TextLens

TextLens — Obsidian Plugin

An Obsidian plugin that scans the active note for image references and inserts the recognized text directly below each image, using a fully on-device OCR engine (PaddleOCR v6 via ONNX Runtime). No server, no API key, no internet connection required after the one-time runtime setup.

Features

• One command — TextLens: OCR Current Note processes every image in the active file
• Fully local — runs PaddleOCR v6 on-device via ONNX Runtime; your images never leave your machine
• Model tiers — Tiny (~5 MB, fastest), Small (~25 MB, balanced), Medium (~60 MB, most accurate)
• Dual syntax support — handles both Obsidian wikilinks (![[image.png]]) and standard Markdown (![alt](https://raw.githubusercontent.com/nexround/obsidian-text-lens/HEAD/image.png))
• Idempotent — skips images that already have an OCR block below them (toggle-able)
• Re-OCR — running the command again on an already-processed note automatically replaces all existing OCR blocks
• Two output formats — collapsible Obsidian callout or plain fenced code block
• Single undo step — the entire batch edit is one Cmd+Z entry
• Merge wrapped lines — automatically joins soft-wrapped OCR lines into natural prose paragraphs

Requirements

• Obsidian ≥ 1.7.2 (desktop only — Windows, macOS, Linux)
• One-time runtime setup (~40 MB download): Settings → TextLens → Local OCR Engine → Setup

Installation

Community Plugin (recommended)

1. Open Settings → Community Plugins and disable Restricted mode
2. Click Browse, search for TextLens, and install it
3. Enable the plugin, then go to Settings → TextLens → Local OCR Engine and click Setup

Manual

1. Download main.js, manifest.json, and styles.css from the latest release
2. Copy them into <your-vault>/.obsidian/plugins/text-lens/
3. Reload Obsidian, enable the plugin, then run Setup from the settings panel

From source

git clone https://github.com/Nexround/obsidian-text-lens.git
cd obsidian-text-lens
npm install
npm run build

Copy main.js, manifest.json, and styles.css into your vault's plugin directory.

Usage

Open any note that contains images, then open the command palette (Cmd/Ctrl+P) and choose:

TextLens: OCR Current Note

The plugin reads all image buffers in parallel, then sends them to the OCR model in a single batch call, and inserts the recognised text immediately below each image.

Callout output (default)

![[screenshot.png]]

> [!note]+ OCR: screenshot.png
> Hello World
> This is recognized text

Code block output

![[screenshot.png]]

​```ocr
Hello World
This is recognized text
​```

Settings

Local OCR Engine

Output

Diagnostics

Development

npm run dev      # watch mode
npm run build    # production build → main.js

Source layout

src/
  main.ts           Plugin entry, commands, settings UI, OCR dispatch
  local-ocr.ts      LocalOcrEngine — wraps ppu-paddle-ocr via CJS bundle
  native-manager.ts Runtime installer — downloads onnxruntime-node & @napi-rs/canvas
scripts/
  build-bundle.mjs  Bundles ppu-paddle-ocr + ppu-ocv + opencv-js → ppu-bundle.cjs
  deploy.mjs        Builds and copies all artifacts to a local vault for testing

Why ppu-bundle.cjs?

Obsidian's Electron renderer loads pages from the app:// protocol. Chromium treats a dynamic import("file://…") as cross-origin and blocks it. build-bundle.mjs pre-bundles ppu-paddle-ocr into a single CJS file so local-ocr.ts can load it with require(), which uses Node.js's CJS resolver and has no protocol restriction.

ppu-bundle.cjs is not committed to the repository. It is built in CI and attached to each GitHub Release as a release asset, then downloaded on-demand by the plugin's Setup flow.

License

MIT