Frequently Asked Questions About YomiNinja

Yes. YomiNinja is 100% free and open source under the GPL-3.0 license. There is no paid tier, no subscription, and no feature gating. Every feature described on this site is available to all users.

You can optionally support the developer on Patreon — this gives you early access to pre-release builds before they are publicly released. All features eventually come to the free version.

YomiNinja is primarily a dictionary lookup tool, not a translator. It extracts text from the screen using OCR, overlays it on your game, and lets you hover any word to get a Yomitan or 10ten popup with readings, definitions, pitch accent, and frequency data.

This makes it ideal for Japanese learners who want to understand what they're reading — rather than simply get a machine translation. Google Translate integration is available as an optional extra if you also want translation output.

Yes, all three platforms are supported:

• Windows 10 & 11 — full feature set, all OCR engines
• Linux — AppImage, .deb, .rpm, .pacman; X11 required (Wayland: experimental)
• macOS — Intel (x64) and Apple Silicon (arm64); Apple Vision OCR available

Download packages for all platforms are available on the download page.

YomiNinja uses screen capture and a transparent overlay — it does not inject code into game processes, read game memory, or interact with game files in any way. This is fundamentally different from text-hook tools like Textractor.

However, no tool can guarantee compatibility with every anti-cheat system. If you're concerned, use YomiNinja with single-player games or games without aggressive anti-cheat. The source code is fully public for inspection.

This is a macOS Gatekeeper quarantine issue, not actual damage. macOS applies a quarantine flag to apps downloaded from outside the App Store.

Fix: open Terminal and run:

xattr -cr /Applications/YomiNinja.app

Then try launching the app again. This removes the quarantine flag. The app is safe — source code is publicly available on GitHub.

Experimental Wayland support was added in v0.9.1, but it is not yet stable. For reliable operation on Linux, X11 is required.

If your Linux setup defaults to Wayland, you can launch YomiNinja under XWayland by setting DISPLAY=:0, or switch your session to X11 in your login manager. Full Wayland support is a known limitation being tracked on GitHub.

As of issue #100 opened on February 15, 2026, the Arch package can fail because of the deprecated http-parser dependency.

The practical workaround is to use the AppImage build instead of the .pacman package until the packaging issue is fixed upstream.

xdotool is a Linux tool that simulates keyboard and mouse input and interacts with windows via X11. YomiNinja uses it to identify and target specific application windows for capture.

Install it via your package manager:

sudo apt install xdotool    # Ubuntu/Debian
sudo dnf install xdotool    # Fedora
sudo pacman -S xdotool      # Arch

When YomiNinja's binary changes (after any update), macOS requires you to re-grant permissions. Go to:

System Settings → Privacy & Security → Accessibility — remove and re-add YomiNinja.

Do the same under Screen Recording. This is a macOS requirement, not a YomiNinja bug.

It depends on the content:

• PaddleOCR (default) — best for standard printed Japanese in most games. Fast, offline, no GPU required.
• MangaOCR — significantly better on stylized, hand-drawn, or manga-style fonts. Use this for visual novels and manga readers.
• Google Cloud Vision — highest overall accuracy, but requires an API key and internet connection.
• Apple Vision — macOS only; excellent for standard text including vertical Japanese.

See the OCR engines comparison for a full breakdown.

Vertical text is harder than standard horizontal dialogue. On macOS, Apple Vision is usually the first thing to try. On other platforms, test Google Cloud Vision if PaddleOCR and MangaOCR are not clean enough.

If the content is manga-like or uses decorative bubbles and rotated text, expect some OCR cleanup to still be necessary. Vertical layout is one of the hardest OCR cases in the YomiNinja ecosystem.

OCR accuracy depends heavily on the font. Games with unusual, pixel-art, or heavily stylized fonts cause the most problems. Common fixes:

• Switch to MangaOCR for stylized fonts
• Install a font mod for the game that replaces unusual fonts with standard ones (e.g., Stardew Valley requires a font mod for best results)
• Try Google Cloud Vision for difficult cases — it handles unusual fonts best
• Increase image preprocessing in YomiNinja settings (contrast, scale)

Furigana (small reading annotations above kanji) can be captured together with the main kanji, causing incorrect text like 「行い(おこな)い」 instead of 「行い」 — which breaks dictionary lookups.

Fix: in YomiNinja settings, adjust the furigana threshold slider. This was added in v0.9.1 specifically for this issue. A higher threshold filters out smaller text (furigana) more aggressively. You may need to experiment with the value for different games.

For core functionality — no. PaddleOCR, MangaOCR, and Apple Vision all run fully offline. Yomitan dictionary lookups also work offline once dictionaries are imported.

An internet connection is only needed if you choose to use Google Cloud Vision or Google Lens as your OCR engine, since these send captures to Google's servers.

No. Yomitan and 10ten Reader both ship pre-installed with YomiNinja. Open the app and the extensions are already there — no browser, no Chrome Web Store, no separate setup.

You will want to import dictionary files into Yomitan for it to show definitions. JMdict is the essential one. The dictionary setup guide walks through this in about 5 minutes.

Start with these essentials:

• JMdict (English) — the standard Japanese–English dictionary. Covers most vocabulary you'll encounter.
• KANJIDIC — kanji readings, meanings, and stroke counts.
• A frequency list (JPDB, iKnow, or VN-frequency) — shows how common a word is, useful for prioritising what to learn.
• Pitch accent dictionary (NHK or Kanjium) — shows the correct pitch pattern, essential if you care about natural-sounding Japanese.

All these are freely available from the jmdict-yomitan GitHub repository.

Common causes:

• No dictionaries imported — Yomitan needs at least JMdict to show definitions. Check Yomitan settings → Dictionaries.
• Overlay is in the wrong position — the hover only works on the YomiNinja overlay text, not the original game text underneath.
• Scanning key held down — by default, Yomitan requires holding Shift while hovering. You can disable this in Yomitan settings.
• Browser overlay mode off — if you're using the PWA/browser overlay, make sure it's active. The standard overlay and browser overlay use different extension contexts.

If OCR is clearly extracting text but the popup never appears, the problem is usually in the dictionary layer:

• JMdict or another dictionary was never imported
• Yomitan scanning is disabled
• A modifier key like Shift is still required for hover lookups
• You are hovering the original rendered game text instead of the overlay text

This pattern also appears in the public issue tracker, so it is a common setup problem rather than an unusual bug.

YomiNinja renders as a transparent window on top of your game using the Windows/macOS/Linux display compositor. Exclusive fullscreen bypasses the compositor entirely, making overlay windows impossible.

Fix: switch your game to Borderless Windowed mode. This is usually a toggle in the game's graphics settings. The game will look identical to fullscreen but the compositor (and thus the overlay) will work.

Yes. Because YomiNinja uses screen capture (not game memory injection), it works with any emulator that runs in a window: Ryujinx, yuzu, RPCS3, PCSX2, RetroArch, mGBA, and others.

Run the emulator in windowed or borderless mode. Select the emulator window in YomiNinja's capture source list, set up an OCR template over the game screen area, and you're set.

Yes, and this is one of YomiNinja's strongest use cases. It works particularly well for VNs where:

• Text is hardcoded as images (text-hook tools can't extract it)
• The VN uses a custom engine with no Textractor hook available
• You're playing a web-based or browser VN

For VNs where Textractor or LunaTranslator work, those tools may give better accuracy since they read text directly from memory. Use YomiNinja when hooking fails. See the comparison guide.

No. YomiNinja collects no personal data and sends no telemetry to any server. There is no analytics, no tracking, no user accounts.

The only exception: if you choose to use Google Cloud Vision or Google Lens as your OCR engine, your screen captures are sent to Google's servers for processing. This is opt-in and disabled by default. PaddleOCR and MangaOCR run fully locally.

The full privacy policy is available at yomininja.com/privacy-policy.

This is a false positive. YomiNinja is fully open source — the entire codebase is on GitHub. Some heuristic scanners flag Electron apps or apps that use screen capture as potentially suspicious.

You can verify the download by comparing file hashes against the checksums published in the GitHub release notes, or by building the app from source yourself. If your antivirus is blocking it, add an exception for the YomiNinja installation folder.

The current workflow uses WebSocket output:

1. YomiNinja sends extracted text over a WebSocket connection in real time
2. A texthooker page (e.g., this one) receives the text and displays it in a browser
3. Yomitan on that browser page lets you create Anki cards from the sentence

Direct Anki integration (without the texthooker intermediary) is on the YomiNinja roadmap. See the Anki mining guide for the full setup.

YomiNinja emits each recognized text result over a local WebSocket server. This allows any connected client to receive the text in real time.

Common uses: feeding a texthooker page for Anki mining, logging game dialogue, piping text to external TTS engines, or building custom integrations like GameSentenceMiner-style workflows. The WebSocket interface is part of YomiNinja's open architecture.

Check these in order:

1. Borderless windowed mode — exclusive fullscreen blocks all overlays. Switch to borderless.
2. Correct window selected — in YomiNinja's capture source list, make sure your game window is selected (not Desktop or another app).
3. OCR returned results — check YomiNinja's status bar. If OCR finds no text, the overlay has nothing to show. Try pressing the hotkey manually.
4. macOS permissions — Screen Recording access must be granted. Check System Settings → Privacy & Security → Screen Recording.

If you're using Google Translate integration, heavy usage triggers Google's rate limiting. This is a limitation of using unofficial access to Google Translate.

Solutions:

• Use DeepL integration instead (more reliable for high-volume use)
• Rely on Yomitan for lookups instead of full-sentence translation — this uses no external API
• If you specifically need Google Translate, use the official Google Cloud Translate API with an API key

Try these steps:

1. Try a different OCR engine — MangaOCR often handles games that PaddleOCR struggles with
2. Check image preprocessing settings — increasing contrast can help with low-contrast text
3. Use an OCR template — limit capture to just the dialogue box to reduce noise from the rest of the screen
4. Font mod — for games with unusual fonts, a font mod that replaces them with standard fonts dramatically improves accuracy
5. Scale up the window — small text is harder to OCR. Make the game window larger if possible

The overlay is designed to follow the game window. If it's behaving erratically:

• Make sure you've selected a specific window as the capture source, not full-screen desktop capture
• Check the overlay position settings in YomiNinja's preferences — you can lock the overlay to a fixed position relative to the window
• On macOS, display scaling issues were fixed in v0.9.3. Make sure you're on the latest version