dominostars/playtranslate

Real-time translation and language learning Android app for games, visual novels, manga, and any other screen text. Supports 26 game languages and 59 user languages!

Download the latest release by clicking here

To report issues, receive support, or make requests, please join the Discord server

• One-tap Translation: Capture the game screen and translate the text on it with one tap
• Auto Translation Mode: Automatically translates as dialogue changes, no tapping required
• Word lookup: Hover the floating lens over any word for immediate dictionary definitions
• Offline: OCR and dictionary lookups work without an internet connection, with optional offline translation models
• Furigana/Pinyin Mode: Show reading hints above characters in real time
• Hotkeys: Configure a physical key to hold-to-preview translations or furigana, great for handhelds with dedicated buttons
• Dual Screen & Split Screen: Works across both screens on dual-display devices like the Ayn Thor, or in Android split-screen alongside windowed games
• Capture regions: Crop to just the dialogue box, subtitles, or any custom area
• Text-to-speech: Hear text spoken aloud. Change the default voice in settings
• Anki export: Save sentences to AnkiDroid with the original text, translation, word list, target words, text-to-speech, and a screenshot. Even record and include game audio! Card type selection with presets for popular decks.
• Yomitan integration: Yomitan dictionaries seamlessly integrate, including pitch accent, frequency chips, kanji enrichment, and merged term definitions everywhere (incl. Anki). Look for deeper integration in the future
• Camera translation: Point your camera at text in the world and read it live, or freeze a frame to tap words and look them up.
• Text History: Keep a record of captured sentences. Off by default.

1. Download the latest release by clicking here
2. On your Android, allow your browser or file manager to install unknown apps: open Settings → Apps → Special app access → Install unknown apps, pick the app, and turn on Allow from this source. Android also offers to take you there when you first open the APK
3. Open the APK and tap Install
4. On first launch, follow the onboarding steps to grant the necessary permissions

On some Android devices, Google Play Protect blocks sideloaded APKs and shows a vague "App not installed" or "harmful app" warning. If that happens, temporarily disable the scanner:

1. Open the Play Store
2. Tap your profile icon (top right)
3. Tap Play Protect
4. Tap the gear icon (top right)
5. Turn off Scan apps with Play Protect

Install the APK, then re-enable Play Protect afterward to keep scanning your other apps.

A few advanced features (like hotkey hold-to-preview) will prompt you to enable accessibility permissions. On Android 13 and later, Android greys out the accessibility switch for any app installed from a downloaded APK, and tapping it shows a "Restricted setting" message. To unblock it:

1. In Settings → Accessibility, select PlayTranslate and tap the greyed-out switch once. Close the "Restricted setting" message.
2. Open the App Info page: Settings → Apps → PlayTranslate.
3. Tap the ⋮ menu and choose Allow restricted settings. This item only appears after step 1.
4. Authenticate if prompted, then go back to Accessibility and turn the switch on.

To report issues, receive support, or make requests, please join the Discord server

You can support PlayTranslate on Ko-fi at https://ko-fi.com/playtranslate

PlayTranslate translates from 26 game languages (the text it can read off the screen) into 59 translation languages (the language shown to you). Both tables are sorted by total worldwide speakers.

By default, translation uses Lingva with ML Kit as an offline fallback. For higher quality translations, you can add any of the following under Settings → Translation services. Add as many as you like — each service is its own entry in the list, so you can keep several configured and reorder them to pick which one translates first:

• DeepL: free tier at deepl.com/en/pro#developer
• OpenAI: platform.openai.com — pick a model at runtime
• Gemini: aistudio.google.com — pick a model at runtime
• DeepSeek: platform.deepseek.com — pick a model at runtime
• Mistral: console.mistral.ai — pick a model at runtime
• Groq: console.groq.com — pick a model at runtime
• OpenRouter: openrouter.ai — pick a model at runtime
• Claude: platform.claude.com — pick a model at runtime
• Bilibili Index-Translate: index-translate.bilibili.com — free, no API key needed. May be temporary
• Custom: any other OpenAI-compatible endpoint — point it at your own base URL

Install AnkiDroid and grant PlayTranslate access in Settings to export cards directly to your decks.

• ML Kit: on-device OCR and translation
• Sudachi: Japanese morphological analysis (Apache 2.0)
• Yomitan: Japanese deinflection rules and their test suite, ported from the browser extension (GPL-3.0)
• HanLP: Chinese word segmentation
• KOMORAN: Korean morphological analysis
• Snowball stemmers via Apache Lucene: Latin/European stemming
• Lingva: online translation
• AnkiDroid: flashcard integration
• MNN: on-device LLM and OCR inference engine (Apache 2.0)
• PaddleOCR: on-device OCR from the bundled PP-OCRv6 text detector and unified recognizer, plus optional per-script recognizers (Apache 2.0)
• OpenCV: image processing for OCR (DBNet contour postprocessing, crop rectification) and for the camera tool's planar tracker (ORB features, pyramidal Lucas-Kanade flow, RANSAC homography fitting) (Apache 2.0)
• Silero VAD: voice-activity detection for the game-audio trimmer, bundled as a converted MNN model (MIT)
• OpenCC4j: Simplified/Traditional Chinese conversion (Apache 2.0)
• slimt: tiny Marian-based NMT engine that runs the Bergamot offline models (GPL-2.0-or-later, with MPL-2.0 Marian components)
• OkHttp: HTTP client for online translation and downloads (Apache 2.0)
• kotlinx.serialization: JSON (de)serialization for the translation backends and data models (Apache 2.0)
• Gson: streaming JSON parsing for Yomitan dictionary banks (Apache 2.0)
• CameraX: camera preview and analysis frames for the camera tool (Apache 2.0)
• Material Components for Android and Material Symbols: UI components, and icons traced from the Outlined symbol set (Apache 2.0)

Work we reimplemented rather than linked. No source was copied verbatim.

• offline-translator (David Ventura): its translator-rs planar tracking engine is the design reference for the camera tool's keyframe-OCR tracker. We took the split between optical flow that sustains correspondences and descriptor re-matching that corrects drift, the Idle/Locked/Lost lifecycle with inlier hysteresis, the anchor cache that re-locks a previously seen scene without re-running OCR, and several tuned thresholds. Independently reimplemented in Kotlin over OpenCV, without the reference's IMU prior and with a smoothing filter of our own (offline-translator GPL 3.0; translator-rs MIT)
• PyThaiNLP: Thai word segmentation, a faithful Kotlin port of its newmm maximal-matcher, run over a word list that includes its CC0 list (Apache 2.0)
• docTR and EasyOCR: line-grouping logic adapted for OCR word-box assembly, following docTR's recognize-then-group architecture with thresholds modelled on docTR _resolve_lines and EasyOCR group_text_box (Apache 2.0)
• Yomitan: the dictionary format our importer and styled renderer target — structured-content glossaries, per-dictionary CSS scoping, and media references. YomitanContentHtml is an independent Kotlin implementation of the format's tag and inline-style whitelists, following Yomitan's render-side sanitisation model (GPL 3.0)

• Gemma 4 E2B (Google): downloadable as an optional offline pack, MNN conversion by taobao-mnn (Apache 2.0)
• Hy-MT2 1.8B (Tencent): translation-specialised model, downloadable as an optional offline pack, MNN conversion by @starsharp06sharp (Apache 2.0)
• Qwen 3.5 2B (Alibaba): downloadable as an optional offline pack, MNN conversion by taobao-mnn (Apache 2.0)
• Retired packs, still recognised while installed: Qwen 2.5 1.5B Instruct (Apache 2.0) and Hunyuan-MT 1.5 1.8B (Tencent HY Community License; not available in the EU, UK, or South Korea)
• PaddleOCR PP-OCRv5+v6 recognizers: optional per-script OCR recognizer packs for additional scripts (e.g. Korean, Arabic, Cyrillic, Thai), downloadable per source language (Apache 2.0)
• Meiki: high-accuracy Japanese OCR model (D-FINE), downloadable as an optional offline pack (LGPL 3.0)
• MangaOCR: Japanese OCR refinement for stylized and vertical text, downloadable as an optional offline pack. manga-ocr-base-2025 by jzhang533, based on manga-ocr by kha-white (Maciej Budyś), both Apache 2.0, converted to fp16 MNN for on-device use. The manga-ocr model family is trained using the Manga109-s dataset, whose use is acknowledged per its terms
• Firefox Translations (Bergamot): Mozilla's offline NMT model pairs, downloadable for offline translation (CC BY-SA 4.0)

• JMdict, KANJIDIC2, and JMnedict: Japanese dictionary, kanji, and proper-name data (EDRDG licence; JMnedict offered as an optional in-app Yomitan download)
• CC-CEDICT: Chinese-English dictionary (CC BY-SA 4.0)
• CFDICT: Chinese-French dictionary, used for French-target glosses (CC BY-SA 3.0)
• HanDeDict: Chinese-German dictionary, used for German-target glosses (CC BY-SA 2.0 DE)
• Wiktionary via kaikki.org: multilingual dictionary entries (CC BY-SA)
• Tatoeba: example sentences (CC BY 2.0)
• PanLex: multilingual translation pairs (CC0)
• wordfreq: word frequency data
• Camel Morph MSA: Arabic morphology, used to map inflected and broken-plural surface forms to dictionary lemmas (© CAMeL Lab, NYU Abu Dhabi; CC BY 4.0, modified)
• Arramooz: Arabic morphological dictionary (© Taha Zerrouki; GPL 3.0)
• Morfologik / PoliMorf: Polish morphology, used to map inflected surface forms to dictionary lemmas (© Marcin Miłkowski; BSD-2-Clause)
• SudachiDict: Japanese tokenizer dictionary bundled for Sudachi, including UniDic (© NINJAL) and part of mecab-ipadic-NEologd (Apache 2.0)
• Jiten: Japanese frequency data, offered as an optional in-app Yomitan dictionary download (CC BY-SA 4.0)
• Wikimedia Commons: pronunciation audio for word playback and Anki cards, fetched on demand. Each clip carries its own author and license (typically CC BY / CC BY-SA / public domain), shown as a credit that travels onto exported cards

GPL 3.0