How Scrollproof works

Most virality tools hide their method behind a sales call. We don't. Here is exactly what Scrollproof measures, how it turns that into scores, and — just as important — what those scores do not mean.

What we measure

For every clip, our engine decodes the frames and the audio and computes real, published signals:

• Visual saliency — a per-frame map of where a typical viewer's eye is pulled, using a spectral-residual saliency model. This is the source of the attention heatmap and the attention curve.
• Motion energy — how much changes frame-to-frame.
• Scene cuts & pacing — where hard cuts land and how fast they come.
• Audio energy — loudness (RMS), onsets, spectral flux, and silence, second by second.
• Human presence — whether a face is on screen and how much of the frame it fills. We detect presence only — no face recognition, no identity, no demographics, nothing stored.

How signals become scores

Hook Strength weighs the first ~1 second — early saliency concentration, an audio onset, motion, and whether a cut lands fast. Hold Rate is a retention estimate built from the attention curve's decay and recovery. The Attention Curve fuses per-frame saliency, motion, and audio energy. The Virality Index is a transparent weighted blend of those channels. Every headline number expands to show the measured signal behind it.

What the scores are — and aren't

Scrollproof is a pre-publish creative diagnostic, like a test screening. It predicts how well a clip is built to stop the scroll and hold attention. It is not a guarantee of views, reach, or revenue — real-world performance also depends on your account, posting time, and the platform's algorithm, none of which we can see.

The Virality Index is currently an uncalibrated composite — an informed heuristic, not a validated probability. We label it that way on purpose. As creators opt in to share their real outcomes, we're building a dataset to calibrate it, and when we do we will publish the honest accuracy number — including where it is weak. We would rather earn a real number than borrow an impressive-sounding one.

The heatmap

The "brain" heatmap is a real visual-attention (saliency) map drawn over your keyframes. It is an illustrative model of where the eye is likely pulled. It is not a medical, neurological, fMRI, or EEG reading, and we never imply one.

Your footage

Your clip is uploaded to private storage, analyzed, and then deleted. We never train any model on your content and we never run face recognition. See our Privacy Policy for details.