RawLens: Autonomous Local-First 4K Video Curator & Rough-Cut Engine on UGOS
Transform raw camera dumps into editable NLE timelines while you sleep—with zero cloud uploads, zero subscriptions, and zero data egress.
Inspiration
Every video creator, YouTuber, and indie filmmaker experiences the same frustrating post-shoot bottleneck:
A single afternoon shoot generates anywhere from 300 GB to well over 1 TB of raw 4K footage. Uploading these massive files to cloud-based AI tools like Descript or Runway takes hours over standard home internet, costs significant monthly fees, and exposes confidential unreleased client footage to external servers.
Meanwhile, editors spend more than half of their post-production time doing repetitive, mechanical scrubbing: cutting out camera operator pauses, blurry takes, accidental black frames, awkward silences, and verbal false starts.
Creators buy high-end UGREEN AI NAS systems specifically for their blazing 10GbE local network speeds and massive storage capacity. We built RawLens to make that hardware work actively for creators overnight, bridging on-premise storage with professional editing software.
What It Does
RawLens is an autonomous, local-first video curation Skill engineered specifically for UGREEN UGOS. Operating continuously in the background of your NAS, it eliminates the tedious early stages of video editing:
- Automatic Folder Monitoring: Watches incoming camera ingestion directories on the NAS without requiring manual intervention.
- Hardware-Accelerated Defect Pruning: Scans video and audio streams locally to detect and flag silences, out-of-focus keyframes, and ruined takes.
- Dialogue Take Alignment: Transcribes spoken audio on-device to isolate valid sentences and discard false starts or verbal stumbles.
- Non-Destructive Rough-Cut Assembly: Instead of re-rendering massive video files (which degrades visual quality and wastes hours), RawLens outputs an industry-standard CMX 3600 Edit Decision List (EDL).
- Instant NLE Import: Editors simply open DaVinci Resolve or Adobe Premiere Pro, import the lightweight EDL file, and start cutting on a timeline where the garbage takes are already gone.
How We Built It: The 4-Stage Pipeline
We designed RawLens as an ultra-efficient, multi-stage pipeline running directly on UGOS Linux environment:
- Ingestion & Stream Analysis: When media is copied from an SD card over 10GbE to the NAS, RawLens inspects the container metadata (frame rate, timecodes, resolution) without decoding pixel data.
- Signal Filtering via Native FFmpeg: We utilize low-overhead FFmpeg filter flags including silencedetect, blurdetect, and scene change detection. Because this process analyzes compressed bitstreams and sampled keyframes rather than rendering full 4K frames, CPU load stays below 20% and memory footprint remains under 2 GB.
- Local Dialogue Alignment via whisper.cpp: The system extracts a lightweight 16kHz mono audio track and passes it through an on-premise, 4-bit quantized Whisper engine. It identifies speech timestamps and determines which takes contain complete, spoken thoughts.
- EDL Sequence Generation: A specialized Python module translates approved timecodes into a clean CMX 3600 EDL sequence with exact frame matching (supporting 23.976, 24, 25, 29.97, and 59.94 FPS).
Why UGREEN AI NAS Is The Key
RawLens could not work effectively on a generic cloud server or a standard desktop PC:
- Local 10GbE Throughput: Ingesting 500 GB takes minutes over a local NAS connection, compared to 10+ hours over typical cloud upload links.
- True Zero-Egress Privacy: Sensitive, unreleased commercial footage never leaves the local network.
- Zero Ongoing Costs: There are no monthly subscriptions, API token meters, or cloud storage rental fees.
- Always-On Background Utility: The NAS runs 24/7, turning dead overnight hours into productive curation cycles so the project is ready the moment the editor sits down at their desk.
Challenges We Ran Into
- Preventing NAS Resource Exhaustion: Running video analytics on a multi-purpose NAS cannot interfere with primary network storage or other container services. We solved this by strictly avoiding full-frame video rasterization and relying on keyframe stream inspection.
- Variable Frame Rate (VFR) Drift: Footage shot on smartphones or consumer cameras often uses VFR, causing audio-video drift when compiled into fixed-frame EDLs. We engineered a normalization layer that aligns timestamps to the nearest standard broadcast timeline base.
- Handling Multi-Camera False Starts: Differentiating between an intentionally quiet pause in acting versus a dead silence required calibrating dynamic noise floors rather than using static decibel cutoffs.
Accomplishments That We're Proud Of
- Built a truly non-destructive workflow that outputs native timeline cuts without altering or re-compressing a single original camera pixel.
- Achieved full compatibility with industry-standard NLEs (tested with DaVinci Resolve 19 and Adobe Premiere Pro).
- Packaged the entire architecture to conform strictly with the Uliya Skill Development Standard for UGOS.
What's Next for RawLens
For Round 2 testing on the physical UGREEN iDX6011 Pro NAS:
- Intel QuickSync Video (QSV) Acceleration: Activating hardware-level decoding on the NAS's integrated GPU to quadruple inspection speed.
- Native UGOS Event Watcher: Hooking directly into UGOS File Manager file-drop notifications for true zero-click automation.
- Multi-Camera Sync: Expanding the EDL compiler to support multi-angle multicam clip groupings automatically.
Open Source & Disclosures
- FFmpeg (v6.0+): LGPL v2.1+ / GPL v2+ (https://ffmpeg.org)
- whisper.cpp: MIT License by Georgi Gerganov (https://github.com/ggerganov/whisper.cpp)
- Python Standard Library: PSF License
- AI Tool Disclosure: Gemini and Claude were used to assist in technical documentation structuring, code review, and English copy proofreading.
Log in or sign up for Devpost to join the conversation.