Please buy me a coffee if you liked my models! Your goodwill will be used for server costs for generating models.
Any Ideas For Enhancements?
Please post any ideas of improvement, if you have. For example, multiple safetensors processing and batch result download, or togglable tree-like metadata viewer, etc.
It's not a commission and I can't guarantee that it will be started, but it would be helpful anyway.
To Make a Long Story Short
The URL is https://safetensors-metadata-extractor.pages.dev
This is a tool that runs in your browser and Extracts metadata information from a safetensors file of Model for AI image generation (Lora, Lycoris, Checkpoint, etc.).
You need no additional installation, sign up, or other preparation. And processing time is noticeably quick.
What is this tool and the motivation to create
Stable Diffusion Web UI (Automatic 1111) has a handful feature to view a metadata of any safetensors files. But it is after launching web UI application when we can view it. And, even worse, for someone like me who uses remote cloud computer resources for image generation, server startup time will bring them suffering. Why should I wait that long just to check the metadata of a file?
So I made up my mind to create the equivalent function as a tool that runs completely on web browser.
What is Metadata?
The data that records training information, such as learning rates, number of epochs, number of images used, base model, type of tags used and total number of each tag, etc.
Features
Extract Metadata and Display its contents (JSON Data) with syntax highlighting enabled.
Can copy the data with just one push.
Noticeably quick execution, Even for 7GB or larger files.
No installation, No sign up.
Entire Processings are Completely done on browser. Any data is not sent to an unknown server. This is thanks to a great browser feature called HTML5 FILE API.
Cavert
In some case when the file has no metadata, is in unsupported formats, or is broken file, then the tool emits an alert.
I have confirmed that it is working properly on Chrome and Microsoft Edge. In some older browser that does not support HTML5 API, it might not work.
About download File
A download file on this page can lead you toward web tool. It's mere HTML file that redirects to the tool site.
機能拡張に関するアイデアをください
もしあれば、機能拡張についてのアイデアをポストしてください（複数ファイル処理、zip形式ダウンロード、開け閉め可能なツリー形式のメタデータ表示など）。
コミッションではないので着手するかはわかりませんが、参考になります。
要約
ウェブツールの場所はこちら⇛https://safetensors-metadata-extractor.pages.dev
Safetensors形式のファイル（Lora、Vae, Checkpoint等）からメタデータを取り出すツールです。
サイトを開いてボタンを押すだけで使えるので、インストールやログインは必要ありません。実行速度も早いです。
ツールについてと作成の動機について
メタデータはWebUIでも見れますが、それを見るためだけにいちいち立ち上げるのはたいへんです。
同様の機能をブラウザのみで動作するツールとして作れないかと思い、作成しました。
メタデータとは？
学習方法、使われたタグの種類とそれぞれの総数、エポック数や繰り返し数、ベースモデルなどを格納しています。
機能
- JSON形式でメタデータを取り出し、色を付けて表示
- ワンプッシュでコピー可能
- 数GBのファイルでも動作は一瞬です。
- インストール等必要なし
注意
- 古いブラウザではHTML5のAPIが動かないかもしれません。推奨Chrome。
- metadataプロパティにメタデータを含まないsafetensorsファイルに対しては機能しません。意図的にメタデータを消している場合などが該当すると思われます。
Description
FAQ
Comments (5)
Thank you!
All great, but I really would like to run it locally, can you give some instructions how to do it please?
Thank you in advance!
In terms of running locally, it seems that there are already a tool created by other developers.
https://civitai.com/models/112019/safetensors-meta-to-txt
One thing to note is that the tool requires a python installation to work.
Hey. Great tool, sadly it does not work completely as i hoped. When i extract the metadata, it shows most of the values. But a lot of the values are hidden/empty with only {} left.
Here is a example from my own lora:
{
"sshs_model_hash": "fa9e631721b36868ca2e0596bcd9615f495b1d35bc2044e358d5e4ed81fdcee7",
"ss_color_aug": "False",
"ss_training_finished_at": {},
"ss_max_token_length": {},
"ss_caption_tag_dropout_rate": {},
"sshs_legacy_hash": "087222cd",
"modelspec.prediction_type": "epsilon",
"ss_network_alpha": {},
"ss_sd_scripts_commit_hash": "31b26d59d3988980e4ecab464ae19efca06b80dd",
"ss_face_crop_aug_range": "None",
"ss_session_id": {},
"ss_batch_size_per_device": {},
"ss_max_grad_norm": {},
"ss_tag_frequency": {
"7_nopts": {
"nopts": 55,
"simple background": 55,
"solo": 55,
"full body": 55,
"purple eyes": 4,
"closed mouth": 34,
"blue background": 14,
"blue eyes": 12,
"white background": 3,
"open mouth": 9,
"smile": 23,
"green background": 14,
"happy": 4,
"standing": 40,
"green theme": 11,
"yellow background": 14,
"yellow theme": 9,
"black eyes": 13,
"yellow eyes": 9,
"pink background": 13,
"claws": 5,
"tail": 8,
"bird": 5,
"wings": 8,
"green eyes": 4,
"sitting": 4,
"red eyes": 4,
"looking at viewer": 20,
"brown eyes": 5,
"horns": 4,
"blue theme": 8,
":d": 3,
"teeth": 4,
"grin": 3,
"monochrome": 5,
"fangs": 4,
"fusion": 4,
"red theme": 3
}
},
"ss_unet_lr": {},
"ss_ip_noise_gamma": "None",
"ss_gradient_accumulation_steps": {},
"ss_optimizer": "prodigyopt.prodigy.Prodigy(decouple=True,weight_decay=0.01,d_coef=2,use_bias_correction=True,safeguard_warmup=True)",
"ss_prior_loss_weight": {},
"ss_seed": {},
"ss_max_train_steps": {},
"ss_bucket_no_upscale": "True",
"ss_caption_dropout_rate": {},
"ss_enable_bucket": "True",
"ss_num_batches_per_epoch": {},
"ss_shuffle_caption": "True",
"ss_learning_rate": {},
"ss_dataset_dirs": {
"7_nopts": {
"n_repeats": 7,
"img_count": 55
}
},
"ss_network_dropout": "None",
"modelspec.sai_model_spec": "1.0.0",
"ss_min_bucket_reso": {},
"ss_base_model_version": "sd_v1",
"ss_max_bucket_reso": {},
"ss_multires_noise_iterations": "None",
"ss_network_dim": {},
"modelspec.architecture": "stable-diffusion-v1/lora",
"ss_min_snr_gamma": {},
"ss_adaptive_noise_scale": {},
"ss_text_encoder_lr": {},
"ss_noise_offset": {},
"ss_full_fp16": "False",
"ss_cache_latents": "True",
"ss_multires_noise_discount": {},
"ss_steps": {},
"modelspec.implementation": "diffusers",
"ss_lr_scheduler": "constant",
"ss_sd_model_name": "anythingV3_fp16.ckpt",
"ss_bucket_info": {
"buckets": {
"0": {
"resolution": {
"0": 384,
"1": 448
},
"count": 385
}
},
"mean_img_ar_error": 0.05837488999706651
},
"ss_reg_dataset_dirs": {},
"modelspec.title": "nopts - Neopets Style v3",
"ss_lowram": "False",
"ss_training_comment": "None",
"ss_random_crop": "False",
"modelspec.resolution": "512x512",
"ss_caption_dropout_every_n_epochs": {},
"modelspec.date": "2023-11-24T22:35:36",
"ss_scale_weight_norms": "None",
"ss_num_epochs": {},
"ss_output_name": "nopts - Neopets Style v3",
"ss_zero_terminal_snr": "False",
"ss_sd_model_hash": "38c1ebe3",
"ss_v2": "False",
"ss_num_train_images": {},
"ss_gradient_checkpointing": "False",
"ss_total_batch_size": {},
"ss_flip_aug": "True",
"ss_epoch": {},
"ss_resolution": "(512, 512)",
"ss_new_sd_model_hash": "812cd9f9d9a0cb62aaad605173fd64dea13cc8ead9a18ca8eca5bc94aba58583",
"modelspec.encoder_layer": {},
"ss_network_module": "networks.lora",
"ss_keep_tokens": {},
"ss_training_started_at": {},
"ss_clip_skip": {},
"ss_num_reg_images": {},
"ss_mixed_precision": "fp16",
"ss_lr_warmup_steps": {}
}
Is this supposed to happen, or might this be a bug? Thank you.
Thank you for reporting!
I have researched and found a bug in the parser algorithm, then it was fixed.
Now the tool runs correctly without printing unintended '{}'. Please try it.
Your kindness assisted this tool to acquire further quality improvements.