A subpixel is a fractional position between pixels — a point that doesn't fall exactly on a pixel of the grid but somewhere in between. The concept matters because real-world things don't conveniently sit on pixel boundaries. A moving car shifts by 3.25 pixels between frames, not 3 or 4. A typed character on a high-resolution screen needs to be drawn at a sub-pixel anti-aliased position to look smooth. Codecs and rendering systems handle these fractional positions through interpolation.
In video encoding specifically, subpixels matter for motion-estimation. When the encoder hunts for the matching patch in the previous frame, it doesn't just check whole-pixel offsets — it checks fractional ones too, like "3.25 pixels right, 1.75 pixels down". The encoder synthesises those fractional positions using interpolation filters, finds the best match, and records that fractional offset in the motion vector. The result is dramatically smoother coding of slow camera pans and gentle motion. H.264 uses quarter-pixel precision (steps of 1/4 of a pixel), HEVC the same, AV1 goes to eighth-pixel for some block sizes, VVC has switchable precisions.
For a product team, subpixel motion estimation is invisible plumbing, but it's the reason slow tracking shots and smooth pans compress so much better than they "should". Without subpixel precision, every slow pan would force the encoder to switch between whole-pixel approximations frame to frame, producing visible jittery edges and inflated bitrates. Subpixel precision is also where modern AI video upscalers and frame interpolators live — they work at sub-pixel positions to synthesise frames that didn't exist in the original capture.
