Inspiration

Independent videographers and digital creators generate hundreds of gigabytes of raw 4K/8K footage on every shoot. Storing this on a NAS is natural, but organizing, logging, and finding key moments requires hours of tedious manual scrubbing. Existing cloud AI tools (Descript, Frame.io) require multi-hour uploads of massive video files, cost expensive recurring monthly subscriptions, and risk leaking client confidential assets on external servers. We were inspired to build a native on-premise AI agent that leverages the UGREEN AI NAS iDX6011 Pro's 96 TOPS NPU and dual 10GbE network to bring instant, zero-cost, completely private post-production automation directly to the creator's desk.

What it does

UGREEN Creator Studio is an autonomous on-premise AI Skill running on UGOS Pro that:

  1. Automated Multi-Cam Ingest: Automatically detects camera media inserted via the front SD 4.0 (UHS-II) card slot or Thunderbolt 4 (40Gbps) ports.
  2. Local NPU AI Acceleration: Utilizes the Intel Core Ultra 7 255H NPU (96 TOPS) via OpenVINO to run quantized Whisper-int8 for speech-to-text timecoded transcription and MobileNet-CLIP for visual scene tagging and clustering.
  3. Smart Bin Organization: Automatically categorizes assets into /Interviews, /B-Roll, and /Highlights with standardized XMP metadata sidecars on BTRFS storage.
  4. NLE Rough-Cut Export: Automatically generates ready-to-edit Final Cut Pro FCPXML and DaVinci Resolve EDL timeline sequences.
  5. Hardware Status Integration: Displays real-time ingestion speed and indexing completion percentage directly on the NAS's front 3.7-inch Touch LCD display and sends mobile push notifications via the UGOS app.

How we built it

  • Core Runtime: Native Python daemon integrated with UGOS Pro CLI (ugos-fs, ugos-ai-service, ugos-lcd, ugos-notify).
  • AI Inference Engine: OpenVINO runtime targeting the 96 TOPS Intel NPU and Arc iGPU, running int8 quantized Whisper and CLIP models for high-throughput, low-power inference.
  • Hardware Decoding: Hardware-accelerated VAAPI/FFmpeg video decoding directly from NVMe/SATA storage pools.
  • Data Sovereignty: 100% offline, air-gapped execution with zero telemetry or outbound network calls.

Challenges we ran into

Balancing multi-stream 4K H.265 hardware decoding with concurrent NPU batch inference while preserving system thermal limits and maintaining real-time status updates on the front-panel 3.7-inch LCD screen.

What we learned

The UGREEN AI NAS iDX6011 Pro's unified architecture (Intel Core Ultra + 96 TOPS NPU + 64GB LPDDR5x RAM) provides desktop-workstation-class AI compute in an energy-efficient 24/7 on-premise server footprint.

What's next for UGREEN Creator Studio

In Round 2 on physical hardware, we will benchmark multi-drive SATA RAID 5 / NVMe read throughput under full NPU load, integrate interactive Uliya voice search ('find sunset drone shots'), and polish the one-click NLE import workflow.

Built With

Share this project:

Updates

Submission history