Skip to content
MresyzzPublic

About

Ultra-fast, local-first lossless video clipper & silence stripper powered by Rust and FFmpeg. Zero re-encoding, instant export.

Resources

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

SnipFlow Logo

SnipFlow

Ultra-fast, local-first lossless video clipper, silence stripper & NLE pre-cut studio powered by Rust and FFmpeg.
Hybrid Smart-GOP rendering. Voice-bandpass silence stripping. Instant EDL & FCPXML timeline export.
简体中文说明文档

Latest Release Crates.io OpsScript Gate Verified License: GPL-3.0 Rust Version Platform Support GitHub Stars


SnipFlow Demo


Why SnipFlow?

Conventional Non-Linear Editors (Premiere Pro, CapCut, DaVinci Resolve) treat every rough trim as a full-scale production. They force heavy project generation, proxy rendering, and destructive re-compression cycles that take minutes or hours while degrading bitrate.

Conversely, traditional open-source stream tools (such as LosslessCut or HandBrake) remain bound to cumbersome Electron wrappers or manual segment picking, offering zero automated intelligence for stream archives, raw camera dumps, or gaming captures.

SnipFlow closes this architectural gap. It merges low-latency native system performance with intelligent, heuristic pre-cut automation—executing stream operations directly at the packet layer or exporting non-destructive .edl / .fcpxml timelines directly into DaVinci Resolve, Premiere Pro, and Final Cut Pro.

Metric / Capability SnipFlow LosslessCut Premiere / CapCut Web-Based Clippers
Cold Start Time < 0.2s (Native Rust) ~2.5s (Electron) 15s – 45s (Heavy NLE) Network dependent
Memory Footprint (Idle) < 45 MB ~280 MB ~1.5 GB+ ~400 MB+
Export Engine Tri-Mode (Stream Copy / Smart GOP / Exact) Stream Copy Re-encode (Lossy) Cloud Re-encode
Export Duration (1hr 4K) ~0.1s – 0.6s ~0.5s – 2s 8 – 25 Minutes Upload + Cloud Render
Automated Silence Stripping Voice-Bandpass + 60ms Breath Padding Manual Only Manual / Complex Plugin Paywalled / Cloud Queue
NLE Timeline Export CMX 3600 .edl & FCPXML v1.9 .fcpxml CSV / Basic Native Lock-in None
Peak Highlight Extraction Multimodal (Audio EBU R128 + Scene Flux) None None None
Headless Batch CLI Built-in (--batch + --nle) Limited Media Encoder Required None

Core Architecture & Engineering Highlights

1. Tri-Mode Cut Engine with Hybrid Smart-GOP Rendering

Pure stream copy (-c copy) is physically constrained to I-frame (keyframe) boundaries. SnipFlow provides three deterministic modes selectable in both GUI and CLI:

  • SMART GOP (CutMode::SmartCopy): Inspects packet PTS alignment. If the In-point falls mid-GOP, SnipFlow micro-renders only the partial leading GOP span [start -> first_keyframe] (typically 0.2s–1.5s) and stitches it seamlessly to a zero-reencode stream copy [first_keyframe -> end] covering 99%+ of the asset.
  • STREAM COPY (CutMode::StreamCopy): Pure zero-reencode demux/remux passthrough with real-time millisecond GOP drift indicators (SNAP (+120ms)) and magnetic keyframe snapping.
  • EXACT FRAME (CutMode::ExactFrame): Ultra-fast full-segment render fallback (-c:v libx264 -preset ultrafast -crf 18 -c:a copy) for heavily damaged variable-framerate streams.

2. Voice-Bandpass Silence Stripper, 60ms Breath Padding & Auto-Calibration

Standard RMS energy detectors struggle with continuous low-frequency HVAC/fan hum or abrupt consonant clipping.

  • Vocal Frequency Isolation: Chains an 85Hz – 7800Hz bandpass filter (highpass=f=85,lowpass=f=7800) prior to silencedetect and volumedetect analysis, rejecting room rumble and coil whine while leaving the exported audio stream 100% untouched.
  • Natural Breath Padding (60ms): Automatically retains 60ms of acoustic lead-in and trail-out around speech boundaries, preventing clipped plosives and robotic transitions.
  • One-Click Auto-Calibration: The AUTO CAL probe analyzes whole-track mean_volume and max_volume distributions to compute the optimal silence threshold automatically.

3. Non-Destructive NLE Timeline Export (CMX 3600 EDL & FCPXML v1.9)

Use SnipFlow as an ultra-fast pre-cut triage station before color grading or VFX:

  • Export active speech segments in < 5 milliseconds as industry-standard CMX 3600 .edl and Final Cut Pro .fcpxml (v1.9) timelines.
  • Import directly into DaVinci Resolve, Adobe Premiere Pro, or Final Cut Pro X with all silent pauses stripped and original camera media linked non-destructively.

4. Multimodal Highlight Discovery (Audio EBU R128 + Visual Scene Flux)

SnipFlow combines two orthogonal signal passes:

  • Band-Filtered EBU R128 Momentary Loudness: Isolates vocal/action transients (120Hz – 6000Hz) while suppressing mechanical keyboard click spikes.
  • Keyframe Scene-Change Detection (-skip_frame nokey): Evaluates visual scene flux (select='gt(scene,0.32)') directly across I-frames without decoding inter-frames, enabling accurate highlight extraction even on completely silent screen recordings.

5. Container Resilience & Multi-Track Stream Preservation

  • Timestamp Sanitization: Injects -fflags +genpts+igndts+discardcorrupt and -avoid_negative_ts make_zero to heal non-monotonic DTS/PTS in raw OBS dumps or surveillance captures.
  • Safe Multi-Stream Mapping: Maps -map 0:v:0? -map 0:a? -map 0:s? to preserve all multi-language / multi-channel audio tracks and subtitles while automatically stripping proprietary camera telemetry streams (gpmd, tmcd) that crash standard MP4 muxers.

Installation

Automated Install (Linux & macOS)

Install the standalone binary into ~/.local/bin using the automated unattended installer—verified across real Debian, Ubuntu, and Alpine containers via OpsScript Gate:

curl -fsSL https://raw.githubusercontent.com/Mresyzz/SnipFlow/main/install.sh | bash

Note: The installer automatically validates host architecture, downloads the matching pre-compiled binary, inspects local FFmpeg availability, and provides exact package installation commands if dependencies are missing.

Pre-built Binaries

Download pre-compiled standalone archives from the Releases page:

  • Windows: snipflow-v*-x86_64-pc-windows-msvc.zip (Zero-configuration portable bundle with static ffmpeg.exe and ffprobe.exe included)
  • macOS (Apple Silicon): snipflow-v*-aarch64-apple-darwin.tar.gz
  • macOS (Intel): snipflow-v*-x86_64-apple-darwin.tar.gz
  • Linux: snipflow-v*-x86_64-unknown-linux-gnu.tar.gz

Cargo (crates.io)

SnipFlow interfaces with FFmpeg via process-isolated pipelines—no C headers, libavcodec-dev, or PKG_CONFIG_PATH configuration required to compile from source:

cargo install snipflow

Headless CLI & Batch Automation

In addition to the desktop studio GUI, SnipFlow ships with a full headless CLI for server workflows and bulk recording triage:

# Launch Desktop Studio GUI
snipflow

# Single-file automated silence strip, smart-GOP seamless merge & NLE timeline generation
snipflow --cli raw_lecture.mp4 clean_lecture.mp4 --nle

# Batch process an entire directory of recordings with auto-calibrated noise floor & EDL/FCPXML output
snipflow --batch ./raw_recordings --out-dir ./cleaned_vods --min-silence 0.5 --mode smart --nle

Key Shortcuts

Shortcut Action Description
Space Play / Pause Toggle playback at active shuttle speed
J / K / L Shuttle Control Rewind 2s / Pause / Cycle playback speed (1x -> 1.5x -> 2x -> 4x)
S Live Skip-Silence Real-time audition mode that jumps over silent intervals during playback
X Blade Slice Split active clip at playhead position
P Snapshot Frame Capture and export full-resolution poster frame at playhead (.jpg)
← / → Step Frame Move playhead backward / forward by 1 frame
Shift + ← / → Seek I-Frame Jump directly to previous / next physical GOP keyframe
I / O Mark In / Out Set start and end boundaries (or drag [ and ] handles directly on timeline)
H Next Peak Jump playhead to the next detected audio/visual highlight
Enter Cut Export Execute export using active engine mode

Roadmap

  • High-performance keyframe-aligned demuxing and remuxing engine
  • Real-time RMS audio waveform envelope & interactive draggable In/Out trim handles
  • Live skip-silence audition mode (S) & 60ms natural breath padding
  • Multimodal audio EBU R128 + visual scene-flux highlight discovery
  • GOP micro-smart-rendering (CutMode::SmartCopy hybrid leading-GOP encode + stream copy)
  • Multi-track audio stream preservation and per-track silence inspection
  • Batch processing CLI interface (--batch) for headless server workflows
  • Vocal bandpass filtering (85Hz–7800Hz) and volumedetect noise-floor auto-calibration
  • Non-destructive NLE timeline export (CMX 3600 .edl & FCPXML v1.9 .fcpxml)

License

Licensed under the GNU General Public License v3.0 (GPL-3.0-or-later).

About

Ultra-fast, local-first lossless video clipper & silence stripper powered by Rust and FFmpeg. Zero re-encoding, instant export.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages