Juzi the mascot

Credits & Licenses

The open projects JuziGenius™ is built on.

Learning Chinese with this app is only possible because other people gave their work away first — stroke-order data traced from decades-old fonts, sentences collected by the Tatoeba Project, dictionaries built over years. They are credited below, with their licenses.

Handwriting

Stroke-order data

Arphic Public License

Every stroke you trace comes from character outlines extracted from Arphic Technology's AR PL fonts, released in 1999.

© 1999 Arphic Technology Co., Ltd. — extracted by Make Me a Hanzi, packaged by hanzi-writer-data. Full license: ARPHICPL.TXT.

The set of characters this app uses is free for anyone to copy under the same license: juzigenius-data.

LXGW WenKai

SIL Open Font License 1.1

The handwritten-style characters on the victory card are set in LXGW WenKai, built on Klee One.

© 2021–2026 LXGW; © 2020 The Klee Project Authors. Source. Subset for this app. Full license: LXGWWenKai-OFL.txt.

Hanzi Writer

MIT

The library that animates stroke order and grades what you write.

Copyright © 2014 David Chanin — chanind.github.io/hanzi-writer

Feather icons

MIT

The settings gear and the grid button that switches between tools, in every tool, are Feather icons.

Copyright © 2013–2023 Cole Bemis — feathericons.com

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Sentences & dictionaries

Tatoeba

CC BY 2.0 FR

Most example sentences are Mandarin–English pairs from the Tatoeba Project. We have edited many of them: converted to simplified characters, corrected where a translation was wrong or a sentence read unnaturally, and with the names Tom and Mary changed to Xiao Ming and Xiao Hong.

tatoeba.org — license

Unicode Han Database (Unihan)

Unicode License

Character definitions, radicals and stroke counts.

© 1991–present Unicode, Inc. — unicode.org

CC-CEDICT

CC BY-SA 4.0

The word and character definitions you see — both the short English shown while you practice and the full list of meanings on the card at the end — are adapted from CC-CEDICT, the community Chinese–English dictionary, © MDBG and the CC-CEDICT contributors. Changes: filtered to the words this app teaches; classifier notes, pronunciation notes and cross-references removed; dictionary shorthand written out in full; tone numbers rewritten as tone marks; and several hundred definitions cleaned up or corrected. The adapted definitions stay under the same license.

mdbg.net/chinese · License

HSK vocabulary lists

Reference data

HSK levels are published by Hanban / Chinese Testing International. Level 4–6 word lists via clem109/hsk-vocabulary, from gigacool/hanyu-shuiping-kaoshi.

HSK 3.0 (nine-level) character and word lists

CC BY-SA 4.0

The levels of the November 2025 HSK syllabus, used for the HSK levels in Handwriting and for the Vocabulary Builder course, transcribed by Mani (krmanik). Joined into one list per level; nothing added, removed or moved.

github.com/krmanik/HSK-3.0 · License

Sound

Google Cloud Text-to-Speech

Google Cloud terms

The audio in Handwriting, Pinyin & Tones and Vocabulary Builder is spoken by two of Google's WaveNet voices, one female and one male, generated for this app. They are computer voices, not recordings of people. Sentences and words you add yourself are read by your own device's Chinese voice, and Character Components has no sound.

cloud.google.com/text-to-speech

Tools used to build the app's data

pypinyin, OpenCC

MIT · Apache 2.0

Pinyin readings come from pypinyin (MIT). Simplified-to-Traditional conversion uses tables from OpenCC (Apache 2.0), © BYVoid.

The data above is not ours to relicense and keeps its original terms — the full breakdown, including chain of provenance for each item, is in THIRD-PARTY-LICENSES.md.

Attribution mistakes are bugs. If something here is miscredited or missing, please write to support@juzigenius.com.

← Back to JuziGenius