What it is
キレ (Kire) is five tools that run inside your browser: image conversion between eleven formats, cropping to a spec — a passport photo, Instagram, YouTube, LinkedIn, each with a page of its own — GIF making, resizing and speeding, video trimming, converting, merging and compressing, and PDF merging, splitting, rotating, page order, images in and out, and compressing. There is no server and no file ever travels over the network. Every result carries a badge that says what happened to its bytes: copied without re-encoding, re-saved with every pixel the same, or re-saved with loss.
How it crops
A JPEG loses a generation of quality every time it is saved. Cropping on its 8 px or 16 px block grid is the exception: the compressed coefficients are copied as they are, so nothing has to be re-encoded. The command-line tool jpegtran does exactly this, and this site compiles the same unmodified C code (libjpeg-turbo’s transupp.c) to WebAssembly and runs it in the browser. For the same photo and the same region, the output has been verified by hash to be byte-identical to jpegtran’s.
Auto mode finds the face with OpenCV.js and the YuNet face-detection model and composes the crop around it. That model also runs only inside the browser and is downloaded only when you choose auto.
Who makes it
IMINADA builds and runs it. The company is named after the Iyo-nada (伊予灘), a stretch of the Seto Inland Sea; the drawing above is the waiting shelter on the platform of Shimonada station, which faces that sea.
License notices
This site includes the open-source software below, acknowledged here as each license requires.
- libjpeg-turbo — the lossless crop engine (transupp.c). IJG License, Modified (3-clause) BSD License, zlib License. This software is based in part on the work of the Independent JPEG Group.
- OpenCV — the face-detection runtime (OpenCV.js). Apache License 2.0.
- YuNet — the face-detection model (face_detection_yunet_2023mar.onnx, OpenCV Zoo). MIT License.
- react-easy-crop — the crop-box UI. MIT License.
- pdf-lib — reading and writing PDF files on the /pdf pages. MIT License.
- PDF.js — drawing PDF pages as images on the PDF to JPG and PDF to PNG pages. Apache License 2.0. It comes with its JPEG 2000 decoder (OpenJPEG, BSD 2-Clause), JBIG2 and fax decoder (PDFium, BSD 3-Clause — the Extract images from PDF page also uses it for black-and-white scans), colour management (qcms, MIT) and the fonts it substitutes for the standard PDF fonts (Foxit, BSD 3-Clause; Liberation, GPL 2.0 with the font exception). Each license text is served next to its files, under /pdfjs/.
- FFmpeg(n5.1.4, a build of this site’s own following ffmpeg.wasm’s recipe) — re-encoding video on the /video conversion, resize, merge and compress pages when the browser’s own codecs cannot. This build includes x264 and is distributed under the GNU General Public License, version 2 or later; the libraries in it (libvpx, LAME, Vorbis, Ogg, zlib) keep their own licences. The licence text and the full list are served next to it (/ffmpeg/COPYING.GPLv2, /ffmpeg/SOURCE.txt), and so is its complete source, every library at the commit it was built from: /ffmpeg/source/.
- libheif(libheif-js) — reading HEIC on the converter, with the libde265 decoder. GNU Lesser General Public License 3.0. It is served as its own files, /wasm/libheif.js and /wasm/libheif.wasm, not built into the site’s code, and its source at the versions built (libheif 1.23.2, libde265 1.0.15) is in /ffmpeg/source/ with the engine’s.
- @jsquash/avif, @jsquash/webp — AVIF and WebP encoding and decoding. Apache License 2.0; the codecs inside are libavif, libaom (BSD 2-Clause) and libwebp (BSD 3-Clause).
- fast-png, fflate, iobuffer, gifenc, UTIF — PNG, GIF and TIFF reading and writing. MIT License; pako (inside UTIF) MIT and zlib License.
- Next.js, React — MIT License.
- Emscripten — the WebAssembly toolchain. MIT License.
- Pretendard — the typeface. SIL Open Font License 1.1.
- Noto Sans SC — the Chinese characters on the Simplified Chinese pages. SIL Open Font License 1.1.