Open Source Credits

FFmpeg is the audio and video engine that powers our video and audio tools. The unmodified ffmpeg.wasm core is served from our static asset CDN (Cloudflare R2) and cached in your browser via the Cache Storage API after the first download, so subsequent video and workflow runs reuse it without re-downloading.

How we comply: We do not modify or statically link FFmpeg. The core is loaded as a separate WebAssembly module fetched at runtime from a static URL, which keeps the binary outside our application bundle and lets advanced users override it with a self-hosted copy. Source code for the upstream build is available at https://github.com/ffmpegwasm/ffmpeg.wasm and https://github.com/FFmpeg/FFmpeg.

libheif is used (via the heic-to wrapper) to decode HEIC/HEIF images into JPEG/PNG entirely in the browser.

How we comply: libheif is loaded as a separate WebAssembly module at runtime. We do not modify the library. Upstream source is available at the link above.

heic-to (v1.4.2) is the JavaScript wrapper around libheif used by our HEIC to JPG and HEIC to PDF tools. We use the CSP-safe build (heic-to/csp), which is libheif compiled with USE_UNSAFE_EVAL=0 so the converter works under our strict Content Security Policy.

How we comply: heic-to is loaded as a separate, dynamically-imported JavaScript chunk at runtime. We do not modify the library. Users can replace the chunk with a self-hosted copy. Upstream source and the LICENSE file are available at the link above.

Favtoo is served as static files from Cloudflare. Cloudflare may keep infrastructure-level access logs governed by their privacy policy. We do not operate our own servers and do not collect or store IP addresses ourselves.

Plausible counts page views without cookies and without storing IP addresses. The IP, user agent, and domain are hashed into a daily-rotating session identifier and the raw IP is discarded.

When enabled, AdSense is loaded only after the user grants consent in our cookie banner. AdSense is proprietary and not open source; it is listed here so all third-party network calls are visible in one place.

React framework for production-grade web applications

JavaScript library for building user interfaces

React renderer for the DOM

Typed superset of JavaScript

Utility-first CSS framework

Icon library used throughout the interface

Utility for constructing className strings

JavaScript wrapper for ffmpeg.wasm; runs FFmpeg in the browser

Utility helpers for the ffmpeg.wasm wrapper

WebAssembly build of FFmpeg loaded at runtime from our static asset CDN and cached in the browser

Create and modify PDF documents in JavaScript

PDF parser and renderer by Mozilla

JavaScript port of Google’s Tesseract OCR engine. Powers our OCR PDF tool. We mirror the unmodified WebAssembly module and its Emscripten loader script (sourced from naptha/tesseract.js-core, https://github.com/naptha/tesseract.js-core) and the gzipped LSTM language models (sourced from naptha/tessdata at the 4.0.0_best_int path, https://github.com/naptha/tessdata) on our own static asset CDN (Cloudflare R2). At runtime the loader script’s embedded relative .wasm path is rewritten in-memory to the absolute R2 URL so it works inside a sandboxed Web Worker; the WebAssembly bytes themselves are served byte-identical to upstream. Files are cached in the browser via the Cache Storage API and HTTP cache after the first download. Each language model (English, Spanish, French, German, Portuguese, Italian, Dutch, Russian, Polish, Turkish, Arabic, Hindi, Japanese, Korean, Chinese Simplified) is fetched on demand only when the user picks that language.

Decrypt password-protected PDFs in the browser

Encrypt PDFs with passwords in the browser

JavaScript text differencing implementation

Markdown parser and compiler

HTML to Markdown conversion

Note: Depends on @mixmark-io/domino (BSD-2-Clause).

QR code generator for browser and Node.js

Read and write Excel (.xlsx) workbooks

Note: Used for spreadsheet parsing. Not using SheetJS (xlsx package).

Raster to SVG tracing and vectorization

Barcode generator

Password strength estimation

Lexicographically sortable unique identifiers

GIF decoding library for parsing and extracting GIF frames

Fast GIF encoding library for creating animated GIFs

YAML parser and serializer for JavaScript

Lightweight TOML parser and serializer

Create, read, and edit ZIP files in the browser

Note: jszip is dual licensed. We use it under the MIT prong.

SQL query formatter supporting multiple SQL dialects

CSS minifier used by our developer tools

EXIF / metadata reader for JPEG, HEIC, TIFF, and more

Read and write EXIF metadata in JPEG files

TIFF image decoder/encoder used by our image tools

Note: Depends on pako (MIT AND Zlib).

Browser-side background removal using a U²-Net-family segmentation model running on ONNX Runtime Web. Powers our AI Background Remover.

Note: AGPL-3.0 obliges us to publish source for any modifications. We do not modify the library; the unmodified npm package and its model chunks are served from our CDN. Users can self-host their own copy by overriding the `publicPath` configuration option.

JavaScript port of HuggingFace Transformers, running ONNX-exported models in the browser. Powers our AI Subtitle Generator and AI Audio Transcriber via the Whisper-tiny model.

Note: Bundles its own onnxruntime-web 1.14.0. Model files (Whisper-tiny config, tokenizer, encoder/decoder ONNX) are mirrored on our R2 CDN under ai/whisper/v1/Xenova/whisper-tiny/. We do not modify the library or the model.

Microsoft ONNX Runtime compiled to WebAssembly. Powers AI Image Upscaler, Pose Detection, Smart Auto-Crop, Silero VAD inference (Silence Cutter, Filler-Word Remover).

Note: Loaded as a separate WebAssembly chunk from our R2 CDN at runtime. Unmodified upstream build. Multi-threading uses the standard SharedArrayBuffer + Web Workers pattern; the cross-origin isolation it requires is set up in public/_headers (COOP/COEP).

Maintained fork of face-api.js. Bundles TensorFlow.js and ships SSD MobileNet face-detection weights. Powers our AI Face Anonymizer and AI Profile Picture Maker.

Note: The library is loaded as a dynamic JS chunk; the model weights (manifest + binary shard) are mirrored on our R2 CDN under ai/face-api/v1/. Detection runs locally; faces are never transmitted off-device. We do not modify the library.

Pure-JavaScript .docx-to-HTML converter. Powers our DOCX to PDF tool — mammoth extracts the document content, then our pdf-lib pipeline renders the HTML to PDF.

Note: Loaded as a separate dynamically-imported chunk. Document parsing happens entirely in your browser; the .docx contents never reach a server.

OpenCV computer-vision library compiled to WebAssembly. Powers our Document Scanner (edge detection, contour finding, perspective warp).

Note: Loaded via a dynamic <script> tag with crossorigin="anonymous" so it satisfies our COEP=require-corp policy via standard CORS fetches. The unmodified opencv.js bundle (which embeds its WebAssembly inline as base64) is mirrored on our R2 CDN.

Recurrent-neural-network speech denoiser by Xiph.org. We use the pre-built jitsi/rnnoise-wasm distribution. Powers our AI Noise Reducer.

Note: The unmodified rnnoise.wasm binary (~110 KB) is mirrored on our R2 CDN. Audio frames are processed entirely in your browser — the denoiser is a pure DSP operation with no network calls.

Silero Voice Activity Detector. Powers our Silence Cutter and Filler-Word Remover.

Note: The unmodified silero_vad.onnx file (~2 MB) is mirrored on our R2 CDN. Inference runs locally via onnxruntime-web. The model has no telemetry; audio frames are classified as speech / non-speech entirely on-device.

Salient-object detection model. We use the ONNX export from danielgatis/rembg. Powers our Smart Auto-Crop tool.

Note: The unmodified u2netp.onnx file (~4.6 MB) is mirrored on our R2 CDN. Saliency maps are computed entirely in your browser via onnxruntime-web; image bytes never leave the device.

Practical Image Super-Resolution by xinntao et al. We use a community ONNX export of the x2plus checkpoint. Powers our AI Image Upscaler.

Note: The unmodified RealESRGAN_x2plus ONNX model (~67 MB) is mirrored on our R2 CDN. Inference runs locally via onnxruntime-web — full upscaling is performed on-device with no server-side rendering.

Google’s real-time pose-estimation model, ported to ONNX by the Xenova HuggingFace mirror. Powers our AI Pose Detection tool.

Note: The unmodified MoveNet Lightning ONNX (~9 MB) is mirrored on our R2 CDN. The 17-keypoint inference runs locally via onnxruntime-web — photos never leave the device.

OpenAI Whisper-tiny speech recognition model, ONNX-exported by the Xenova HuggingFace community. Powers AI Subtitle Generator, AI Audio Transcriber, and Filler-Word Remover.

Note: The unmodified config + tokenizer + ONNX weights (~42 MB total) are mirrored on our R2 CDN under ai/whisper/v1/Xenova/whisper-tiny/. Transcription runs locally via @xenova/transformers — audio bytes never reach a server.

Self-hosted under public/fonts/. The OFL license file ships alongside the .woff2 files.