NoAdsTools

NoAdsTools is a privacy-first, client-side image editor and batch processor in the NoAds suite. Image files never leave the browser. Edit one image or a hundred — crop, resize, rotate/flip, brightness/contrast/saturation, blur, color-to-transparent, text overlays, freehand brush, shapes, blur/pixelate redaction, ML background removal, smart compression and "smallest size" preset, EXIF metadata stripping, auto-trim of transparent or solid-color borders, HEIC/HEIF import, and export to PNG, JPG, WebP, or PDF — per image or as a ZIP. All locally.

Privacy

Image files never leave the browser. The site fetches its own code, fonts, and ML model assets from this origin only. No third-party CDNs, no analytics, no tracking, no cookies. See privacy.html or the in-app privacy panel for the full disclosure (including the exact localStorage keys we set and which vendored libraries load lazily on first use).

Vendored ML assets (repo size note)

This repo ships ~118 MB of pre-vendored binary assets under js/vendor/bgremove/ so that the in-browser "Remove background" feature works end-to-end after git clone && npm install with no separate model-download step. With the v1.1 additions (jsPDF ~420 KB and libheif ~1.1 MB), the total disk footprint of the working tree is ~135 MB; including .git history the on-disk repo is ~245 MB.

That's:

• The @imgly/background-removal ESM bundle (~170 KB, AGPL-3.0) and its license text.
• The ISNET fp16 segmentation model — content-addressable chunks fetched from staticimgly.com/@imgly/background-removal-data/1.7.0/ once at install time, then vendored.
• The ONNX Runtime Web SIMD WASM kernels — both the CPU path and the WebGPU/JSEP path — also as content-addressable chunks alongside the model. The browser only downloads the variant it actually uses at runtime.
• The matching onnxruntime-web@1.21.0 ESM bundles (CPU + WebGPU, ~800 KB total, MIT) under js/vendor/onnxruntime-web/, resolved at runtime via an import map in index.html.
• The jsPDF UMD bundle (~420 KB, MIT) under js/vendor/jspdf/ for image-to-PDF export.
• The libheif-js decoder (~1.1 MB JS + WASM, LGPL-3.0) under js/vendor/heic/ for HEIC/HEIF import. Loaded on first .heic open, gated by a one-time consent modal.
• The JSZip library (~97 KB, MIT) at js/vendor/jszip.min.js for batch ZIP export.
• The MediaPipe BlazeFace face-detection ONNX export (~600 KB, Apache-2.0) under js/vendor/blazeface/, re-distributed by Qualcomm AI Hub Models. Powers the "Auto-detect faces" button in the redact tool. Tile-based multi-scale scanning catches small faces in group photos.
• The Tesseract.js v7 OCR engine (~22 MB across 6 LSTM-only WASM variants + English eng.traineddata.gz, all Apache-2.0) under js/vendor/tesseract/. Powers the "Detect text" button + preview-select mode with PII regex auto-marking.
• The ONNX Runtime Web threaded JSEP WASM kernels (~36 MB across ort-wasm-simd-threaded.{jsep,}.{wasm,mjs}) also vendored under js/vendor/onnxruntime-web/ — required by both BlazeFace and the existing background-removal model at runtime.

A clone of this repo is consequently larger and slower than a typical static-site repo. The trade is intentional: zero third-party CDN reliance, self-hosted everything, and a deploy that works immediately on a freshly cloned site. See THIRD-PARTY.md and js/vendor/bgremove/.notice for the full inventory + re-vendoring instructions.

What we ship

• vanilla JS, no framework, no build step — native ES modules and a single CSS file.
• Self-hosted Onest font (variable, OFL) under fonts/.
• Vendored libraries under js/vendor/: bgremove/ (@imgly/background-removal + ISNET model + ONNX runtime WASM), onnxruntime-web/ (peer ESM bundle + the threaded JSEP WASM kernels shared by BlazeFace + bg-remove), blazeface/ (face detection ONNX, ~600 KB), tesseract/ (OCR engine + English language data, ~22 MB), jspdf/ (PDF export), heic/ (HEIC decoder), jszip.min.js (ZIP batch export).
• PWA manifest + icons — installable on desktop and mobile. Icons live at img/logo.svg, img/icon-192.png, img/icon-512.png, img/icon-512-maskable.png, img/apple-touch-icon.png (all generated from img/logo.svg via node scripts/build-icons.mjs).
• No analytics, no telemetry, no third-party requests at runtime.

License

This project is licensed under the GNU AGPL v3.0 — see LICENSE. The AGPL's source-availability requirement means the running site links to its own source repository in the footer and privacy panel.

Source: https://github.com/Dan512/NoAdsTools

Local dev setup

Requires Node 22+.

• nvm use && npm install — install dev deps (Playwright, axe-core)
• npm test — unit tests via Node's built-in test runner
• npm run test:browser — Playwright browser tests (after npm run test:browser:install to fetch browser binaries)
• npm run serve — local static dev server on http://localhost:4173
• node scripts/build-icons.mjs — re-render img/icon-*.png from img/logo.svg (committed to the repo; only re-run if the logo changes)
• node scripts/install-blazeface.mjs — re-fetch the BlazeFace face-detect ONNX from Qualcomm AI Hub's S3 release bucket (already committed; only re-run after bumping QAI_VERSION)
• node scripts/install-tesseract.mjs — re-fetch Tesseract.js v7 + tesseract.js-core v7 + English eng.traineddata (already committed; only re-run after bumping the pinned versions)
• node scripts/install-ort.mjs — re-fetch the ONNX Runtime Web threaded JSEP WASM kernels (already committed; only re-run after bumping ORT_VERSION)
• node scripts/measure-weight.mjs — print the initial-load wire byte count (raw + gzipped) for the static page