Visual Information Fidelity (VIF) is an information-theoretic full-reference quality metric that treats the original frame as a source of visual information and asks how much of that information survives compression to reach the viewer's eye, modelled on how the human visual system takes in a scene. It is computed at several spatial scales — four, coarse to fine in the VMAF implementation — so it notices both broad and fine losses, and it is the judge that reacts when compression strips away the subtle texture of skin or foliage. VIF earned its place as one of the elementary features fused inside the classic VMAF (v0) design, alongside the detail-loss metric and a motion feature, where a trained model learned how to weigh it. Its notable twist is that the 2026 VMAF v1 generation drops VIF: it was costly to compute and, once other features improved, no longer added meaningful accuracy, so removing it made VMAF both more accurate and faster — proof that even a strong feature can become redundant.
