Skip to content

OpenCut v1.9.0

Choose a tag to compare

View all tags
@SysAdminDoc SysAdminDoc released this 26 Mar 05:11
v1.9.0
37afc7e

OpenCut v1.9.0 — AI Detection Upgrades, OTIO Export, Emotion Highlights

New Features

Silero VAD — Neural Silence Detection

  • AI-powered voice activity detection as an alternative to energy threshold
  • Three modes: energy (classic FFmpeg), vad (Silero neural), auto (tries VAD first, falls back to energy)
  • Far more accurate in noisy environments — uses a 1.8MB ONNX model, <1ms per 30ms chunk
  • Dropdown selector in the Silence Removal panel

CrisperWhisper — Verbatim Filler Detection

  • New filler detection backend using the nyrahealth/CrisperWhisper model
  • Transcribes speech verbatim, explicitly marking fillers as [UH] and [UM] with precise timestamps
  • Standard Whisper actively omits fillers — CrisperWhisper captures them all
  • Selectable via dropdown in the Filler Word Detection panel

Robust Video Matting (RVM)

  • New background removal backend with temporal consistency across frames
  • Uses a recurrent neural network that exploits temporal information — no flickering between frames
  • Two models: mobilenetv3 (fast) or resnet50 (higher quality)
  • Selectable alongside existing rembg backend in the Video AI panel

Florence-2 Auto Watermark Detection

  • New /video/auto-detect-watermark endpoint
  • Uses Florence-2 vision-language model to automatically locate watermark regions
  • Detects logos, text overlays, and watermarks with a natural language prompt
  • Edge-density fallback when Florence-2 is not installed

Emotion-Based Highlights

  • New emotion_highlights.py core module using deepface facial expression analysis
  • Builds an emotion curve over time, identifying peaks of emotional significance
  • New "Emotion Highlights (Face AI)" button in the Captions tab
  • Complements LLM-based highlight extraction with visual emotion signals

Engagement Scoring

  • Multi-signal highlight quality scoring: hook strength, emotional peak, pacing, quotability
  • Blended with LLM scores (60% LLM, 40% engagement) for better highlight ranking
  • Engagement breakdown returned in shorts pipeline and highlight extraction responses

OpenTimelineIO Export

  • New /timeline/export-otio endpoint with three modes: cuts, segments, markers
  • Universal timeline interchange format supported by DaVinci Resolve, Avid, FCP, and more
  • Export silence/filler cuts as an OTIO file importable into any compatible NLE

Improved Thumbnail Scoring

  • Composition balance analysis via edge density in quadrants
  • Center interest bonus for well-framed subjects
  • Motion blur penalty using gradient variance detection

Frontend & UXP Improvements

  • Detection method dropdown for silence removal (energy / VAD / auto)
  • Filler backend selector (Whisper / CrisperWhisper)
  • Background removal backend selector (rembg / RVM) with model list update
  • Project media scan debouncing prevents redundant API calls
  • Capability cache resets on disconnect so buttons re-evaluate on reconnect
  • Non-2xx HTTP responses properly surfaced as errors in API layer
  • UXP panel: AbortSignal.timeout() polyfill for older UXP runtimes
  • UXP panel: PProBridge.getProjectItems() for project media browsing
  • ExtendScript: import verification polling (replaces single-check), indentation cleanup

Backend

  • 7 new dependency checks: silero_vad, crisper_whisper, sam2, propainter, otio, deepface, rvm
  • Capabilities reported in /health response (silero_vad, crisper_whisper, otio)
  • New docs: RESEARCH.md, ROADMAP.md (moved to docs/), UXP_MIGRATION.md

Install

Download OpenCut-Setup-1.9.0.exe below — handles everything (server, FFmpeg, CEP extension, registry keys, desktop shortcut). No Python or FFmpeg required.

Or from source:

git clone https://github.com/SysAdminDoc/OpenCut.git
cd OpenCut
pip install -e ".[all]"
python -m opencut.server
Assets 3
Loading