LabelPair

A booru-style tagging tool in browser, which used for training stable diffusion lora, etc.

All files will process in your local browser.

Features

Browser only. No installation required. Run at any devices.
Different view to tag like doing a captcha easily.
Batch tagging and merge to check. Useful for incremental tagging.
Tag distribution analysis.
Organize your trigger words in order.
Retrieve images with same semantic meaning.
Simplify workflow. No project concept. Drag files and start tagging.
Check unreasonable tags with rules. eg. indoors and outdoors are opposite and should not be in one image. (plan in future)
AI pre-tagging automatically in browser. (plan in future)
Organize your trigger words in natural language. (plan in future)

Prepare dataset

Use your favorite tagging model or method to tag your images automatically first. Personally i use this which base on kohya_sdscript.

No matter what method you use, your dataset should look like this:
```
dataset
    1.png
    1.txt
    2.png
    2.txt
    ...
```
The txt contains tags of the image with the same filename, its content looks like this:
```
solo, looking at viewer, open mouth, sitting, outdoors, collar, chair, cat, orange fur, male focus, cat, cute, blush, full body, smile, :3, sunlight
```
Upload dataset

Open Demo Link and drag images and texts(*.txt) together to specific area of the page.

All files process in your local browser, don't worry. I don't have money to buy a server.
Happy tagging

In MultiSelect mode, you can select some images to add/remove/rename tag. If you don't select any images, will use all images by default.

Retrieve can retrieve images that doesn't contain this tag. Useful to balance you tags.
Save your world

At the top:

Analysis shows your tags distribution and adjust trigger words order.

Export save result as txt files.

With txt files, You may consider override source tag files or use diff compare tool to double check.

If you don't know code, you don't have to build.

The section is for those who want to change to source code.

This project doesn't use webpack. Only one thing you need to do is to run tsc --build.

And then use http server to run index.html.