CaptionBurn

Captions, on-device.

Auto-caption your videos and burn the subtitles in — entirely on your phone. No cloud, no uploads, no waiting.

Visit the website →  ·  Latest release →  ·  ☕ Buy me a coffee

---

Why CaptionBurn

• 🔒 Privacy-first — Your video never leaves the device. Speech recognition runs locally with Whisper. Translation runs locally with Google ML Kit. No accounts, no uploads, no telemetry.
• ⚡ Fast — On-device transcription with word-level timing. Pick a video, get captions in seconds.
• 🎨 You're in control — Edit every caption before you export. Pick the font size, position, outline, color, and word-highlight style. Translate to a second language if you want.

Features

🎤 99 languages transcribed	Whisper auto-detects the language of your video. Three model sizes (Tiny / Base / Small) — pick speed or accuracy.
🌐 50+ languages translated	Tap a language, get a translated caption track on the same video. ML Kit downloads the language pack once.
🪄 Burned-in captions	Captions are baked into the pixels — they show everywhere, on every platform, with no separate subtitle file.
✏️ Edit anything	Tap any caption to fix a word. Adjust word timing. Re-render in seconds.
🎨 Style controls	Size 36–110, 9 positions, outline 0–8 px, 5 text colors, 5 highlight colors, 4 word-effect styles (none / fill / grow / underline).
📂 Stays in your gallery	Exports to your device's Movies/CaptionBurn folder. Share, post, or save normally.

How it works

1. Import	2. Transcribe	3. Style	4. Export
Pick any video from your gallery	Whisper transcribes on-device with word-level timing	Choose font, color, position, word effects	Captions burned into the video, saved to your gallery

Privacy in plain English

• Your video stays on your device. It is never uploaded.
• Your transcript stays on your device. It is never uploaded.
• The Whisper speech model is downloaded once from public CDNs (HuggingFace), then runs offline forever.
• ML Kit downloads small translation packs (one per language) from Google's CDN. The pack is downloaded — your text is not uploaded.
• The app shows a banner ad and an occasional interstitial ad via Google AdMob. AdMob receives standard ad-platform info (advertising ID, IP, app interactions). That's it.

Read the full privacy policy →

Install

Until the Google Play Store listing is live, install directly from the GitHub Releases page.

1. Go to Releases.
2. Download the right APK for your phone:
   • app-arm64-v8a-release.apk — modern Android phones (most users)
   • app-armeabi-v7a-release.apk — older 32-bit ARM phones
   • app-x86_64-release.apk — emulators / Chromebooks
3. Open the APK on your phone and follow the install prompt. (You may need to allow installs from unknown sources for your browser.)

Support the work

CaptionBurn is free, MIT-licensed, and built by one person. If it helps you, buy me a coffee ☕ — it keeps the lights on.

---

For developers

Build locally

git clone https://github.com/chartmann1590/captionburn.git
cd captionburn
# Create local.properties with your AdMob test keys (see app/build.gradle.kts for the keys read).
./gradlew :app:assembleDebug

Java 17 + Android SDK 35 + NDK 28.2.13676358 required. The project pins these via the Gradle wrapper and ndkVersion.

CI/CD

Every push to main triggers .github/workflows/release.yml:

1. test — runs unit tests (./gradlew testDebugUnitTest)
2. build — assembles signed release APKs (per-ABI) + AAB, gates on 16 KB page alignment
3. release — publishes a GitHub Release with all artifacts and a mapping.txt for crash deobfuscation

versionCode is auto-incremented from github.run_number + 100; versionName becomes 0.1.<run_number>.

Required GitHub Secrets

Secret	Purpose
KEYSTORE_BASE64	Upload key (base64 of the .jks file)
KEYSTORE_PASSWORD	Keystore password
KEY_ALIAS	Key alias inside the keystore
KEY_PASSWORD	Key password
ADMOB_APP_ID	AdMob app ID
ADMOB_BANNER_AD_UNIT_ID	Banner ad unit ID
ADMOB_INTERSTITIAL_AD_UNIT_ID	Interstitial ad unit ID
ADMOB_NATIVE_ADVANCED_AD_UNIT_ID	Native advanced ad unit ID

16 KB page size

All native libraries (FFmpeg suite, FFmpegKit, whisper.cpp, ML Kit, libc++_shared) are aligned at 0x4000 (16 KB) per Android's 16 KB page size requirement, thanks to NDK r28+. The CI pipeline gates every release on zipalign -c -P 16.

---

License

App source: MIT. Native dependencies (FFmpegKit Full GPL, whisper.cpp MIT, ML Kit per Google ToS) are licensed under their own terms — see each project for the exact requirements when redistributing compiled binaries.

---

_{Built with care by Charles Hartmann · Buy me a coffee ☕}