Credits and licenses

ImgTagger is made of other people's work: open-source libraries that run in your browser, models that people trained and published, and photographs that their authors gave to everyone. This page names all of it, with the author, the license and a link to the source, because the people behind it deserve the credit and because you should be able to check the terms. We use all of it as published and do not change it.

Models · Code libraries · 7-Zip and libheif · Data · Pictures · Full license texts

Models

A model is a file of numbers that a program learned from examples. Your browser downloads each one from Hugging Face, a public model host, the first time you use a tool that needs it, and keeps it for later. The links point to the copy we load, pinned to one version. Two small sets of files are our own copies instead, marked below.

The tools that turn a picture into a sketch, make a color transparent, make edge and line-art maps (the depth map is the exception: see Depth Anything above), find duplicates, fix color, rotate, crop, resize, compress and combine pictures use ordinary image processing, with no model.

A note on how these models were trained: most were trained on large collections of pictures gathered for research (for example Places2 for the inpainting models, ImageNet for colorization and Danbooru for the taggers). The authors license the trained files as shown above. The terms of the datasets themselves are set by their owners, not by us; we mention them so that you know.

Code libraries

These make up the pages themselves or are loaded by them, unchanged. "From jsDelivr" means that your browser fetches the file from that public code host; everything else is served from this site. The text of every license is in third-party-notices.txt (below).

The site is built with Vite (MIT license), which is not part of what your browser loads. The stored-ZIP writer, the PDF writer and the image-metadata reader are our own code.

7-Zip and libheif (LGPL)

Two of the libraries are under the GNU Lesser General Public License, which lets anyone use them, including in a free site like this one, as long as the library stays replaceable and its source stays available. Here is how that works for each:

• 7-Zip (used by the 7z option of the Encrypted ZIP Maker) is by Igor Pavlov. Its WebAssembly build is the 7z-wasm package. We serve its two files, 7zz.es6.js and 7zz.wasm, exactly as published, from /vendor/7z-wasm/ (with its license and the unRAR license text), and our code loads them as separate files, so a different build of the library can take their place. The source code is at the 7z-wasm repository and 7-zip.org. The unRAR restriction forbids using the code to make a RAR compressor, which this site does not do.
• libheif (used to open iPhone HEIC photos in browsers that cannot) comes inside heic-to, loaded unmodified from jsDelivr, as its own file, only when a HEIC photo is added. The source code of libheif is at github.com/strukturag/libheif.

Data

• Tag lists. The tagger models' list of 9,083 tags (selected_tags.csv, Apache-2.0, by SmilingWolf), and the list of 2,132 character names on the character tagger page that comes from it. The tags themselves are the work of the volunteers who tag pictures on Danbooru; the current tag names and post counts shown by the Danbooru tagger were read from Danbooru's public API.
• Language data for reading text (19 languages): the Tesseract project's trained data, Apache-2.0, served from this site.
• Face models: see the model table above.

Pictures

The example pictures on the tool pages, and the pictures inside the demo clips, come from Wikimedia Commons. Every one is marked there as CC0 (dedicated to the public domain by its author) or in the public domain, so no credit is required, and we give one anyway. Some were first published on Unsplash before June 2017 or on Pixabay before 2019, when both sites released their pictures as CC0; Commons volunteers checked each of those against the original page. There are 40:

We crop and resize the pictures to fit, and a few clips use copies that were changed for the recording (tilted, given example camera data, saved at another size); the caption under each clip says so. The paintings in the face sorter clip are by Vermeer, Leonardo da Vinci and Rembrandt, long in the public domain. The pictures of drawings are by jsks, made with the ComiPo manga-making software and published on Pixabay. We chose paintings and drawings, not photographs of people, for the clip about faces. If you appear in any of these pictures and would like it removed, tell us and we will take it down. The round badge in the Image to SVG and Favicon examples, the pages of text in the PDF and text-reading clips, and the sample files made for the clips were drawn or written by us for this site.

Full license texts

third-party-notices.txt (177 KB, plain text) holds the license text of every library above and of the packages they depend on, with its author's copyright line, exactly as published, plus the list of models with their licenses. If something is missing or wrong on this page, please tell us and we will fix it.