Crop Video

H.264 does not store a colour value for every pixel. Human vision is far more sensitive to brightness than to colour, so the format keeps full-resolution brightness and one colour sample per 2×2 block — a trick called chroma subsampling that saves a great deal of space for no visible cost.

It does mean a frame has to divide evenly into those blocks. A crop 401 pixels wide, or one starting at x=17, cuts a colour block in half, and the encoder refuses the job rather than guessing. Tools that pass the selection straight through produce a failure that talks about pixel formats and says nothing about the rectangle you drew.

So the rectangle is rounded down to even numbers on every side before it reaches the encoder, silently. At most one row and one column go missing, which nobody can see, and the alternative is an error nobody can act on.

Aspect presets and the fraction trap

Your selection is held as fractions of the frame rather than pixels, so that it survives the video being displayed at whatever size fits your screen. That is the right way to store it and the wrong way to apply a ratio: a selection half as wide and half as tall as a 16:9 frame is not square, it is 16:9.

The presets convert into real pixels, apply the ratio there, and convert back. A 1:1 selection is a genuine square in the output file, which is what a social platform will measure it as.

Cropping never scales up

You get exactly the pixels inside your rectangle, at the resolution they already had. A tight crop of a 4K clip is still sharp; the same crop of a 480p one is not, because there was never any more detail there. No amount of processing invents it, and anything claiming otherwise is guessing at pixels.