I just watched the third video diary and was pretty surprised to hear about the texture compression Lionhead is apparently using:
To quote Tim Rance: "squeeze them down to just one per cent of their original size".
Even wavelet, LBT or weighted least squares based compression algorithms have trouble achieving such ratios for natural images without quite noticeable artifacts. Even then, they seem too expensive (LBT based approaches less so) to decompress in real-time.
If someone has some pointers as to what exactly he is referring to it'd be very much appreciated

The only way this'd make sense to me is to really take his "as a character moves around the world" literally and assume that he is either referring to a streaming system where a low-res approximate is kept in memory (the 1%) and refinements are streamed from disk or if he is talking about things such as wrinkle maps where only differences for spatially localized regions need to be stored (probably even using scalar quantization).

Could we maybe get a more hands-on tech video diary episode?

The 1% comment refers to the animation compression.

Ah, that makes more sense
Thanks for the reply! I really didn't expect someone from Lionhead to post about this; especially not this quick

If you don't mind me asking: I suppose you're using the hardware supported texture compression formats in that case?