768",
"learning_rate": "0.0001",
"lr_scheduler": "constant",
"lr_warmup": "0",
"train_batch_size": 4,
"epoch": "1",
"save_every_n_epochs": "1",
"mixed_precision": "bf16",
"save_precision": "bf16",
"seed": "1234",
"num_cpu_threads_per_process": 2,
"cache_latents": true,
"caption_extension": ".txt",
"enable_bucket": true,
"gradient_checkpointing": false,
"full_fp16": false,
"no_token_padding": false,
"stop_text_encoder_training": 0,
"use_8bit_adam": false,
"xformers": true,
"save_model_as": "safetensors",
"shuffle_caption": false,
"save_state": false,
"resume": "",
"prior_loss_weight": 1.0,
"text_encoder_lr": "5e-5",
"unet_lr": "0.0001",
"network_dim": 128,
"lora_network_weights": "",
"color_aug": false,
"flip_aug": false,
"clip_skip": 2,
"gradient_accumulation_steps": 1.0,
"mem_eff_attn": false,
"output_name": "ChisatoHandstand",
"model_list": "custom",
"max_token_length": "75",
"max_train_epochs": "",
"max_data_loader_n_workers": "",
"network_alpha": 128,
"training_comment": "",
"keep_tokens": "0",
"lr_scheduler_num_cycles": "",
"lr_scheduler_power": "",
"persistent_data_loader_workers": false,
"bucket_no_upscale": true,
"random_crop": false,
"bucket_reso_steps": 64.0,
"caption_dropout_every_n_epochs": 0.0,
"caption_dropout_rate": 0,
"optimizer": "AdamW",
"optimizer_args": "",
"noise_offset": ""
}
@Draff thank you!
@Draff Did you crop or edit the images you pulled at all?
@d0m1n1c nope. Left them all as is and let bucketing take care of it. The only thing I did was manually remove bad data.
hii... can you answer this i need help!!!
do you use wd tagger to make caption .txt? or manual tagging?
It depends where I am getting images from.
If its from a booru site like danbooru or gelbooru I will just use the tags they have already set
If its from twitter or pixiv I will use WD tagger and do some manual edits to add tags or remove ones I think are wrong
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.