VMAF — Video Multi-Method Assessment Fusion — is the modern industry-standard quality metric, designed by Netflix to measure perceived video quality on a single understandable scale. It outputs a number from 0 to 100, deliberately calibrated against thousands of real human ratings: viewers consistently rated content scoring 20 as "bad", and 100 as "excellent". A VMAF of 93 means "looks essentially perfect to most people"; 80 means "noticeably compressed but acceptable"; 50 means "clearly compromised".
What VMAF actually does is fuse several measurements into one — visual fidelity, detail loss, motion across frames — and run them through a model trained against real subjective viewer ratings. It tracks human perception much better than PSNR or SSIM, especially for the kinds of quality differences that matter to streaming services: scene-cut visibility, motion smoothness, banding on gradients, faces and skin tones. Netflix, YouTube, Meta, Disney+ and basically every major streaming service uses VMAF as the authoritative metric for encoding decisions.
For a product team, VMAF is the metric to plan around. Two practical anchors. VMAF 93 is the standard target for premium streaming — at that level, the vast majority of viewers will not perceive any quality loss vs the master. VMAF 80 is the standard target for the lowest rung of an adaptive bitrate ladder — acceptable on a phone screen, mediocre on a TV. Encoding choices (codec, bitrate, preset, per-title settings) are usually evaluated by their VMAF curve: "at what bitrate does this codec hit VMAF 93?". A 30 % reduction in bitrate at the same VMAF is a 30 % saving on CDN costs — that's the language streaming engineering speaks in 2026.
