NovaCut

A professional Android video editor built with Kotlin and Jetpack Compose. Open alternative to CapCut, PowerDirector, and DaVinci Resolve -- with AI-assisted tools, GPU-accelerated effects, and desktop NLE interoperability.

NovaCut PREMIUM MOBILE VIDEO EDITING

Release history is maintained in git tags and the development checkout's local CHANGELOG.md.

Project planning

Planning files are local-only in the development checkout:

• ROADMAP.md is the only source of truth for incomplete actionable work.
• RESEARCH.md stores consolidated product, platform, and ecosystem research.
• Shipped work lives in git history and the local CHANGELOG.md.

Features

Timeline Editing

• Multi-track timeline with video, audio, overlay, text, and adjustment layers
• Trim, split, merge, crop, rotate with visual handles
• Slip/slide editing — drag clip body to slide (reposition) or slip (shift source window)
• Magnetic snapping — clips snap to edges, playhead, and markers (8dp threshold with diamond indicators)
• Clip grouping — select multiple clips, group/ungroup, move as a unit
• Speed control (0.1x-16x) with bezier speed ramping curves and presets
• Keyframe animation for position, scale, rotation, opacity, volume with 12 easing types (linear, ease in/out, spring, bounce, elastic, back, circular, expo, sine, cubic)
• 14 speed presets including time freeze, film reel, heartbeat, crescendo
• Undo/redo (50 levels) with full state restoration + command-based undo foundation
• Long-press multi-select for batch operations
• Pinch-to-zoom + zoom in/out/fit buttons
• Timeline scrubbing with frame-accurate seeking
• Colored timeline markers — 6 colors (red/orange/yellow/green/blue/purple) with labels, notes, and jump navigation
• Sticker/GIF/image overlays — position, scale, rotate, opacity with timeline placement
• Favorites & recent effects — mark effects as favorites, track recently used for quick access
• Multi-cam sync — audio-based clip synchronization across tracks
• Clip reorder & move — reorder clips within a track or move between tracks
• Haptic feedback — tactile response on trim handle grab and magnetic snap
• Waveform caching — LRU cache avoids redundant audio decoding on timeline recomposition
• Clip color labels — 7 Catppuccin colors (red, peach, green, blue, mauve, yellow, none) with colored top border on Timeline
• Track collapse/expand — Per-track chevron + collapse/expand all toggle, collapsed tracks show thin 24dp colored bars
• Track height cycling — Long-press track type icon to cycle 48→64→80→96dp
• Keyboard shortcuts — Space, Ctrl+Z/Y, arrow keys, M, S, +/-, Delete, Ctrl+S, Ctrl+C/V for external keyboard editing
• Snap-to-beat/marker — Beat markers and timeline markers as additional snap targets (settings-driven)
• Marker list panel — Searchable, filterable marker list with color chips, inline label editing, jump-to-time

Effects & Transitions

• 37 GPU-accelerated GLSL transitions with unique Material icons per type — dissolve, wipe, zoom, spin, flip, cube, ripple, pixelate, morph, glitch, swirl, heart, dreamy, plus 12 new: door open, burn, radial wipe, mosaic reveal, bounce, lens flare, page curl, cross warp, angular, kaleidoscope, squares wire, color phase
• 40+ video effects — brightness, contrast, saturation, hue, sharpen, vignette, mosaic, fisheye, wave, chromatic aberration, radial blur, motion blur, tilt shift
• Film grain — perceptual-aware (more in shadows, less in highlights), animated blue noise pattern
• VHS/Retro — scanlines, chroma bleeding, tracking distortion, posterized color depth
• Glitch — RGB channel splitting, 8x8 block corruption, horizontal line displacement
• Light leak — procedural animated warm gradient with screen blend mode
• 9-tap Gaussian blur — separable kernel with proper sigma-based weights
• 18 blend modes (normal, multiply, screen, overlay, soft light, hard light, difference, exclusion, etc.)
• Freehand/rectangle/ellipse/gradient masks with feather, expansion, and motion tracking
• Professional chroma key — YCbCr color space keying with smoothstep feathering and green/blue spill suppression

Color Grading

• Lift/gamma/gain color wheels with continuous control
• RGB curves and HSL qualifier
• LUT import (.cube/.3dl) with file picker and intensity control
• Color matching — per-channel gamma correction between reference and target clips
• Video scopes — histogram, waveform, vectorscope with animated overlay (GPU compute shader ready for ES 3.1+)

Audio

• Full audio mixer with per-track volume faders, pan slider, mute/solo, smoothed VU meters (ballistic attack/decay)
• 15 DSP effects — parametric EQ, compressor (corrected attack/release), limiter, delay, chorus, de-esser, pitch shift, noise gate
• Waveform visualization with fade envelope overlay
• Beat detection — spectral flux onset detection with adaptive thresholding and BPM estimation (aubio NDK ready)
• Auto-duck — speech-aware volume keyframing (analyzes voice track, creates keyframes on music track)
• EBU R128 loudness normalization — K-weighted measurement with 6 platform presets:
  • YouTube/Spotify (-14 LUFS), TikTok (-14 LUFS), Podcast/Apple (-16 LUFS), Broadcast EBU R128 (-23 LUFS), Cinema (-24 LUFS), Loud (-9 LUFS)
• True-peak limiting to prevent clipping
• Voiceover recording with automatic timeline placement
• Fade overlap protection — fade in + fade out constrained to clip duration
• Noise reduction — Spectral gate heuristic (5 modes: off/light/moderate/aggressive/spectral gate). DeepFilterNet ML path planned

AI Tools

Tool	Engine	On-Device?
Auto Captions	ONNX Runtime Whisper tiny.en (English; multilingual Sherpa/Whisper path gated)	Yes
Background Removal	MediaPipe Selfie Segmentation (~1-7MB, ~30fps)	Yes
AI Green Screen	Planned -- RobustVideoMatting (requires model integration)	Planned
Object Removal	LaMa-Dilated inpainting (40ms/frame @ 512x512 on flagship devices)	Yes
Video Upscaling	Planned -- Real-ESRGAN (requires model integration)	Planned
Frame Interpolation	Planned -- RIFE v4.6 (requires NCNN dependency)	Planned
Style Transfer	Planned -- AnimeGANv2 + Fast NST (requires model integration)	Planned
Stabilization	Planned -- OpenCV (requires dependency)	Planned
Smart Reframe	EMA-smoothed crop trajectory, 3 strategies (face/pose detection is stub)	Partial
Tap-to-Segment	Planned -- SAM 2.1 Hiera Tiny target with MobileSAM fallback	Planned
Scene Detection	Content-aware frame difference analysis with auto-split	Yes
Auto Color	Histogram-based brightness/contrast/saturation/temperature	Yes
Motion Tracking	Template matching with position keyframe generation	Yes
Audio Denoise	Spectral gate heuristic (DeepFilterNet ML planned)	Yes

Text & Titles

• Rich text overlays with 10+ animation styles
• Static templates — lower thirds, title cards, end screens, CTAs
• Animated Lottie templates — 10 built-in (slide-in lower third, bounce title, typewriter, glitch reveal, neon glow, fade subtitle, circle logo reveal, countdown, subscribe button). Render frame-by-frame for export via LottieDrawable
• Caption editor with start/end time sliders (mutually constrained)
• Caption style gallery with karaoke, word-pop, bounce, typewriter, minimal styles
• Continuous caption positioning via BiasAlignment (not 3-zone snap)
• Text on path (straight, curved, circular, wave)
• Shadow, glow, letter spacing, line height controls

Text-to-Speech

• System TTS — Android built-in voices with mutex-protected synthesis
• Piper TTS (planned) — near-human quality VITS voices via Sherpa-ONNX (stub, requires dependency integration)
  • 10 voice profiles defined: Amy (US), Ryan (US), Alba (UK), Thorsten (DE), Dave (ES), Siwis (FR), Takumi (JP), Huayan (CN), Sunhi (KR), Faber (BR)
  • Currently falls back to Android System TTS
• System/Piper engine toggle in TTS panel

Export

• GIF export — Self-contained GIF89a encoder with LZW compression, configurable frame rate (10/15/20fps) and max width (320/480/640px)
• Frame capture — PNG/JPEG single-frame export from current playhead position
• 480p to 4K Ultra HD
• 4 codecs — H.264, H.265 (HEVC), AV1, VP9 with hardware capability detection via MediaCodecList
• HDR export confidence — HEVC, AV1, and VP9 preflight reports HDR10+, Dolby Vision Profile 10, Ultra HDR source gain maps, and device-tier hardware encode support before render
• One-tap platform presets — YouTube 1080p, YouTube 4K, TikTok, Instagram Reels, Instagram Square, Threads
• Multi-sequence Media3 Composition export for visible video and overlay tracks, with dedicated audio-track mixdown
• Batch export with multiple presets simultaneously
• Background export with progress notification, ETA display, and cancel
• Timeline interchange — OTIO (OpenTimelineIO) JSON export/import + FCPXML export for desktop NLE round-tripping (DaVinci Resolve, Premiere Pro, Final Cut Pro)
• EDL export (CMX 3600) with sanitized reel names and proper timecodes
• Chapter markers and subtitle export (SRT, VTT with word-level cues, ASS/SSA with full styling)
• Burned-in subtitle rendering — Canvas-based with ASS/SSA file generation for FFmpeg integration
• Audio-only and stems export modes
• Export error cleanup — partial output files deleted on failure/timeout

Effect Library

• Copy/paste effects between clips
• Export effects to .ncfx file for sharing
• Import effects from .ncfx with portable LUT references (filename-based, not absolute paths)

Project Management

• User template system (save/load/delete project templates, preserves non-media track clips)
• Project snapshots with version history and auto-generated default names
• Project archive (ZIP export)
• Auto-save with configurable interval, format versioning, rotating backups
  • Full serialization: all clip fields, compound clips, 9 caption style properties, mask bezier handles, clip group IDs
• Command-based undo/redo foundation — sealed class with AddClip, RemoveClip, TrimClip, MoveClip, SetClipSpeed, ApplyEffect, CompoundCommand
• 3-tier proxy workflow — thumbnail (scrubbing) / proxy (540p editing) / original (export) with auto-switch and storage management
• Cloud backup UI (backend pending)
• First-run tutorial — auto-shows on first launch, dismissable, resettable from Settings

Settings

• Default resolution, frame rate, aspect ratio, export codec
• Auto-save toggle + interval (15-300s)
• Proxy resolution selector
• Reset first-run tutorial
• Show waveforms — Global waveform visibility toggle
• Snap to beat / snap to markers — Timeline snap behavior toggles
• Default track height — 48/64/80/96dp chips
• Confirm before delete — Gate clip deletion dialog
• Thumbnail cache size — 64/128/256 MB
• Default export quality — LOW/MEDIUM/HIGH
• All settings persist via DataStore

Tech Stack

Component	Technology
Language	Kotlin 2.1.0
UI	Jetpack Compose + Material 3 (Catppuccin Mocha theme)
Video	Media3 1.10.1 (Transformer + ExoPlayer)
Effects	OpenGL ES 3.0 (37 GLSL transitions, 40+ effect shaders)
Audio DSP	Custom engine (EQ, compressor, chorus, delay, pitch shift)
Speech-to-Text	ONNX Runtime 1.26.0 (Whisper)
Noise Reduction	Spectral gate fallback (DeepFilterNet planned)
Beat Detection	Spectral flux onset detection (aubio NDK ready)
Loudness	EBU R128 / ITU-R BS.1770 measurement
Segmentation	MediaPipe Tasks Vision 0.10.35
Video Matting	Planned (RobustVideoMatting, ONNX Runtime)
Object Removal	LaMa-Dilated (ONNX Runtime, neighbor-fill fallback)
Upscaling	Planned (Real-ESRGAN)
Frame Interpolation	Planned (NCNN + Vulkan)
Style Transfer	Planned (AnimeGANv2 + Fast NST)
Stabilization	Planned (OpenCV)
TTS	Android System TTS (Piper via Sherpa-ONNX planned)
ASR acceleration target	Sherpa-ONNX v1.13.2 AAR + Moonshine v2 Tiny EN policy (native backend still gated)
Animated Titles	Lottie Compose 6.7.1 and Media3 Lottie overlay support
Startup performance	AndroidX Baseline Profile / Macrobenchmark 1.4.1
Timeline Exchange	Planned (OpenTimelineIO)
DI	Hilt / Dagger
Database	Room (v4 with migration chain 1→4)
Settings	DataStore Preferences
Architecture	MVVM, single-activity Compose navigation, StateFlow

Architecture

com.novacut.editor/
├── ai/                     # AI features (captions, scene detect, stabilize, auto-edit)
├── engine/                 # Core engines (29 injectable singletons)
│   ├── VideoEngine          # Media3 playback + export
│   ├── AudioEngine          # Waveform extraction + PCM processing
│   ├── AudioEffectsEngine   # DSP chain (EQ, compressor, chorus, etc.)
│   ├── ShaderEffect         # GLSL fragment shader pipeline
│   ├── KeyframeEngine       # Bezier/hold interpolation
│   ├── ProjectAutoSave      # JSON serialization with format versioning
│   ├── ExportService        # Foreground service for background export
│   ├── BeatDetectionEngine  # Spectral flux onset + BPM estimation
│   ├── LoudnessEngine       # EBU R128 measurement + normalization
│   ├── NoiseReductionEngine # Spectral gate (DeepFilterNet stub)
│   ├── FrameInterpolationEngine  # RIFE v4.6 slow-motion (stub)
│   ├── InpaintingEngine     # LaMa object removal (ONNX Runtime + NNAPI)
│   ├── UpscaleEngine        # Real-ESRGAN video upscaling (stub)
│   ├── VideoMattingEngine   # RVM AI green screen (stub)
│   ├── StabilizationEngine  # OpenCV optical flow (stub)
│   ├── StyleTransferEngine  # AnimeGAN + Fast NST (stub)
│   ├── SmartReframeEngine   # Subject-tracking auto-crop
│   ├── TapSegmentEngine     # SAM 2.1 / MobileSAM target metadata (stub)
│   ├── PiperTtsEngine       # Piper VITS TTS (stub, system TTS fallback)
│   ├── LottieTemplateEngine # Animated title rendering
│   ├── FFmpegEngine         # FFmpegX fallback encoder (stub)
│   ├── SubtitleRenderEngine # Canvas + ASS subtitle rendering
│   ├── GenerativeVideoPolicy # Cloud-only trust gates for large video generators
│   ├── TimelineExchangeEngine  # OTIO/FCPXML interchange
│   ├── ProxyWorkflowEngine  # 3-tier media management
│   ├── EditCommand          # Command-pattern undo/redo
│   ├── db/ProjectDatabase   # Room database with migrations
│   ├── whisper/WhisperEngine     # Built-in Whisper (ONNX)
│   ├── whisper/SherpaAsrEngine   # Sherpa-ONNX ASR target metadata + fallback
│   └── segmentation/        # MediaPipe selfie segmentation
├── model/                  # Data classes (Project, Clip, Track, Effect, etc.)
├── ui/
│   ├── editor/             # Main editor (EditorScreen, EditorViewModel, 40+ panels)
│   ├── export/             # ExportSheet, BatchExportPanel
│   ├── mediapicker/        # MediaPickerSheet
│   ├── projects/           # ProjectListScreen, ProjectTemplateSheet
│   ├── settings/           # SettingsScreen, SettingsViewModel
│   └── theme/              # Catppuccin Mocha theme
├── MainActivity.kt         # Single activity, Compose navigation, permission handling
└── NovaCutApp.kt           # Application class, notification channels

Build

# Debug build
./gradlew assembleDebug

# Release build (requires keystore.properties or env vars)
./gradlew assembleRelease

# Managed-device startup/editor performance gate
./gradlew :baselineprofile:pixel6Api36BenchmarkReleaseAndroidTest :baselineprofile:collectNonMinifiedReleaseBaselineProfile

Manual QA: audio focus

Before release, verify audio focus on a physical device:

• Start music in another app, open NovaCut, and play timeline preview. The external app should pause or duck while NovaCut plays.
• Connect headphones, start preview playback, then unplug them. NovaCut preview should pause instead of continuing through the speaker.
• Start timeline preview, then start a voiceover recording. Preview should pause before recording starts and focus should release when recording stops.
• Start a TTS preview, then leave the panel or close the editor. Preview speech should stop and other audio should be able to resume.

Requirements

• Android Studio Ladybug+ (2024.2+)
• AGP 8.7.3, Gradle 8.9, JDK 21
• Android SDK 36

Release Signing

Configure via keystore.properties:

storeFile=path/to/your.jks
storePassword=yourpass
keyAlias=youralias
keyPassword=yourpass

Or via environment variables: NOVACUT_STORE_FILE, NOVACUT_STORE_PASSWORD, NOVACUT_KEY_ALIAS, NOVACUT_KEY_PASSWORD

If release credentials are not configured, assembleRelease falls back to debug signing so CI and local verification can still produce a testable release artifact without relying on an embedded keystore.

Release Checksums

CI publishes a .sha256 file next to every uploaded APK. To reproduce the release checksum locally after a build:

python scripts\write_release_checksums.py --check
Get-FileHash app\build\outputs\apk\release\app-release.apk -Algorithm SHA256

APK Size Budget

CI checks debug, release, and androidTest APK sizes against scripts/apk_size_baseline.json with a 2 MB per-output growth allowance. After an intentional dependency or asset-size change, refresh the baseline from a verified build:

python scripts\check_apk_size.py --update-baseline
python scripts\check_apk_size.py

Distribution Readiness

GitHub Releases are the direct APK distribution channel for this checkout. Google Play listing metadata, privacy disclosures, Data safety worksheet, and screenshot assets are committed under fastlane/metadata/android/en-US/ and validated in CI.

F-Droid-compatible Fastlane metadata is present in the same source tree. F-Droid publication still needs a final reproducible-build metadata pass, including AllowedAPKSigningKeys after the release signing key policy is fixed.

Android developer verification is not complete. Starting in September 2026, Google requires apps installed on certified Android devices in initial regions to be registered by a verified developer, and package names must be registered with a signed APK. NovaCut can keep shipping direct APKs locally, but broad sideload/F-Droid continuity depends on completing that account/package-name step or documenting a limited-distribution fallback.

Dependencies

Key external dependencies currently in build.gradle.kts:

Dependency	Version	Purpose
ONNX Runtime	1.26.0	Whisper ASR + LaMa inpainting
Sherpa-ONNX	1.13.2 target	Future native Moonshine v2 ASR path; official AAR is a GitHub release asset, not a Maven dependency
SAM 2.1 ONNX	Targeted	Future tracked-mask path via explicit model download; MobileSAM remains the small-device fallback
MediaPipe	0.10.35	Selfie segmentation
Lottie Compose	6.7.1	Animated title templates
AndroidX Benchmark/ProfileInstaller	1.4.1 / 1.4.1	Baseline Profile generation and release profile install
OkHttp	5.3.2	Future opt-in cloud APIs
FFmpegKit 16 KB	6.1.1	FFmpeg-backed fallback export paths not covered by Media3 Transformer
Android DeepFilterNet	0.0.8	On-device voiceover noise reduction

Distribution and Third-party Notices

Open-source notices are available in Settings > Third-party notices > Open source licenses. Redistributed builds that include com.moizhassan.ffmpeg:ffmpeg-kit-16kb:6.1.1 must keep the packaged FFmpegKit / FFmpeg license resources and source-offer text. The fork source is published at https://github.com/moizhassankh/ffmpeg-kit-android-16KB, and the bundled source-offer resource points to FFmpegKit source access at https://github.com/arthenica/ffmpeg-kit/wiki/Source.

Supported Devices

• Min SDK: 26 (Android 8.0 Oreo)
• Target SDK: 36 (Android 16)
• Required: OpenGL ES 3.0
• Recommended: 4GB+ RAM, Snapdragon 7-series or better for AI features
• AV1 hardware encoding: Pixel 8+, Snapdragon 8 Gen 3+, Dimensity 9200+

Permissions

Permission	Purpose
RECORD_AUDIO	Voiceover recording
FOREGROUND_SERVICE	Background export processing
FOREGROUND_SERVICE_MEDIA_PROCESSING	Android 14+ foreground export classification
POST_NOTIFICATIONS	Export progress notifications
INTERNET	Model downloads (Whisper), cloud inpainting API
ACCESS_NETWORK_STATE	Respect Wi-Fi-only model download settings
VIBRATE	Haptic feedback

Media access uses the system Photo Picker (ActivityResultContracts.PickVisualMedia) and ACTION_OPEN_DOCUMENT exclusively — NovaCut requests no broad READ_MEDIA_VIDEO / READ_MEDIA_IMAGES / READ_MEDIA_AUDIO / READ_EXTERNAL_STORAGE / WRITE_EXTERNAL_STORAGE permissions, so the per-URI grant model survives background kill without the Android 14 Selected Photos compatibility-mode loss.

Known Limitations

• Multi-sequence export now honors track opacity through Media3 compositor settings, and all 18 fallback blend modes render distinctly; true source-over-destination blend math still needs a custom programmable compositor because Media3's public settings only expose alpha/transform
• clip.isReversed works in preview but not in export (Media3 Transformer has no reverse playback support)
• SmartRenderEngine analysis results not used for actual export bypass
• 11 AI/ML engine stubs awaiting dependency integration (see ROADMAP.md)

License

MIT