Tag Management
To add a tag through the Manager GUI, enter it within the entry field located in the lower-left corner.
As you type in a tag, you may see bold red words appear in the box to the right; this is the manager's autocomplete feature. It will suggest words that are not in the current caption that you're working on, but that do exist in other captions within the dataset cache.
To make use of the autocomplete feature, you can either:
- Press the <TAB> key to autocomplete your tag entry to the top suggested word or phrase in the autofill box, OR;
- Use the <Up> and <Down> arrow keys to navigate between the three options, and press <ENTER> to select the one that you've highlighted.
After pressing the <ENTER> key, the tag entry field will be emptied, while your written tag appears in the caption for your current image (displayed right). To view the caption, you can navigate to the namesake tab on the left, just next to the Tag Editor tab.
While the Tag Editor offers a useful look at how your caption will appear in your TXT file when you save it, it isn't where you'll make any further modifications. To do this, instead navigate back to the Tag Editor tab.
In Tag Editor, you will see a grid of boxes containing the tags which pertain to the current caption (which, again, pertains to the image displayed on the right).
The very first tag, colored in gold and accompanied by a 🔒 icon, is your trigger word. This allows all images within your chosen dataset to be attributed a common name. If you've worked with ML concepts such as Low Rank Adaptations (LoRA) before, you may be aware that this word is (usually) integral to them being useful. As such, this tag is protected during your session as you edit the dataset. It may not be removed, nor may you change it mid-session. This tag is established when you first load your chosen directory, based on the first word that the Manager encounters and within the first `.txt file that it encounters when searching recursively through dataset directories.
Above the grid, you should see some radio buttons. Your settings here determine the behavior of your mouse events when clicking on a tag. By default, the behavior is set to merely delete the selected tag from the current caption, but where the Manager becomes most useful is the ability to delete/add a tag from/to ALL other existing captions in the dataset. So, if you knew that every image in your dataset contains trees, and you don't wish these trees to define what your Trigger Word refers to, you can extrapolate a tag from the current caption to all of the others!
There is a more efficient way of doing this, however. If you navigate to the tag entry widget again, you'll find that similar radio buttons exist for its own behavior, and they will work exactly as you would expect. You may notice that no deletion method exists here; but there is a way!
To negate your tag entry, simply precede your word or phrase with one or more dashes (-), as we are now subtracting tags! You will also notice that, while doing this, the autocomplete feature will switch its behavior to look for tags within the current entry that can be deleted.
To navigate to other captions to edit, press the <Control> key with a <Left> or <Right> arrow key.
Alternatively, you can press the left and right arrow buttons within the GUI above the image display:
Be aware that during a single session in the Manager, you are only ever editing an internal cache within the program, loaded into memory. None of your changes are being made directly to the TXT files! To save your changes, press the Save Dataset button under the Options tab. This will save all of your captions to their appropriate TXT paths.
Should you ever need to make further changes to the dataset, you can always load the same directory again as you did before.
Save early, and save often!