Originally based on OpenDivX, it's a Mpeg-4 compliant codec. (fully opensource project).
MPEG-4 is very effective in compression, because it reuses data from one image to generate the next one (still images of a video are only 1/25 of a second or even less apart from each other, in almost all cases they look very similar).There are different ways how to detect which part of the picture can be reused, and QPel = 'Quarterpel Motion Compensation' and GMC = 'Global Motion Compensation' are two advanced ways of doing this. Normally you only use 'Halfpel Motion Compensation', so the resolution is higher with Quarterpel.
A different step in compression is to divide some coefficients by a number to gives them smaller values (and hopefully zero). This is called 'Quantization'. Normally, every 16x16 block in the image is divided by the same value, but with 'adaptive quantization', for every block a different value can be chose.
This _can_ lead to better quality, when e.g. important parts of a picture like faces are compressed less, and less importants part (background) are compressed stronger, but it's still somewhat experimental.