Last update: 10-4-2025.

Note 1: If you're reading this, I'm assuming you're at least a bit familiar with training a LoRA. If not, I have some steps/tips in https://civitai.com/articles/7483/civitais-trainer-a-simple-beginners-guide-to-training-character-lora-using-it.

Note 2: Like this guide? Consider supporting me at https://www.patreon.com/c/MetalChromeX so I can continue making cute anime girl models (and better pay off crippling college and credit card debt)!

Intro

This is a relatively quick and straightforward guide on some of the tips I've learned/have been taught on how to tag images in a dataset. Additionally, since several auto-tagging programs are available to help you tag images automatically, I'd like to mention this in the article. I was recommended TagGUI a while ago and have been using it ever since. I thought I'd type up a quick, no-nonsense guide not only on some best tagging practices I've learned, but also on how to use TagGUI.

I intend to update this article with new information and suggestions. If you notice any information I've missed, something that seems incorrect, something that's unclear, etc., please feel free to comment so I can improve the article. Obviously, though, just don't be rude and keep any criticisms/feedback constructive!

Downloading TagGUI

This is simple: go to https://github.com/jhc13/taggui and download the file.

From here, unzip the folder. This will have the files/application you need to run TagGUI.

Getting started with TagGUI

Presuming you've prepared your dataset, you will now want to tag your images. In my previous guide (https://civitai.com/articles/7483/civitais-trainer-a-simple-beginners-guide-to-training-character-lora-using-it), I showed how to use CivitAI's autotagger. This guide, of course, will not focus on that, as I will be using TagGUI's Python-based autotagger.

After downloading TagGUI and extracting it, you'll be prompted to load a directory. You will want to navigate to the location of your dataset (the folder/folders you saved your images into). Keep in mind that, if you're at the root folder, it'll load all your images. If your character has multiple outfits/appearances, I definitely recommend creating separate folders for each one.

I'll also briefly mention here that if some of your images are lower resolution, it may benefit you to use an upscaler to render them at a higher resolution. I usually don't do this since I tend to have higher resolution images in the first place, but I've done it from time to time. I already have Topaz Gigapixel AI (a paid application), but chainNNer is a free alternative.

Anyway, once you've navigated to the location, you'll be presented with a screen that resembles the screenshot below. It's been so long since I set it up, but it should ask you which model to use. I've been using SmilingWolf/wd-eva02-large-tagger-v3 with no issues.

Regarding the differences between the models, CivitAI user draugrghost provided the following URL (warning: this is some pretty technical stuff!): https://l4.pm/wiki/Personal%20Wiki/AI%20stuff/understanding%20anime%20taggers.html

From here, you'll want to tag your images. For specific outfits, I'd recommend tagging them with a keyword. For instance, for this Riselia Ray Crystalia model, you may want to add 'R1s3l1aun1f0rm' to all images that include her uniform. Just type it out and slap that Enter button. You can also Shift-Click or Control-A to select multiple images and mass-add a single tag or a series of tags (use a comma to separate them, like you would when prompting for an image).

Once that's all good, select all images in question (click any image and press Ctrl+A) and then click the 'Start Auto-Captioning' button. This'll automatically start detecting the elements in your photo. TagGUI will start to create a .txt file that pairs with all your images in the folder you saved it to.

You can (and almost certainly will need to) manually delete any unneeded tags by selecting them from the menu on the top-right and pressing the delete key on your keyboard (or manually opening the accompanying .txt file and manually deleting and saving it).

After this is done, I like to create a folder called dataset and copy all the images from each folder into it. Then I right-click the folder > Compress To > and choose ZIP to create a .zip file. This .zip file can be used on sites like CivitAI itself, Google Colab, etc.

What not to do when tagging

Let's start with what not to do. Firstly, and this is something I've been guilty of for a while (I've been course-correcting recently), is including pre-tagged character names in your dataset. Websites like https://www.downloadmost.com/NoobAI-XL/danbooru-character will provide a list of characters with Danbooru-based tags that Illustrious already knows. TagGUI will often automatically grab them; I recommend typing in something for which there'll clearly be no tag. For example, in that screenshot above, I tagged Kaede Otori as K&3d30t0r1. While I don't think Kaede is even a pre-trained character, the principle should apply in general.

I'd also avoid getting too vague with your tagging. So, for example, instead of tagging an image with shirt, consider black shirt if that's what your character is wearing in the image. For certain outfits, I'd recommend adding a specific tag that, again, wouldn't be something that'd be pretrained on. For my recent Orihime Inoue character model, for example, I tagged all images of her white dress as 0r1h1m3dr3ss in tandem with other tags, like navel, cleavage, cleavage cutout, clothing cutout, midriff, hair ornament, and crop top. Tags like this, which are more specific, can help you generate more consistent outfit ideas by avoiding vagueness. To provide another example, if your character is, ahem, well-endowed, TagGUI may tag both breasts and large breasts on an image. Generally speaking, it'd make more sense to remove breasts in favor of large breasts since there's a bit of redundancy there.

I'd also remove tags that TagGUI inserts that don't really mean anything or are simply incorrect. For example, I've shown TagGUI add tags like virtual youtuber, 1boy, male focus, etc., when tagging an image of a single girl from an anime; these are clearly incorrect, so removing them ought to help with stability.

What to do when tagging

To say the inverse of what I said a moment ago, try to be relatively specific with your tags. However, I wouldn't get overly granular. So you probably don't need to type 'red shirt with a pocket on the left-hand side and there's a pencil sticking out of it' or something like that. Obviously, I just made up this example on the spot. In this case, tagging red shirt, shirt pocket, pencil should suffice, as the model ought to be able to learn that your character is wearing a red shirt with a pencil in their shirt pocket from these three separate tags.

Final thoughts