Open source notices

This page is about the open-source components inside Heeler, not about the licence that unlocks your copy. For Heeler's own terms, see the End User License Agreement.

Heeler is built on other people's work. This page says whose, under what terms, and where to get the parts you are entitled to.

LibRaw

Heeler reads RAW files with LibRaw 0.22.2, used under the Common Development and Distribution License, version 1.0 (CDDL-1.0). LibRaw is offered under either the LGPL or the CDDL; Heeler takes the CDDL option, and contains no GPL-family code.

The CDDL asks that the source for the covered code be available to anyone who receives a binary built from it. The exact source Heeler compiles, unmodified from the upstream 0.22.2 release, is published at:

https://www.heeler.app/source/libraw/

That archive is the tree this build was made from, not a pointer at whatever upstream happens to be today. Its SHA-256 is published beside it so you can check the two match.

The full licence text is in this app, on the CDDL 1.0 page beside this one, and it ships with the source as LICENSE.CDDL in the root of that archive. Both are the same text: the page is generated from the tree Heeler is built from, and a test compares them.

zlib and libjpeg

Deflate-compressed and lossy DNG files need these two, and both are linked in.

zlib, © 1995-2024 Jean-loup Gailly and Mark Adler, under the zlib licence.
mozjpeg, a libjpeg-turbo derivative, © the Independent JPEG Group and contributors, under the IJG and BSD-3-Clause licences.

The IJG licence asks for one sentence in particular, so here it is in the words it asks for:

This software is based in part on the work of the Independent JPEG Group.

Neither library has been altered.

JPEG XL

Adobe's newer DNG files, both the lossy ones and every Lightroom HDR merge, store their pixels as JPEG XL. Heeler reads those with jxl-oxide, © its authors, under the MIT or Apache-2.0 licences, at your option.

The library has not been altered.

Lensfun lens database

Heeler ships a snapshot of the lensfun project's lens database: the XML files that describe lenses and their optical corrections, used to recognise the lens a photograph was taken with. The database is used under the Creative Commons Attribution-Share Alike 3.0 licence.

Only the database is included. The lensfun software library is not part of Heeler: the files are read by Heeler's own code. The bundled snapshot is unmodified; if Heeler ever ships an edited copy, the edits will be published under the same licence, as Share Alike requires. The snapshot's origin and revision are recorded in the lensfun-db folder that ships inside the app.

Machine-learning models (downloaded, not bundled)

The smart selection, depth and noise-reduction tools use machine-learning models that are not part of the installer. Each is downloaded only when you ask, from the source the download dialog names, and verified against a checksum pinned in the app before it is trusted. Every model also has a second source, a mirror of the same published files on Vagabond Burro's own release page, tried when the first does not answer; the checksum is the same either way. A downloaded model lives in Heeler's data folder and can be installed, updated or removed from Preferences (see Downloaded models).

MobileSAM (the Segment Anything image encoder, distilled) and the Segment Anything mask decoder, © Meta AI and the MobileSAM authors, under the Apache License 2.0: code and weights alike. The training data (SA-1B) is published by Meta under its own stated terms.
BiRefNet Lite (the one-shot subject matte), © Zheng Peng and the BiRefNet authors, under the MIT license: code and weights alike. Its training corpus (DIS5K and related academic datasets) is academic in origin; the weights are published by the author under MIT, and Heeler uses them as published.
LaMa (the fill behind Remove and the Finish Fill brush), © Samsung AI Center and the LaMa authors, under the Apache License 2.0: code and weights alike; ONNX packaging by Carve, also Apache-2.0. Its training corpus (Places2) is academic in origin, with research-oriented terms; the weights are published under Apache-2.0 and Heeler uses them as published. The fill only ever synthesizes pixels inside a mask you made on purpose, and only on your machine.
ViTMatte (the Smart matte in Selection Polish), © the Hust Vision Lab and the ViTMatte authors: the code under the MIT license, the published weights under the Apache License 2.0; ONNX packaging by Xenova. Its training corpus (Composition-1k) is academic in origin; the weights are published as stated on the author's model card, and Heeler uses them as published. The matte reads only the photograph and your own selection, on your machine.
Depth Anything V2 Small (the scene-depth plane behind Fog, Depth Lighting, Depth of Field, and depth-range selections), © the Depth Anything authors, under the Apache License 2.0: code and weights alike; ONNX packaging by the Hugging Face onnx-community. Only the Small variant is used: the Base and Large weights carry a non-commercial (CC-BY-NC) license and are deliberately not offered. The plane is relative depth computed from your photograph, on your machine.
SCUNet (Noise Reduction's Model method), © Kai Zhang and the SCUNet authors, under the Apache License 2.0: code and weights alike (the weight release in the KAIR repository is under MIT); ONNX packaging by Heliosoph, also Apache-2.0. It was trained on synthesised noise over public image sets, academic in origin; the weights are published as stated and Heeler uses them as published. The model reads only your photograph, on your machine.

The models are run by ONNX Runtime (© Microsoft, MIT) through the ort bindings (MIT or Apache-2.0). Nothing your photographs contain is ever sent anywhere: every model runs on your machine.

Rust and JavaScript libraries

Heeler links a number of open-source Rust crates and bundles a few JavaScript libraries, essentially all of them under the MIT or Apache-2.0 licences. Both ask only for attribution.

Terms and warranties

The components above are provided by their own authors as is, with no warranty of any kind. Their licences say so, and nothing here changes that.

Any warranty, support or liability that Heeler offers you is offered by Heeler alone. It does not come from LibRaw LLC, the Independent JPEG Group, or any other author whose work is named on this page, and none of them takes on any obligation to you because their code is inside Heeler. Heeler indemnifies them against any claim arising from terms Heeler has offered.

Nothing in Heeler's own licence limits or alters your rights in the source code covered by the CDDL. Where the two ever appear to disagree about that source, the CDDL wins.

Heeler itself

Heeler is not open source. Nothing on this page grants any right to Heeler's own code; it describes the third-party components inside it and the terms those components carry.

HEIC / HEIF photographs

Heeler contains no HEVC decoder of its own. HEIC files are decoded by the operating system's own codec, under the licence the OS vendor already pays for:

macOS: Apple's ImageIO framework, the same system codec Photos and Preview use, licensed by Apple as part of the operating system.
Windows: Windows Imaging Component, which decodes HEIC when Microsoft's HEVC Video Extensions and HEIF Image Extension are installed from the Microsoft Store. Without them Heeler reports HEIC files as unsupported; installing them is all that is needed, and nothing about Heeler changes.

On platforms whose operating system provides no HEVC codec, Heeler does not open HEIC files rather than shipping a decoder it would need patent licences to distribute.