Remove Background: cut out or change a photo's background, on the device

.gitattributes

@@ -0,0 +1,4 @@
+# AI model files are binary: never touch their line endings.
+*.onnx binary
+*.onnx.part* binary
+*.wasm binary

README.md

@@ -8,15 +8,15 @@
 <h3 align="center">Free, private tools for PDF, images, audio, Word, Excel and PowerPoint that never see your files.</h3>
 
 <p align="center">
-  64 tools — compress, convert, edit, redact, merge, split, cut, OCR, transcribe — running <strong>100% in your browser</strong>.<br>
+  65 tools — compress, convert, edit, redact, remove backgrounds, merge, split, cut, OCR, transcribe — running <strong>100% in your browser</strong>.<br>
   <strong>No uploads. No sign-up. No watermark. No limits. Free forever. 16 languages.</strong>
 </p>
 
 <p align="center"><sub>Open source. Edits PDF text in the file’s own font, transcribes audio with Whisper on your device, and never hands your files to anyone.</sub></p>
 
 <p align="center">
   <a href="https://fizzdoc.com"><strong>Open Fizzdoc →</strong></a> &nbsp;·&nbsp;
-  <a href="#-all-64-tools"><strong>All tools</strong></a> &nbsp;·&nbsp;
+  <a href="#-all-65-tools"><strong>All tools</strong></a> &nbsp;·&nbsp;
   <a href="#-how-it-works"><strong>How it works</strong></a> &nbsp;·&nbsp;
   <a href="CONTRIBUTING.md"><strong>Contribute</strong></a>
 </p>
@@ -27,7 +27,7 @@
   <a href="LICENSE"><img src="https://img.shields.io/badge/license-Apache--2.0-22c55e?style=flat-square" alt="Apache-2.0 license"></a>
   <img src="https://img.shields.io/badge/bytes%20uploaded-0-16a34a?style=flat-square" alt="0 bytes uploaded">
   <img src="https://img.shields.io/badge/price-free%20forever-16a34a?style=flat-square" alt="Free forever">
-  <img src="https://img.shields.io/badge/tools-64-e5322d?style=flat-square" alt="64 tools">
+  <img src="https://img.shields.io/badge/tools-65-e5322d?style=flat-square" alt="65 tools">
   <img src="https://img.shields.io/badge/languages-16-2f6fdb?style=flat-square" alt="16 languages">
   <a href="CONTRIBUTING.md"><img src="https://img.shields.io/badge/PRs-welcome-8b5cf6?style=flat-square" alt="PRs welcome"></a>
 </p>
@@ -53,7 +53,7 @@ Most free online file tools ask you to upload your contract, payslip, bank state
 | What you get | |
 |---|---|
 | 🔒 **Your file stays on your device** | No upload endpoint exists, and the CSP blocks connections to any other site |
-| 🧰 **64 tools in one place** | PDF, Word, Excel, PowerPoint, images, audio, JSON and Mermaid |
+| 🧰 **65 tools in one place** | PDF, Word, Excel, PowerPoint, images, audio, JSON and Mermaid |
 | ✏️ **Edit PDF text in its own font** | Changed lines are rewritten inside the PDF with its embedded font, size and colour |
 | 🎙️ **Speech to text on the device** | Whisper runs in the browser: TXT, SRT or VTT, offline after a one-time download |
 | 🆓 **Free forever** | No account, no daily limits, no watermark, no premium tier |
@@ -108,14 +108,15 @@ Most free online file tools ask you to upload your contract, payslip, bank state
 | **Edit text** in a PDF, or **black out** personal details for good | [Edit PDF](https://fizzdoc.com/edit-pdf/) · [Redact PDF](https://fizzdoc.com/redact-pdf/) |
 | Turn a PDF into an **editable Word** file, or Word into PDF | [PDF to Word](https://fizzdoc.com/pdf-to-word/) · [Word to PDF](https://fizzdoc.com/word-to-pdf/) |
 | **Remove the password** from a bank statement, or add one | [Unlock PDF](https://fizzdoc.com/unlock-pdf/) · [Protect PDF](https://fizzdoc.com/protect-pdf/) |
+| **Remove the background** from a photo, or put white, a colour, a blur or another photo behind, at full resolution | [Remove Background](https://fizzdoc.com/remove-background/) |
 | **Copy text** from a scanned PDF, a photo or a screenshot | [OCR PDF](https://fizzdoc.com/ocr-pdf/) · [Image to Text](https://fizzdoc.com/image-to-text/) |
 | Combine **photos or screenshots into one PDF** | [JPG to PDF](https://fizzdoc.com/jpg-to-pdf/) · [PNG to PDF](https://fizzdoc.com/png-to-pdf/) |
 | Convert **everyday formats**: PNG ↔ JPG, WebP → JPG, Excel ↔ CSV | [Convert Image](https://fizzdoc.com/convert-image/) · [Excel to CSV](https://fizzdoc.com/excel-to-csv/) · [CSV to Excel](https://fizzdoc.com/csv-to-excel/) |
 | Turn a spreadsheet into **JSON for an API**, or open JSON in Excel | [Excel to JSON](https://fizzdoc.com/excel-to-json/) · [JSON to Excel](https://fizzdoc.com/json-to-excel/) |
 | Export a **Mermaid diagram as PNG or SVG** for slides, docs or a README | [Mermaid to PNG](https://fizzdoc.com/mermaid-to-png/) · [Mermaid to SVG](https://fizzdoc.com/mermaid-to-svg/) |
 | **Remove hidden author and company names** before sharing an Office file or PDF | [Word](https://fizzdoc.com/remove-word-metadata/) · [Excel](https://fizzdoc.com/remove-excel-metadata/) · [PowerPoint](https://fizzdoc.com/remove-powerpoint-metadata/) · [PDF](https://fizzdoc.com/remove-pdf-metadata/) |
 
-## 🧰 All 64 tools
+## 🧰 All 65 tools
 
 | Format | Tools |
 |---|---|
@@ -124,7 +125,7 @@ Most free online file tools ask you to upload your contract, payslip, bank state
 | **Word** | [Word → PDF](https://fizzdoc.com/word-to-pdf/) · [Compress](https://fizzdoc.com/compress-word/) · [Remove metadata](https://fizzdoc.com/remove-word-metadata/) · [Extract images](https://fizzdoc.com/extract-images-from-word/) |
 | **Excel** | [Excel → CSV](https://fizzdoc.com/excel-to-csv/) · [CSV → Excel](https://fizzdoc.com/csv-to-excel/) · [Excel → JSON](https://fizzdoc.com/excel-to-json/) · [JSON → Excel](https://fizzdoc.com/json-to-excel/) · [Compress](https://fizzdoc.com/compress-excel/) · [Remove metadata](https://fizzdoc.com/remove-excel-metadata/) · [Extract images](https://fizzdoc.com/extract-images-from-excel/) |
 | **PowerPoint** | [Compress](https://fizzdoc.com/compress-powerpoint/) · [Remove metadata](https://fizzdoc.com/remove-powerpoint-metadata/) · [Extract images](https://fizzdoc.com/extract-images-from-powerpoint/) |
-| **Images** | [Compress](https://fizzdoc.com/compress-image/) (to [20 KB](https://fizzdoc.com/compress-image-to-20kb/), [50 KB](https://fizzdoc.com/compress-image-to-50kb/), [100 KB](https://fizzdoc.com/compress-image-to-100kb/) for forms) · [Resize](https://fizzdoc.com/resize-image/) (larger or smaller) · [Convert](https://fizzdoc.com/convert-image/) · [PNG → JPG](https://fizzdoc.com/png-to-jpg/) · [JPG → PNG](https://fizzdoc.com/jpg-to-png/) · [WebP → JPG](https://fizzdoc.com/webp-to-jpg/) · [JPG → WebP](https://fizzdoc.com/jpg-to-webp/) · [Image → Text](https://fizzdoc.com/image-to-text/) (select text right on the photo, like Live Text) · [Mermaid → PNG](https://fizzdoc.com/mermaid-to-png/) · [Mermaid → SVG](https://fizzdoc.com/mermaid-to-svg/) (diagrams from code, with a preview and ready-to-use examples) |
+| **Images** | [Remove background](https://fizzdoc.com/remove-background/) (AI on your device, full resolution) · [Compress](https://fizzdoc.com/compress-image/) (to [20 KB](https://fizzdoc.com/compress-image-to-20kb/), [50 KB](https://fizzdoc.com/compress-image-to-50kb/), [100 KB](https://fizzdoc.com/compress-image-to-100kb/) for forms) · [Resize](https://fizzdoc.com/resize-image/) (larger or smaller) · [Convert](https://fizzdoc.com/convert-image/) · [PNG → JPG](https://fizzdoc.com/png-to-jpg/) · [JPG → PNG](https://fizzdoc.com/jpg-to-png/) · [WebP → JPG](https://fizzdoc.com/webp-to-jpg/) · [JPG → WebP](https://fizzdoc.com/jpg-to-webp/) · [Image → Text](https://fizzdoc.com/image-to-text/) (select text right on the photo, like Live Text) · [Mermaid → PNG](https://fizzdoc.com/mermaid-to-png/) · [Mermaid → SVG](https://fizzdoc.com/mermaid-to-svg/) (diagrams from code, with a preview and ready-to-use examples) |
 | **Audio** | [Cut & Split Audio](https://fizzdoc.com/split-audio/) (trim or split an MP3/M4A on a waveform, preview every part) · [Merge Audio](https://fizzdoc.com/merge-audio/) (listen before you download) — no re-encoding, so no quality loss · [Audio to Text](https://fizzdoc.com/audio-to-text/) · [MP3 to Text](https://fizzdoc.com/mp3-to-text/) · [Video to Text](https://fizzdoc.com/video-to-text/) · [Subtitle Generator](https://fizzdoc.com/subtitle-generator/) (SRT/VTT) — speech to text with OpenAI's Whisper running on your device; the model downloads once, then works offline |
 
 **Google Docs, Sheets and Slides** work too: download as .docx / .xlsx / .pptx and use the matching tool. Nothing is sent to Google.
@@ -289,6 +290,10 @@ Fizzdoc would not exist without these open-source projects and the people who ma
 | [pdf.js](https://github.com/mozilla/pdf.js) | Renders pages, extracts text, powers the editor | [Mozilla](https://github.com/mozilla) and contributors |
 | [pdf-lib](https://github.com/Hopding/pdf-lib) | Writes and edits PDFs, stamps, the OCR text layer | [@Hopding](https://github.com/Hopding) |
 | [Tesseract](https://github.com/tesseract-ocr/tesseract) and [tesseract.js](https://github.com/naptha/tesseract.js) | Text recognition, compiled to WebAssembly | [tesseract-ocr](https://github.com/tesseract-ocr) and [naptha](https://github.com/naptha) |
+| [BiRefNet](https://github.com/ZhengPeng7/BiRefNet) | The AI model that removes backgrounds (BiRefNet-lite, 512 px export by [studioludens](https://huggingface.co/studioludens/birefnet-lite-512)) | [@ZhengPeng7](https://github.com/ZhengPeng7) and co-authors |
+| [U²-Net](https://github.com/xuebinqin/U-2-Net) | The light background model for phones with little memory | [@xuebinqin](https://github.com/xuebinqin) |
+| [ONNX Runtime Web](https://github.com/microsoft/onnxruntime) | Runs the AI models in the browser, on the GPU (WebGPU) or CPU | [Microsoft](https://github.com/microsoft) and contributors |
+| [Whisper](https://github.com/openai/whisper) and [Transformers.js](https://github.com/huggingface/transformers.js) | Speech to text, on the device | [OpenAI](https://github.com/openai) and [Hugging Face](https://github.com/huggingface) |
 | [fflate](https://github.com/101arrowz/fflate) | Reads and writes Word, Excel, PowerPoint and ZIP files | [@101arrowz](https://github.com/101arrowz) |
 | [Inter](https://github.com/rsms/inter) | The typeface | [@rsms](https://github.com/rsms) |
 | [Archify](https://github.com/tt-a1i/archify) | Generates the interactive architecture map | [@tt-a1i](https://github.com/tt-a1i) |

THIRD_PARTY_NOTICES.md

@@ -15,6 +15,7 @@ The deployed site ships these third-party components in `assets/`:
 | English language data (@tesseract.js-data/eng, tessdata 4.0.0 best_int) | Apache-2.0 data, MIT package | https://github.com/tesseract-ocr/tessdata |
 | Mermaid 12.0.0 (Mermaid to Image page only), bundling d3, dagre-d3-es, cytoscape, chevrotain, DOMPurify, KaTeX, marked, roughjs, dayjs, lodash-es, stylis, khroma, elkjs and others | MIT, ISC, BSD-3-Clause, Apache-2.0, Unlicense; DOMPurify used under Apache-2.0 (dual MPL-2.0 OR Apache-2.0); elkjs EPL-2.0, source at https://github.com/kieler/elkjs | https://github.com/mermaid-js/mermaid |
 | Whisper tiny speech model (OpenAI; ONNX conversion by onnx-community) and Transformers.js 4.3.0 with ONNX Runtime Web (Audio to Text pages only), bundling @huggingface/jinja, @huggingface/tokenizers, protobufjs, flatbuffers, long, guid-typescript, platform | Apache-2.0 (model, Transformers.js, tokenizers, flatbuffers, long); MIT (ONNX Runtime, jinja, platform); BSD-3-Clause (protobufjs); ISC (guid-typescript) | https://github.com/openai/whisper, https://github.com/huggingface/transformers.js, https://github.com/microsoft/onnxruntime |
+| BiRefNet-lite background model (ZhengPeng7/BiRefNet; 512 px ONNX export by studioludens, weights stored as 16-bit) and U²-Net small (xuebinqin/U-2-Net; ONNX by BritishWerewolf), run with ONNX Runtime Web's WebGPU build (Remove Background page only) | MIT (BiRefNet, ONNX Runtime); Apache-2.0 (U²-Net) | https://github.com/ZhengPeng7/BiRefNet, https://github.com/xuebinqin/U-2-Net, https://github.com/microsoft/onnxruntime |
 | Inter typeface (@fontsource-variable/inter) | SIL Open Font License 1.1 | https://rsms.me/inter |
 | Archify (generator and runtime of `public/architecture.html`) | MIT | https://github.com/tt-a1i/archify |
 | JetBrains Mono typeface (embedded in `public/architecture.html`) | SIL Open Font License 1.1 | https://github.com/JetBrains/JetBrainsMono |

docs/ARCHITECTURE.md

@@ -77,6 +77,7 @@ sequenceDiagram
 | `src/engine/office.ts` | Word → print HTML, PDF → Word, PDF → PowerPoint | pdf.js, fflate |
 | `src/engine/ocr.ts` | English OCR for images and searchable PDFs | tesseract.js (self-hosted via `vite-plugins/ocr-assets.ts`), pdf-lib |
 | `src/engine/pdf-text.ts` | Rewrites changed lines inside the page content stream with the PDF's own font | pdf-lib |
+| `src/engine/background.ts`, `matte.ts` | Remove Background: BiRefNet-lite (512 px) in a worker on WebGPU or the CPU, U²-Net small for low-memory phones; the mask is fitted to the full-size photo with a fast guided filter and edge colours are cleaned (blur-fusion), so output keeps the photo's resolution | ONNX Runtime Web |
 | `src/engine/transcribe.ts` | Speech to text, SRT and VTT with Whisper in a worker; model cached once in Cache Storage | transformers.js, ONNX Runtime |
 | `src/tools/edit-pdf.ts` | In-page PDF editor: change text, add text, white-out | pdf.js, pdf-lib, `pdf-text.ts` |
 | `src/tools/ocr-viewer.ts` | Image with selectable recognized text (Live Text style) | – |

public/bg/birefnet-lite/manifest.json

@@ -0,0 +1,18 @@
+{
+  "source": "https://huggingface.co/studioludens/birefnet-lite-512/tree/4a3c40c36c94093cc1e724d9ea428b8fa4b57dc7",
+  "license": "MIT",
+  "files": [
+    {
+      "path": "model.onnx",
+      "size": 102785931,
+      "sha256": "9a62e3bee0d30df4686bc3d5e66011b8514ac4bf3e4aa39d0cce5175efb7636d",
+      "parts": [
+        "model.onnx.part0",
+        "model.onnx.part1",
+        "model.onnx.part2",
+        "model.onnx.part3",
+        "model.onnx.part4"
+      ]
+    }
+  ]
+}