Image tagging & dataset guides
Practical explainers written alongside the ImgTagger tools: how booru tags work, how to pick a tagger model and threshold, how to caption a training set, and how to clean a messy image folder before you start.
Booru tags & auto-tagging
How the Danbooru tag vocabulary works and how to get good results from a WD14 tagger.
Danbooru tags, explained
The categories, ratings, naming rules and conventions behind the tag vocabulary most anime image models are trained on.
Read the guide →Common booru tags: a reference glossary
Plain-English meanings, real post counts and common mix-ups for the everyday Danbooru tags a WD14 tagger outputs.
Read the guide →WD14 tagger models compared: ConvNeXt V2 vs SwinV2 vs ViT
Download sizes, tag lists, character support and published validation scores for the three WD14 models ImgTagger offers.
Read the guide →How to choose a confidence threshold for auto-tagging
What the scores mean, how the cut-off trades missed tags for wrong ones, and sensible starting values for different jobs.
Read the guide →Captioning training datasets
Turning raw tags into captions a LoRA or fine-tune can actually learn from.
How to caption a LoRA training dataset with tags
A start-to-finish workflow: trigger words, what to tag and what to leave out, reviewing the batch, and exporting one .txt per image.
Read the guide →Pruning and blacklisting tags: what to keep and what to cut
What to cut from auto-tagger output before training, how a blacklist differs from removing tags later, and starter blacklists.
Read the guide →Caption .txt sidecar files: format, naming and bulk editing
How sidecar caption files are named and formatted, how ImgTagger reads and writes them, and how to fix the usual file mistakes.
Read the guide →Booru tags vs natural-language captions: which should you use?
Tags or sentences? How each caption style handles counting, relationships and token limits, and how to match it to your base model.
Read the guide →Cleaning image collections
Removing duplicates, filtering by faces, and evening out color before you tag.
A practical workflow for cleaning an image dataset
The order we recommend for turning a messy image folder into a clean, captioned training set, with what to check at each stage.
Read the guide →How perceptual hashing finds near-duplicate images
How average and DCT hashes spot resized and re-saved copies, with a worked example and what the threshold numbers really mean.
Read the guide →Using face detection to curate image collections
Detection vs recognition, what the settings trade off, and recipes for keeping portraits, dropping crowd shots and balancing expressions.
Read the guide →White balance and histogram normalization for batch photos
Color casts, gray-world auto white balance, histogram stretching and gamma, with worked numbers and the cases where you should leave color alone.
Read the guide →Beyond anime tags
Other kinds of image tagging, and how ImgTagger keeps your files on your device.
Image keywords for SEO: alt text, filenames and metadata
Alt text, filenames, captions and IPTC metadata: what Google says it uses, and why one image needs different tags for different jobs.
Read the guide →How in-browser AI works, and why your images stay local
What ImgTagger downloads, where the model is stored, what it costs in memory and CPU, and how to confirm in DevTools that nothing is uploaded.
Read the guide →