by Meierhans » 08 Feb 2010, 09:11
Thats a good question. I just started with mapping, did build several branches with big inputs on top to take just a region out of frame, then run the keystones in these branches just at the resolution of this part of the picture to save CPU. Finally add things together again with several ML Advanced that exactly take the input size of the biginput from before... This is not really elegant but works for simple shapes. If mangle would just do a proper bilinear resize, i think it would be much better suitable as you could have a bunch of keystones in one node (as script) which are all taking small parts out of the picture.