Getting Started

VMAF Studio is a web-based interface for running VMAF (Video Multi-Method Assessment Fusion) video quality analysis on OSC. It connects to the EasyVMAF service to compare a reference video against a transcoded version and compute a perceptual quality score.

VMAF is Netflix's open source quality metric, widely used to validate encoded renditions before publishing to end-users. A score of 100 indicates the distorted video is perceptually identical to the source.

Prerequisites

  • An OSC account (sign up here)
  • Your OSC personal access token (from Settings → API) — VMAF Studio uses it to call the EasyVMAF service on your behalf
  • Video files accessible via a URL or stored in a MinIO bucket on OSC

Step 1: Create a VMAF Studio instance

Via web console

  1. Go to app.osaas.io/dashboard/service/ablindberg-osc-vmaf-studio
  2. Click Create studio
  3. Fill in the fields:
Field Description
name A short identifier for the instance, e.g. myvmaf
oscAccessToken Your OSC personal access token. Used server-side only — never sent to the browser
  1. Click Create and wait for the instance to show status Running

Via CLI

Store your token as a service secret first:

npx -y @osaas/cli secret set vmafstudio osctoken <your-personal-access-token>

Then create the instance:

npx -y @osaas/cli create ablindberg-osc-vmaf-studio myvmaf \
  -o oscAccessToken="{{secrets.osctoken}}"

Step 2: Run a quality analysis

Open the VMAF Studio URL in your browser. The UI lets you:

  1. Enter the URL of a reference video (the original, uncompressed or high-quality source)
  2. Enter the URL of a distorted video (the transcoded rendition to evaluate)
  3. Click Analyze to start the job

VMAF Studio submits the job to EasyVMAF, polls for completion, and displays the aggregate VMAF score and per-frame score chart when ready.

Interpreting results

Score range Interpretation
95–100 Excellent — visually transparent
85–94 Good — suitable for most streaming delivery
70–84 Acceptable — noticeable quality loss on large screens
< 70 Poor — visible artefacts, consider re-encoding

Resources