How video compression works

Uncompressed video is absurd. A single 1920×1080 frame is about six megabytes of raw pixels, and at 30 frames a second that is 180MB per second — roughly 11GB a minute. No camera stores that, and no network moves it. Every video you have ever watched was compressed before you saw it.

The big saving: storing change, not frames

The single largest idea in video compression is that consecutive frames are nearly identical. In a shot of someone talking, the wall behind them is the same in frame 100 as it was in frame 99. Storing that wall 30 times a second is waste.

So encoders store occasional complete frames — keyframes — and describe every frame in between as a set of differences from its neighbours: this block moved four pixels left, this region is unchanged, this patch got slightly brighter. A static interview shot compresses enormously because almost nothing changes. Confetti, rain, a handheld pan across foliage — where every pixel changes every frame — compresses very little.

The second saving: throwing away detail you won't miss

Within each frame, the encoder discards information selectively. Human vision is far more sensitive to brightness than to colour, so colour is stored at lower resolution than brightness — a trick called chroma subsampling that has been standard since analogue television. Fine gradients get simplified. High-frequency detail in busy areas gets rounded off, because you are unlikely to notice a missing texture in a bush.

Push this too far and you see the failure modes: blocky patches in flat areas like skies, smearing in fast motion, and halos around hard edges. Those are the encoder running out of bits and rounding harder.

The three controls that matter

Frame rate is a fourth lever in principle — halving it halves the number of frames to store. It is not offered here, because dropping a 60fps clip to 30fps visibly changes how motion reads, and that is a decision worth making deliberately rather than having a compressor make it quietly.

What Dexta actually does

Compress Video re-encodes with H.264 when you export MP4 or MOV, and VP9 when you export WebM. The quality presets set a CRF value — 22, 27 or 33 for H.264; 28, 34 or 40 for VP9 — and the resolution cap scales the height down without ever scaling up. Audio is re-encoded to AAC at 128 kbps, or Opus at 96 kbps for WebM, and a clip with no audio track gets none added.

Mistakes worth avoiding

• Compressing something that is already compressed. Anything downloaded, exported from an editor, or received through a messaging app has been encoded at least once. A second pass costs quality for very little size.
• Compressing before you edit. Every edit that re-encodes costs a little quality; starting from a compressed file means those losses stack. Compress last.
• Compressing hard before uploading to social media. The platform re-encodes anyway, and it will do so from your already-damaged file.
• Assuming a percentage. Nothing here predicts your file. Run one export and read the before-and-after figure the tool reports.