tojiru

Turn a fixed-page document into a self-contained static web reader. Point it at a PDF, comic archive, or DjVu; get a folder you can drop on any static host or open locally. No server, no client-side PDF engine — pages are pre-rendered once and lazy-loaded.

tojiru (綴じる, "to bind a book") is the descendant of a PHP PDF reader written in 2009. The idea that kept it alive for 17 years: pre-render each page, ship it static, let the browser do nothing but display. This rewrite generalizes it to more formats and packages it as a CLI.

What it does

tojiru book.pdf --out site/
# → site/  (index.html, reader.js, manifest.json, pages/, thumbs/)
# serve it, or open site/index.html

• PDF pages become crisp vector SVG (via poppler/mupdf). Text stays sharp at any zoom, and for born-digital PDFs the bundle is often smaller than the source.
• Comics (CBZ, CB7, CBR) and DjVu become image pages with thumbnails.
• The reader lazy-loads pages as you scroll (IntersectionObserver), has a thumbnail sidebar, keyboard navigation, reading-position memory (localStorage), and #page=N deep links.
• Full-text search (Ctrl+F) for PDFs with a text layer, a table of contents built from the PDF outline, a dark theme that follows the system, a phone-friendly layout, and an optional double-page / right-to-left (manga) view.
• SVG pages are gzipped and inflated in the browser (DecompressionStream), so they render correctly on any host regardless of its Content-Encoding configuration.

Supported formats

| Format | How | Extra dependency | |---|---|---| | PDF | one SVG per page (vector) | pdftocairo (poppler) or mutool (mupdf) | | CBZ | zip of images | none (bundled) | | CB7 | 7-Zip archive of images | 7z (p7zip) | | CBR | RAR archive of images | none (bundled, WASM) | | DjVu | rendered page by page | ddjvu, djvused (djvulibre) |

Comics work out of the box. Only PDF and DjVu need a system tool, detected at runtime — if it's missing, tojiru tells you which package to install instead of crashing.

Two PDF extras are picked up automatically when present: pdftotext (poppler, usually installed with pdftocairo) builds the search index, and mutool (mupdf) reads the outline for the table of contents. Both are optional — without them you just lose that one feature.

Install

Requires Node ≥ 20.

npm install -g @gheop/tojiru

Or run it once without installing:

npx @gheop/tojiru book.pdf --out reader/

From source:

git clone <repo-url> tojiru
cd tojiru
npm install
npm run build
npm link        # makes the `tojiru` command available

Optional system tools (install only what you need):

# Debian/Ubuntu
sudo apt install poppler-utils djvulibre-bin p7zip-full mupdf-tools
# Fedora
sudo dnf install poppler-utils djvulibre p7zip mupdf

Usage

tojiru <input> [options]
tojiru serve <dir> [options]

Generation shows per-page progress on stderr (e.g. Converting 12/30).

Convert options

| Option | Description | |---|---| | -o, --out <dir> | Output folder (default: the input name without extension) | | -t, --title <title> | Document title shown in the reader | | -f, --force | Overwrite a non-empty output folder | | --serve | Start a preview server on the output folder after converting | | --single-file [file] | Output a single portable HTML file instead of a folder | | --image-format <fmt> | Raster page encoding: keep (as-is, default) or webp | | --quality <n> | WebP quality 1-100 for lossy raster pages (default: 80) | | --spread | Lay pages out two-up (double-page spread) | | --rtl | Right-to-left reading order (manga); pairs with --spread | | --paged | Default to the paged view (one page per screen); the reader can switch back |

Reading

The generated reader works the same whatever the source format:

• Search — Ctrl+F (or /) opens a text search for PDFs that shipped a text layer. Matches list the page with a highlighted excerpt; click one to jump there.
• Table of contents — when a PDF has an outline, it shows at the top of the thumbnail column; click an entry to jump to its page, and the current section is highlighted as you read.
• Dark mode — follows the system by default; the ◐ button toggles it and remembers your choice.
• Paged view — the ≣ / ▭ button switches between continuous scroll and one page per screen (horizontal, arrow-key or swipe navigation). --paged makes the paged view the default; the choice is remembered.
• Double-page / manga — --spread shows two pages per row; add --rtl for right-to-left order (page 1 on the right, left arrow advances).

tojiru manga.cbz --out manga/ --image-format webp --spread --rtl

Single file

tojiru book.pdf --single-file book.html

Produces one .html file with all pages, thumbnails, and the reader bundled in. Double-click it to read offline — no server, no extra files. Works on any OS that has a browser.

Capped at 30 MB of page data. For large documents, use the folder output and host it.

Smaller comics

tojiru comic.cbz --out comic/ --image-format webp

Comics keep their original images by default (keep), so the bundle matches the source. Pass --image-format webp to re-encode raster pages as WebP — useful for colour comics shipped as large PNGs. On the Pepper&Carrot sample this drops the bundle from 1.2 MB to ~300 KB with no visible loss. Pages already in WebP are left untouched.

--quality <n> sets the WebP quality (default 80). Lower it for maximum compression (--quality 75 is ~20% lighter again and still clean on most artwork); raise it for archival (--quality 92). It applies to lossy raster pages — comics re-encoded with --image-format webp and auto-rasterized scanned/comic PDFs. DjVu scans stay lossless.

Preview server

tojiru serve out/         # serve an existing bundle
tojiru serve out/ -p 9000 # on a custom port

| Option | Description | |---|---| | -p, --port <n> | Port to listen on (default: 8000; auto-increments on EADDRINUSE) |

The server uses Node built-ins only — no extra dependencies. It opens the browser automatically (best-effort).

Examples:

tojiru novel.pdf --out reader/ --title "My Novel"
tojiru comic.cbz --out comic/
tojiru scan.djvu --out book/ --force

# preview immediately after converting
tojiru novel.pdf --out reader/ --serve

# or convert first, serve later
tojiru novel.pdf --out reader/
tojiru serve reader/

Where it shines (and where it doesn't)

tojiru is built for vector text. Born-digital text PDFs render as compact, razor-sharp SVG. Scanned PDFs and image-heavy pages are rasterized, so the bundle grows — for those, the format is the wrong fit. Measured numbers are in examples/BENCH.md, with a reproducible sample corpus in examples/.

How it works

input → detect format (magic bytes) → extractor → pages → manifest.json → static bundle + reader

Each format is a small plugin (src/extractors/) producing a normalized list of pages; everything downstream is format-agnostic. Adding a format means adding one extractor.

Development

npm test            # unit + integration (vitest)
npm run test:e2e    # reader rendering (Playwright)
npm run dev -- <input> --out <dir>   # run the CLI from source via tsx

License

MIT — see LICENSE.

Changelog

v0.9.0 — Paged view, contents highlight, leaner comics (2026-06-26)

• Paged reading view — the ≣ / ▭ button switches the reader between continuous scroll and one page per screen (horizontal, arrow-key or swipe navigation), and the choice is remembered. --paged makes it the build default; --rtl pages it right-to-left.
• The table of contents follows along — the outline entry covering the current page is highlighted as you read.
• --image-format webp never inflates a comic — each page keeps WebP or its original, whichever is smaller. An already-compressed scan that used to grow ~2× now stays at its source size, while colour comics still drop to ~×0.3 (measured points in examples/BENCH.md).
• Tests run in CI on Node 24 (the Node 20 deprecation warning is gone), and a flaky reader e2e was de-flaked.

v0.8.0 — Search, dark mode, contents and manga layout (2026-06-26)

• Full-text search (Ctrl+F or /) for PDFs that carry a text layer. pdftotext builds an index at conversion time (a search.json file, or inlined in single-file mode); matches list the page with a highlighted excerpt and jump there on click. Scans and comics keep the browser's native find.
• Dark theme that follows the system by default, with a ◐ button that toggles and remembers the choice. The reader now has a phone layout (the thumbnail column collapses to an overlay, full-width pages, larger tap targets), and the toolbar controls and thumbnails are real buttons reachable by keyboard.
• Table of contents built from the PDF outline via mutool, shown at the top of the thumbnail column; click an entry to jump to its page. Falls back to nothing when mutool or the outline is absent.
• Double-page spread (--spread) and right-to-left manga order (--rtl): two pages per row, page 1 on the right, the left arrow advances. Pairs wrap to one page per row on narrow screens.
• Tests now run in CI on GitHub and GitLab, and --serve opens a browser on Windows too.
• Internals: the single-file output is derived from index.html instead of duplicating the markup, and buildManifest takes an options object.

v0.7.0 — Lighter vector text (2026-06-26)

• SVG page coordinates are now rounded to 1 decimal (0.1 pt) instead of 2 — measured ~19% lighter (gzipped) on glyph-heavy text PDFs, verified visually lossless even when zoomed (0.1 pt is ~0.13 px on screen).

v0.6.0 — WebP quality control (2026-06-26)

• --quality <n> (default 80) sets the lossy WebP quality for comic and auto-rasterized PDF pages. The default dropped from 82 to 80 — measured ~8% smaller with no visible loss.
• Thumbnails are now encoded at WebP effort 6 (slightly smaller, free).
• DjVu stays lossless: benchmarking near-lossless against lossless on bitonal scans showed an identical size (no smooth content to quantize), so there was nothing to gain.

v0.5.0 — WebP comics (2026-06-26)

• --image-format webp re-encodes raster (comic) pages to WebP — the Pepper&Carrot sample drops from 1.2 MB to 324 KB. Default stays keep (images copied as-is). Pages already in WebP are left untouched.
• Re-converting into an existing folder now clears the old pages/ and thumbs/ first, so a format switch never leaves orphaned page files behind.
• The repo's comic demo is now WebP, cutting the committed site/demo from 1.4 MB to ~550 KB.

v0.4.0 — Single-file output (2026-06-26)

• --single-file [file] option: bundles all pages, thumbnails, and the reader into one portable HTML file
• Double-click the file to read offline — no server, no folder needed
• Size guard: rejects documents whose page data exceeds 30 MB and tells you to use folder output instead
• Folder mode and all existing tests unchanged

v0.3.0 — Smaller PDF bundles (2026-06-26)

• Vector PDF page coordinates are rounded to 2 decimals — ~25% smaller pages, no visible change
• Image-only PDF pages (scans, comic PDFs) are auto-detected and rendered as WebP instead of SVG-wrapped bitmaps — e.g. a comic PDF dropped from ×34 to ×2.7

v0.2.0 — Preview server and progress (2026-06-25)

• Built-in tojiru serve <dir> command previews a bundle locally — uses Node built-ins only, no extra install needed
• --serve flag on the convert command starts the server immediately after generating
• Per-page progress shown on stderr during generation (Converting i/N, Processing i/N)
• Browser opens automatically after serve / --serve (best-effort, swallows errors)

v0.1.2 — Version reporting (2026-06-25)

• tojiru --version now reads the version from package.json, so it always matches the published version

v0.1.1 — Scoped package name (2026-06-25)

• Published on npm as @gheop/tojiru (the unscoped tojiru name was taken); the installed command is still tojiru
• README install instructions corrected to the scoped name

v0.1.0 — Initial release (2026-06-25)

• PDF → vector SVG pages; CBZ/CB7/CBR/DjVu → image pages
• Static folder bundle with a vanilla-JS lazy-loading reader (thumbnails, keyboard nav, resume, deep links)
• Format detection by content; graceful degradation when an optional system tool is missing