Fidelity is how closely the video you delivered matches some original, the natural question when you have a pristine master on disk, such as a movie, a finished render, or a recorded stream. Compression throws information away to make files smaller, and fidelity asks how far the result drifted from the source. It is the quantity full-reference metrics like PSNR, SSIM, and VMAF measure, comparing impaired video against the reference frame for frame. The key catch is that fidelity is not the same as how good the video looks: it is possible to have low fidelity and high perceptual quality at once, as when an AI upscaler invents plausible detail that pleases viewers but was never in the original. Fidelity also presumes you possess the master, so it cannot be measured for live broadcast or user-generated content with no reference. Hold the distinction in your pocket: most of quality measurement scores fidelity against a reference, but the hardest real-world cases are about perception with no original at all.
The handbook · 7 volumes
Fora Soft Video Engineering Handbook
Fidelity measures how closely delivered video matches its source, which is not the same as looking good once enhancement or denoising is involved. We wrote seven volumes covering the whole video stack, and for fidelity start with Volume 1: full-reference metrics, compression and detail loss, with Volume 6 on AI enhancement that can raise appeal while lowering fidelity. Written by Nikolay Sapunov, CEO at Fora Soft.
Using AI enhancement and unsure whether fidelity metrics still tell the truth?
Share your source and processed clips: we measure fidelity and perceived quality separately and tell you which metric should gate releases. 250+ video products shipped since 2005.
