Netflix VMAF

Perceptual video quality scoring with Netflix VMAF, run over SSH on your own VM.

Open sourceSLA: oss-best-effortv3.2.0Image
VMAF predicted vs subjective quality correlation
VMAF predicted vs subjective quality correlation

Netflix VMAF (Video Multimethod Assessment Fusion) is the open-source, full-reference perceptual video quality metric used across the streaming industry to measure how good an encode looks to the human eye. Score a distorted or transcoded clip against its pristine reference and get a VMAF score that correlates with real viewer perception, so you can tune encoders, compare codecs, and set quality gates with confidence.

This is a headless command-line appliance, operated over SSH, with no web console. It ships the vmaf CLI, the libvmaf library, and a lean FFmpeg 7.0.2 built with the libvmaf filter, all compiled from upstream source, plus a tiny bundled reference and distorted sample so you can run a demo score the moment the instance boots.

Key features

  • The vmaf CLI plus libvmaf, with the standard VMAF models bundled (vmaf_v0.6.1, the no-enhancement-gain vmaf_v0.6.1neg, and the 4K model)
  • An FFmpeg build with the libvmaf filter, so you can score directly from any input FFmpeg can decode (H.264, HEVC, VP9, AV1, and more)
  • A bundled reference and distorted sample clip for an instant demo score
  • A first-boot self-check (vmaf version plus a real sample score) that fails closed on a broken install
  • A co-located landing page that documents the exact commands to run

Use cases

  • Measure and compare encode quality across codecs, bitrates, and presets
  • Add an automated VMAF quality gate to a transcode or packaging pipeline
  • Validate a new encoder or CDN transform against a golden reference

Why run it on the NetActuate edge Scoring is CPU-bound and works on the raw video you already move through your pipeline. Running VMAF on a NetActuate VM close to where your media is produced or stored keeps large mezzanine files off the public internet and gives you dedicated compute for batch scoring on infrastructure you control.

Getting started Deploy the appliance, then SSH in and run vmaf against the bundled sample clips (or your own reference and distorted files). The landing page served on the instance IP shows the installed version, a demo score, and copy-paste commands for both the vmaf CLI and the ffmpeg libvmaf filter.

Frequently asked questions

How do I deploy Netflix VMAF on NetActuate?

Use the Deploy on NetActuate button to talk to our team. We provision Netflix VMAF on a NetActuate virtual machine in the point of presence you choose, close to your audience. Each deployment runs in the single location you select.

What does it cost to run Netflix VMAF?

Netflix VMAF is open-source software and free to run on the marketplace - you pay only for the NetActuate compute (virtual-machine size and location) it runs on, at standard rates. There are no usage-based software fees.

Where can Netflix VMAF run?

Netflix VMAF can be deployed in any of NetActuate's 45+ points of presence across every continent, so you can place ingest, transcode, delivery and caching close to your contributors and viewers.

Who supports Netflix VMAF?

NetActuate operates and supports the underlying infrastructure - the virtual machine, network and edge platform. The application itself is supported by its upstream open-source project.