A real-time game translation Android app, built for both language learners and people who just want to play. Supports 23 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

PlayTranslate with Persona 3 Reload

Features

• One-tap Translation: Capture the game screen and translate Japanese text 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 definitionss
• 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.

How to Use

1. Download the latest release by clicking here
2. On your Android, enable Settings → Security → Install unknown apps for your file manager or browser
3. Open the APK and tap Install
4. On first launch, follow the onboarding steps to grant the necessary permissions

Won't install?

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.

Can't enable accessibility?

A few advanced features (like hotkey hold-to-preview) will prompt you to enable accessibility permissions. Some Android OEMs block sideloaded apps from receiving accessibility permissions by default, and the toggle in Settings might be grayed out or show a "Restricted setting" message. To unblock it:

1. Open Settings → Apps → PlayTranslate
2. Tap the ⋮ menu (top right)
3. Tap Allow restricted settings
4. Authenticate when prompted

You can now turn on accessibility for PlayTranslate.

Support

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

Supported Languages

PlayTranslate translates from 25 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.

Game languages (read from the screen)

Language	Native name	Code
English	English	en
Chinese (Simplified)	简体中文	zh
Chinese (Traditional)	繁體中文	zh-Hant
Hindi	हिन्दी	hi
Spanish	Español	es
Arabic	العربية	ar
French	Français	fr
Portuguese	Português	pt
Russian	Русский	ru
Indonesian	Bahasa Indonesia	id
German	Deutsch	de
Japanese	日本語	ja
Turkish	Türkçe	tr
Vietnamese	Tiếng Việt	vi
Korean	한국어	ko
Italian	Italiano	it
Thai	ไทย	th
Dutch	Nederlands	nl
Romanian	Română	ro
Hungarian	Magyar	hu
Swedish	Svenska	sv
Catalan	Català	ca
Danish	Dansk	da
Finnish	Suomi	fi
Norwegian	Norsk	no

Translation languages (translated for you)

Language	Native name	Code
English	English	en
Chinese	中文	zh
Hindi	हिन्दी	hi
Spanish	Español	es
Arabic	العربية	ar
French	Français	fr
Bengali	বাংলা	bn
Portuguese	Português	pt
Russian	Русский	ru
Urdu	اردو	ur
Indonesian	Bahasa Indonesia	id
Swahili	Kiswahili	sw
German	Deutsch	de
Japanese	日本語	ja
Marathi	मराठी	mr
Telugu	తెలుగు	te
Turkish	Türkçe	tr
Vietnamese	Tiếng Việt	vi
Korean	한국어	ko
Tamil	தமிழ்	ta
Persian	فارسی	fa
Italian	Italiano	it
Thai	ไทย	th
Gujarati	ગુજરાતી	gu
Polish	Polski	pl
Ukrainian	Українська	uk
Tagalog	Tagalog	tl
Malay	Bahasa Melayu	ms
Kannada	ಕನ್ನಡ	kn
Dutch	Nederlands	nl
Romanian	Română	ro
Hungarian	Magyar	hu
Greek	Ελληνικά	el
Czech	Čeština	cs
Swedish	Svenska	sv
Belarusian	Беларуская	be
Hebrew	עברית	he
Bulgarian	Български	bg
Catalan	Català	ca
Slovak	Slovenčina	sk
Haitian Creole	Kreyòl Ayisyen	ht
Croatian	Hrvatski	hr
Danish	Dansk	da
Finnish	Suomi	fi
Norwegian	Norsk	no
Albanian	Shqip	sq
Galician	Galego	gl
Slovenian	Slovenščina	sl
Lithuanian	Lietuvių	lt
Latvian	Latviešu	lv
Afrikaans	Afrikaans	af
Macedonian	Македонски	mk
Estonian	Eesti	et
Georgian	ქართული	ka
Welsh	Cymraeg	cy
Maltese	Malti	mt
Icelandic	Íslenska	is
Irish	Gaeilge	ga
Esperanto	Esperanto	eo

Optional: Online Translation Backends

By default, translation uses Lingva with ML Kit as an offline fallback. For higher quality translations, you can plug in an API key for 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
• Custom: any other OpenAI-compatible endpoint — point it at your own base URL

Optional: Anki Flashcards

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

Credits

Libraries and services

• ML Kit: on-device OCR and translation
• Sudachi: Japanese morphological analysis (Apache 2.0)
• 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)
• 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)

Adapted from other projects

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)

(Optional) Downloadable Offline Models

• TranslateGemma 4B (Google): translation-tuned Gemma 3, downloadable as an optional offline pack (Gemma terms of use)
• Qwen 2.5 1.5B Instruct (Alibaba): downloadable as an optional offline pack (Apache 2.0)
• Gemma 4 E2B (Google): downloadable as an optional offline pack (Gemma terms of use)
• Hunyuan-MT 1.5 1.8B (Tencent): translation-specialised model, downloadable as an optional offline pack (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)

Linguistic data

• 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)
• 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

License

GPL 3.0