logo
logo

VC-6 is a multi-resolution codec that means no-more-proxies. This interactive demo allows you to explore the costs of retrieving content for AI training.

Doing no work is faster and greener than doing **some work. With VC-6, all the resolutions you need are stored in layers in the initial compression. Traditional solutions require proxies. Here, we measure

  • a VC-6 multi-resolution content-at-rest workflow
  • an FFMPEG on-demand proxy workflow

The demo uses a MetaRex worker node to intercept API calls and dynamically change the API so that you can transparently deploy VC-6 into existing workflows

Cost calculation is based on an April 2025 snapshot of AWS pricing.

Each time the workflow is run, the mrx-worker looks at the costs of sending the proxies to an AI endpoint:

  • FFMPEG:
    • compute cost of subsampling and creating proxies with FFMPEG
    • network costs of reading the source to do subsampling
    • increased storage cost of (proxy+source)
  • VC-6 costs:
    • difference between original FFMPEG compression & VC-6 compression (almost zero)
    • no networks additional costs - proxies already in original compression
  • common costs ignored
    • original compression cost (almost identical)
    • retrieval API costs - MetaRex makes them identical by design
    • AI processing costs - identical by design
    • latency differences between solutions ignored - hard to cost

The overall ROI workflow is based on

  • 1000 hr/day of new content ingested at full res (4k or HD)
  • VC-6 is 40% more expensive than H.264 for the original ingest
  • viewed content traverses an AWS NAT cost point
  • all content is stored for 1 month on S3
  • all content is software encoded on an AWS t3a instance
  • 1000 hr/day of new content analysed via Lo res proxy
  • 100 hr/day of new content re-analysed via Mi res proxy
  • 10 hr/day of new content re-re-analysed via Hi res proxy
  • the actual AI costs don’t vary because the AI API does not change
  • only the storage/retrieval of the ingested-proxy content is used for costs
  • For NAB, costs are a snapshot of AWS prices in March 2025
  • Real Time analytics pending a supporter to put them into MetaRex open-source instrumentation

If there are ways you’d like the ROI metrics to controlled with user parameters, please contact MetaRex


& view
frame: 2
  • 3840×2160 4k
  • 1920×1080 HD
  • 960×540 qHD Hi
  • 480×270 q2D Mi
  • 240×135 q3D Lo
  • 1920×1080 HD
  • 960×540 qHD Hi
  • 480×270 q2D Mi
  • 240×135 q3D Lo

Run the workflow to generate some stats