Free image taggers and tools that run in your browser
Three taggers (one picture at a time, character recognition, and a batch tagger for whole folders) plus three tools for cleaning up image collections. They all open your files locally, so nothing is uploaded.
Anime & Danbooru Tagger
Drop in one picture and get its Danbooru-style tags in seconds, ready to copy into a prompt or a caption.
Open the tool →Anime Character Tagger
Find out which anime character is in a picture. The model recognizes 2,132 popular characters.
Open the tool →Batch WD14 Tagger & Editor
Run WD14 tagger models on a whole folder, then search, filter and bulk-edit the tags before exporting .txt caption files.
Open the tool →Duplicate Image Finder
Group visually similar images with perceptual hashing. It catches resized, re-saved and lightly edited copies that file hashes miss.
Open the tool →Face & Expression Sorter
Split a batch into “has faces” and “no faces”, count faces per image, and optionally filter by detected expression.
Open the tool →Batch Color Fixer
Apply white balance, histogram normalization, brightness, contrast, saturation and gamma to many images at once.
Open the tool →Which tool do I need?
| If you want to… | Use |
|---|---|
| Get the Danbooru tags for one picture, e.g. for an AI art prompt | Anime & Danbooru Tagger |
| Find out which anime character is in a picture | Anime Character Tagger |
| Auto-tag a whole folder and save a caption file per image | Batch WD14 Tagger |
Edit, search or bulk-change existing .txt captions | Batch WD14 Tagger (drop images and captions together) |
| Remove resized, re-saved or nearly identical copies | Duplicate Image Finder |
| Keep only images with faces, or drop crowd shots | Face & Expression Sorter |
| Fix a color cast or even out exposure across a batch | Batch Color Fixer |
Using them together
If you are building a training set, the tools work best in a set order: deduplicate first so you don't tag the same image twice, filter by faces if your subject is people, correct color only if your source images are inconsistent, and tag last, once the set is final. A practical workflow for cleaning an image dataset walks through each stage and the mistakes to avoid.
What runs where
Every tool reads your files in the browser and processes them on your device. Some need a model download first:
| Tool | How it works | Extra download | What you get back |
|---|---|---|---|
| Anime & Character Taggers | The WD14 ConvNeXt V2 model run by ONNX Runtime Web in a background worker | 388 MB model from Hugging Face when you add your first picture, cached in your browser | Tags to copy, formatted for prompts or captions |
| Batch WD14 Tagger | A WD14 model run by ONNX Runtime Web in background workers | 365–455 MB model from Hugging Face, cached in your browser (shared with the taggers above) | A ZIP of .txt captions, with or without the original images |
| Duplicate Image Finder | SHA-256 for exact copies, then a perceptual hash | None | Your chosen originals, unchanged |
| Face & Expression Sorter | face-api.js face detector and optional expression model | Small model files from the face-api.js project | Your chosen originals, unchanged |
| Batch Color Fixer | Per-pixel math on a canvas | None | Every image re-encoded with your corrections |
There are no accounts, quotas or watermarks. The tools open on phones, but memory and screen space make a computer far more comfortable for real batches. How in-browser processing works covers the details, including how to check that nothing is uploaded.