PDF Composer

PDF Compose is an Obsidian plugin that lets you combine PDF pages and blank notebook pages into a single document, draw and annotate on them, search everything (including your handwriting), and export the result back to a standalone PDF — all stored as a normal Markdown note in your vault.

Table of contents

• Key features
• Getting started
• Working with pages
• Annotation tools
• Text blocks
• PDF annotations (highlight + comment)
• Handwriting recognition (OCR)
• Full-text search
• Dark mode & color filters
• Zoom, layout & touch/stylus support
• Exporting to PDF
• Undo / redo
• Settings reference
• Compatibility notes
• Known limitations
• License

Key features

• 📄 Compose documents from multiple PDFs and blank pages — mix pages from any number of source PDFs plus built-in blank/grid/lined templates, in any order.
• ✍️ Freehand drawing with pressure-sensitive pens (fineliner, fountain pen, pencil, charcoal, brush, two highlighters).
• 🔷 Shapes & connectors — lines, arrows, polygons, rectangles, triangles, diamonds, ellipses, with fill, styling, and text labels; arrows can be attached to a shape so they follow it when moved.
• 📝 Markdown text blocks placed freely on top of a page.
• 🖍️ PDF text highlighting + comment boxes, connected with a line, similar to margin notes on paper.
• 🔍 Offline handwriting recognition (OCR) for your pen strokes — no cloud services involved.
• 🔎 Fuzzy full-text search across PDF text, handwriting (OCR), text blocks, and PDF annotations, with adjustable typo tolerance.
• 🌓 Dark mode & color filters, with per-page override and automatic protection of embedded images from inversion.
• 🗂️ Full page management — reorder, multi-select, copy/cut/paste, rotate, insert, and swap a page's source.
• 📤 Export to a standalone PDF, vector-based, including your strokes, shapes, text blocks, and comments.
• ↩️ Undo/redo for essentially every action.
• 🔎 Smooth zooming & panning, pinch-to-zoom, and an optional "stylus only draws" mode so fingers can scroll while a pen draws.

A PDF Compose document is a regular Markdown note (marked with pdfcompose: true in its frontmatter). This means it stays searchable, versionable with git, and can always be opened as plain Markdown (pane menu → Open as Markdown) if you ever need the raw content.

Getting started

1. Create a new document — ribbon icon or command "Create new PDF Compose file".
2. Add a source PDF — Sources tab in the sidebar → Add PDF source → pick a name and the PDF file.
3. Add pages — Add page in the Pages tab (or a page's context menu → Insert page before/after) to pick pages from a source PDF, a blank template, or a reusable template file.
4. Draw / annotate — pick a tool from the annotation toolbar and start marking up the page.
5. Export — use the pane menu or command palette's Export as PDF whenever you want a standalone file to share.

Working with pages

• Sidebar page list — thumbnails, drag & drop reordering, multi-select (Shift/Ctrl-click), and a context menu (⋮) for delete, copy, rotate, insert, paste, change source, and show extracted page text.
• Add page dialog — choose a source PDF (with a page picker and thumbnails), a built-in blank template, or a file from your configured template folder; multiple pages can be selected and inserted at once, at a chosen position.
• Template folder — a vault folder (configured in settings) of PDF files offered as reusable "blank page" templates everywhere pages can be added.
• Rotate / invert per page — pages can be rotated in 90° steps and individually flagged to override the global color mode.
• Change source / page — swap which PDF page a Compose page points to without losing your annotations.
• Cross-page move — drag a selection of drawn objects or text blocks from one page onto another.

Annotation tools

Additional useful behaviors:

• Pressure sensitivity per pen, with a configurable minimum width and response curve.
• Curved / stepped connectors for lines, arrows, and polygons, editable per point via double-click (add point) or right-click / long-press (delete point).
• Endpoint binding — drag a line/arrow endpoint onto a shape to bind it; the connector then follows the shape if it's moved, resized, or rotated.
• Labels — double-click a shape, line, or arrow to add a text label.
• Transform handles — rotate and scale a selection; hold Shift for 15° rotation steps and uniform scaling.
• Copy / cut / paste / delete for the current selection, including across pages.

Text blocks

Freely positioned Markdown content rendered directly on top of a page (notes, links, images), with adjustable width and font size, edited through a small pop-up editor. An empty block is discarded automatically if you cancel without typing anything.

PDF annotations (highlight + comment)

Select text inside a PDF page (mouse drag, touch long-press, or pen drag) to get:

• an "Annotate" action that highlights the selected text, plus
• a comment box in a column to the right of the page, connected to the highlight by a line (style configurable: none / straight / curve / step).

Comment boxes support Markdown content, adjustable width and font size, and can be dragged vertically without moving the underlying highlight. A "Copy" action on the text-selection toolbar copies the selected PDF text to your clipboard.

Handwriting recognition (OCR)

PDF Compose can recognize your handwritten pen strokes into searchable text, fully offline — no cloud calls.

Setup

1. Open plugin settings → OCR (handwriting recognition) and enable it.
2. Download the model and character table via the built-in Download buttons (defaults point to a ready-to-use model), or supply your own compatible model and character table.
3. Fine-tune recognition with the Line factor, Word factor, Fragment rescue factor, Interpolation spacing, and Minimum confidence sliders — the defaults work well for typical handwriting; see the in-app tooltips for what each one does.
4. Choose whether OCR should run in the background (when idle) and/or automatically before a search.

You can also trigger OCR manually per page (page context menu → Run OCR), for the whole document (OCR for all pages / Re-run OCR completely), or via the command palette. Recognized words are searchable both through PDF Compose's own search and Obsidian's regular full-text search.

The Search tab has a "Show bounding boxes" toggle to visualize which strokes were recognized as which word — useful for tuning the line/word factors above.

Training your own handwriting model

The command "OCR: Collect training data (own handwriting)" opens a modal that shows you a series of words to trace (several built-in word lists, or your own custom list) and exports the raw samples as JSON, for training a model on your own handwriting style.

For a step-by-step guide on training your own OCR model, see: https://github.com/TMK314/TrainOcrModel-Guide

Model license

The pre-filled default model is licensed non-commercial (CC BY-NC-SA 4.0), with an additional "No Distribution" clause for its underlying training dataset. This applies only to the downloaded model file, not to the plugin itself — see Settings → OCR → Model license for the full summary and link. Commercial use requires training or supplying your own model.

Full-text search

The Search tab searches across:

1. PDF text,
2. Handwriting (OCR) results,
3. Text blocks, and
4. PDF annotations (comment content).

Search is typo-tolerant, with independently configurable tolerance for regular text vs. handwriting (handwriting recognition is more error-prone, so a higher tolerance is recommended there). Results are grouped by category with a preview, and can be stepped through with Previous / Next.

Dark mode & color filters

Configurable in settings, under Dark mode:

• Toggle whether color filters apply in light mode, dark mode, or both.
• Shift hue — rotate the display hue; 0° is a classic inversion.
• Monochrome tint — render dark mode in a single hue instead of full-color inversion.
• Dim white / Lighten black sliders for finer brightness control.
• Any page automatically gets readable dark-mode colors regardless of whether its source PDF is naturally light or dark, and embedded images are excluded from inversion so photos don't look like negatives.
• Any individual page's color handling can be manually overridden via the "Invert brightness" action.

Zoom, layout & touch/stylus support

• Zoom via Ctrl/Cmd + Scroll, Ctrl/Cmd + +/-/0, the on-screen zoom control, or pinch-to-zoom.
• Scroll direction — switch between vertical (default) and horizontal page layout.
• Stylus-only drawing — when enabled (default), finger touches only scroll/pan/pinch-zoom; only a stylus or mouse draws, erases, or selects.
• Long documents stay responsive thanks to on-demand page rendering.

Exporting to PDF

Export as PDF (pane menu, ribbon, or command palette) produces a standalone, vector-based PDF:

• Source PDF pages are copied unchanged, preserving text selectability and image quality.
• Your strokes, shapes, and labels are exported as vector graphics, not as a raster image.
• Text blocks keep basic Markdown formatting (bold, italic, inline code, headings, bullet lists).
• PDF annotations become real PDF comment annotations, plus the highlight.
• You choose which pages to include and which color mode (original / light / dark) to bake into the export.

Undo / redo

Nearly every action (drawing, erasing, moving, transforming, page operations, style and color-mode changes, etc.) can be undone with Ctrl/Cmd+Z / Ctrl/Cmd+Shift+Z (or Ctrl/Cmd+Y), the toolbar buttons, or the command palette.

Settings reference

Additional quick-access toggles (save PDF text as a searchable note, color mode, "stylus only", scroll direction, invert brightness) are available directly in the document view's "more options" (⋮) menu.

Compatibility notes

• A PDF Compose file is a regular Markdown note — Obsidian's search, backlinks, and graph view all work on it normally, and you can always fall back to Open as Markdown to read or edit it directly.
• The plugin recognizes and keeps working with files created by older versions of itself.

Known limitations

• Curved/stepped lines, arrows, and polygons are exported to PDF as straight segments.
• The bundled default OCR model is licensed non-commercial; commercial use requires your own model (see the training guide linked above).
• OCR accuracy depends on handwriting style and the model behind it.
• Very large or heavily annotated documents may run slower on low-end mobile devices.

License

This README describes the PDF Compose Obsidian plugin. The plugin's own license is independent of the license of any downloaded OCR model — see Settings → OCR → Model license for details on the bundled handwriting model.