The guide is short as you will see: The number of people who do not know that safestensors can be opened as txt to extract the metadata is impressive. They can steal everything, even the number of images used and repeats, the only thing that is inaccessible is the database
Parámetros Importantes:
"ss_sd_model_name": "animefull-final-pruned-fp16.safetensors",
"ss_clip_skip": "2",
"ss_num_train_images": "384",
"ss_tag_frequency": {
"Casas": {
"ss_epoch": "10",
"ss_face_crop_aug_range": "None",
"ss_full_fp16": "False",
"ss_gradient_accumulation_steps": "1",
"ss_gradient_checkpointing": "False",
"ss_learning_rate": "0.0002",
"ss_lowram": "True",
"ss_lr_scheduler": "cosine_with_restarts",
"ss_lr_warmup_steps": "96",
"ss_max_grad_norm": "1.0",
"ss_max_token_length": "225",
"ss_max_train_steps": "1920",
"ss_min_snr_gamma": "5.0",
"ss_mixed_precision": "fp16",
"ss_network_alpha": "16",
"ss_network_dim": "32",
"ss_network_module": "networks.lora",
"ss_new_sd_model_hash": "06bece8e771a9ec16474db4c7d601d2f04fd08d1b1611072e7dd97c18cec3a09",
"ss_noise_offset": "None",
"ss_num_batches_per_epoch": "192",
"ss_cache_latents": "True",
"ss_caption_dropout_every_n_epochs": "0",
"ss_caption_dropout_rate": "0.0",
"ss_caption_tag_dropout_rate": "0.0",
"ss_dataset_dirs": {
"Casas": {
"n_repeats": 8,
"img_count": 48
"ss_num_epochs": "10",
"ss_num_reg_images": "0",
"ss_optimizer": "bitsandbytes.optim.adamw.AdamW8bit",
"ss_output_name": "Casas",
"ss_prior_loss_weight": "1.0",
"ss_sd_model_hash": "0b1893f6",
"ss_sd_scripts_commit_hash": "5050971ac687dca70ba0486a583d283e8ae324e2",
"ss_seed": "42",
"ss_session_id": "1694177070",
"ss_text_encoder_lr": "4e-05",
"ss_training_comment": "None",
"ss_training_finished_at": "1684514891.2831492",
"ss_training_started_at": "1684513546.7678096",
"ss_unet_lr": "0.0002",
"ss_v2": "False",
"sshs_legacy_hash": "d8afa111",
"sshs_model_hash": Hello everyone!
This is Tomas Agilar speaking, and I'm thrilled to have the opportunity to share my work and passion with all of you. If you enjoy what I do and would like to support me, there are a few ways you can do so:
Every donation, whether it's through Ko-fi or my Patreon membership, has a significant impact on my ability to keep creating and sharing content that you love. You are an essential part of my creative journey, and your support fuels me to reach new heights.
From the bottom of my heart, I want to thank you for considering supporting me. Every act of generosity inspires me and gives me the confidence to keep pursuing my dreams and projects. I encourage you to explore the provided links to learn more about my work and find a way to contribute.
Thank you for being here and for being a part of my community. Together, we can make a difference and achieve wonderful things. Have an amazing day!
With gratitude, Tomas Agilar
Description
FAQ
Comments (16)
PSA ALERT
you can also just click on "metadata" while hovering with your mouse cursor over a model in the extra networks list to access this info. also, i wouldn't call using that info stealing, it's formatted as a clear text json for a reason, to be easily readable by other people/tools
Be careful. This is false information.
For styles you don't need to use lots of repeats, in facts several people use 5-15 and with insanely good results. Epochs you need to use the ones that give you enough steps to work on, again dependant. Characters balanced? Depends. There are no rules other than knowing what you're doing and the images you're tagging and processing, end of the story.
Sorry to say this but you're very wrong dude. Feel free to check popular LORA makers like Konyconi, for example. In fact he even made a guide and it's the contrary you're saying.
I wouldn't call this stealing. Metadata is incredibly useful even for regular users because many of the most common tags used in training are often not included as 'trigger words' by the author and they might very well be what you need to achieve the best results when the trigger words alone are not enough.
@Tomas_Aguilar Well yeah that's the whole point!
But as someone who makes LoRAs you can remove specific tags from showing in the metadata or even the entire metadata altogether - if you want. This is a script I just wrote that will remove all metadata from a given .safetensors model (it saves a new one without it):
from safetensors.torch import load_file, save_file, safe_open
original_model_path = r"C:\folder\example.safetensors" # change this path <<<
new_model_path = r"C:\folder\example-without_metadata.safetensors" # change this path <<<
model = load_file(original_model_path)
save_file(model, new_model_path)
print("Done.")
If you would like a script that allows you to set which tags/parameters you would like to remove (ex: 'ss_learning_rate') - I can write one for you.
also, move mouse on one lora in lora list of WEBUI, and a circled "i" will appear, click for info.
@Tomas_Aguilar true
Good info, the virus Trojan Horse image is the sus though. 😂
Knew about this for a while. I've been trying to figure out why my LoRAs only have one tag in the metadata.
damn... what have u done? that sorta information should've close from public use, delete this guide immediately
I hope you're being ironic lol
@LDWorksDavid hehe, yep ;)
Really?
This is not new at all, this is the same as inspecting the safetensors using additional metadata tab in automatic1111...
This is neither stealing by the way.