PLEN

Licenses in Fumiko

Models

  • Hy-MT2 7B and Hy-MT2 1.8B (Tencent) – Apache 2.0. License text: Hy-MT2 7B, Hy-MT2 1.8B (Hugging Face site).
  • Index-Translate 2B and Index-Translate 2B (small) – Apache 2.0. Model card: Index-Translate 2B, Index-Translate 2B (small) (Hugging Face site).
  • Speech bubble detection: comic-text-and-bubble-detector (ogkalu) – Apache 2.0.
  • Manga text recognition: PP-OCRv6_manga (Kellenok) – Apache 2.0. The model uses the Manga109-s dataset.
  • Other text recognition: PaddlePaddle PP-OCRv6 models (tiny, small) and PP-OCRv5 mobile models (Korean, Latin, Cyrillic, Arabic, Devanagari, Thai, Greek, Tamil, Telugu) – Apache 2.0.

Libraries

  • llama.cpp and ggml – MIT
  • cpp-httplib and nlohmann/json (part of llama.cpp) – MIT
  • subprocess.h (part of llama.cpp) – public domain
  • KleidiAI – Apache 2.0
  • ONNX Runtime – MIT
  • jsoup – MIT
  • libc++ (LLVM) – Apache 2.0 with LLVM exceptions
  • Kotlin, kotlinx.coroutines and AndroidX – Apache 2.0
  • Clipper (parts ported to Kotlin) – Boost Software License 1.0
  • OpenCV (parts ported to Kotlin) – Apache 2.0
  • OpenCL headers (Khronos) – Apache 2.0, only when building the app. On the phone, Fumiko uses the phone's own OpenCL driver.

Data

  • The language detection tables were built from Tatoeba data (CC BY 2.0 FR, part CC0) and Unihan data (Unicode License v3).

Fonts

  • Zen Maru Gothic (the shape of the 文 character in the logo) – SIL Open Font License 1.1
  • Comic Neue Bold (typesetting of the translation in speech bubbles) – SIL Open Font License 1.1