Manga Converter — Matcha Reader Tools

📖 Manga Converter

AI panel detection for manga, in your browser. Finds the panels, optionally reads and translates their text with Gemini, and exports for the Matcha Reader, as EPUB, PDF, or XTC.

1. Choose your manga

Tap to choose a .cbz / .zip / .epub / .pdf — or select page images

Pages are ordered by filename, cover and copyright first. EPUBs use their spine order and table of contents. PDFs use document order, rasterized at 2×.

2. Export format

Matcha Reader format Panel-by-panel navigation, with offline dictionary lookup and translations. EPUB Fixed-layout .epub. Reads panel by panel on any e-reader, without dictionary, translations or panel zoom. PDF Same pages and panels as the EPUB, one per PDF page, for any PDF viewer. XTC Xteink's own format for the stock firmware, black-and-white. Pages are pre-rendered, so pick a target resolution below. XTCH Same, in 4-level grayscale. Twice the size.

3. Panel detection

Book type

Sets the order panels are read in.

Panels only (no full pages) Panel crops only. Much smaller, and opens straight into panel reading. The cover keeps its full page, as does any page whose panels miss part of the artwork.

Target resolution

Downscales to the device screen. Smaller download, faster page turns, no visible quality loss. Never upscales.

Advanced

AI panel detection Runs the YOLO26 model in your browser, ~21 MB once, then cached. Untick for the faster white-gutter heuristic. Rotate wide panels to portrait EPUB, PDF and XTC/XTCH only. Those carry pre-rendered images, so a wide panel is turned upright to fill the screen. Untick to keep panels as drawn. Matcha Reader format rotates them on the device instead. 1-bit BMP (Floyd–Steinberg dithering) Black-and-white pages that turn faster on the device. Best for line art. Screentone becomes dither patterns.

Dither brightness

Some scans dither darker than the original. Lighten them here. Affects 1-bit BMP and XTC/XTCH only; JPEG and PNG pages are untouched.

4. Text recognition (Gemini)

Skip text recognition Finds the panels but doesn't read the text. No API key, nothing leaves your browser, much faster. You lose dictionary lookup and translations on the device. Untick to have Gemini read each panel.

Advanced

Gemini model Panel crop margin (px) Only convert first N pages (for testing)

5. Book details (optional — auto-detected from EPUB/ComicInfo when present)

Title

Author

Language

Splits your reading stats by language, and tells text recognition which language to expect. Only EPUBs and CBZs with a ComicInfo.xml declare one; fill it in yourself for anything else. It is read at conversion time, so a book converted without it counts as "unknown" until you convert it again.

Chapters — one per line: page | title, 0-based final page number. Overrides the EPUB's own chapters.

OCR sends each panel image to the Gemini API. A full volume takes a while, so keep this tab open. The screen stays awake during conversion. Gemini's rate limits can make long volumes slow; the converter retries automatically.

Install on the device

The download is a zip holding one folder, named after the title. Put it anywhere on the SD card, by card reader or over the reader's Wi-Fi file transfer. /manga/YourTitle/ is a nice convention. The device finds any folder containing panels.idx, at any depth.