HOW IT WORKS

How CutScore reads a video.

Give it a file or a link. Everything measurable is computed deterministically; AI is reserved for the few judgments a number can't make. You get a 0–100 score, the evidence, and concrete fixes — as a coaching report.

THE PIPELINE · 3 STEPS

From a file to fixes.

STEP 01

Give it a video

Drop a file or paste any link — YouTube, Vimeo, a raw export, up to 4K. CutScore demuxes the streams and detects the genre on its own, which sets the weighting for everything that follows.

STEP 02

It reads the craft

Frames, audio and the edit are measured against published standards — the same way every time. The only step that uses AI is the subjective read: aesthetics, text design and synthesis.

STEP 03

You get the fixes

A global score, per-family scores, prioritized fixes with timestamps and a how-to for each, plus the evidence behind every call — delivered as a shareable coaching report.

METHOD

Objective by design.

Anything that can be measured is measured deterministically — the same way every time. AI is reserved for the judgment calls a number can't make, and your footage is never handed to a third-party model to produce a score.

Measured, not guessed

ffmpeg, OpenCV, PySceneDetect and an EBU R128 meter do the quantitative work — exposure, sharpness, stability, loudness, cut detection. Re-score the same file and you get the same numbers.

AI only where it belongs

Aesthetics, on-screen text design and the written synthesis are subjective — that's where a model helps. Everything objective stays deterministic and reproducible — and your footage isn't sent to a third-party model to be scored.

Held to real standards

We grade against published references, not house opinion — so the targets mean something outside CutScore.

EBU R128 loudness WCAG ≥ 4.5:1 text −1 dBTP true peak
GENRE-AWARE

One engine, the right bar for each video.

CutScore detects whether you made a vlog, talking-head, tutorial, product demo, reaction, podcast, short or brand film, then shifts the weighting of the 13 families to match. A handheld vlog isn't penalized for not looking like a brand film, and a brand film is held to a higher image bar. The detection and weighting happen automatically — you don't pick a mode.