This just screams challenge to me. You have a 16*16, 8 colour image in 768 bits. So the question is, how much can the format be changed to allow for a wider

This just screams challenge to me. You have a 16*16, 8 colour image in 768 bits. So the question is, how much can the format be changed to allow for a wider range of expressibility? The obvious place to start is increasing the resolution by somehow reducing the bits per pixel. 768/(16x16) = 3 (current resolution makes 3 bits per pixel) 768/(17x17) = 2.657 768/(18x18) = 2.370 768/(19x19) = 2.127 768/(20x20) = 1.920 768/(21x21) = 1.741 768/(24x24) = 1.333 768/(26x26) = 1.136 768/(32x32) = 0.75 Of course, once you get a higher resolution with fewer bits you lose the ability to be able to make each individual pixel any colour you want. Once you go below 1 bit per pixel you lose the ability to refer to pixels individually altogether. Most pixel artists would take some degree of hit though, in fact much of the beauty of pixel art is in massaging the restrictions. A lot of 8 bit systems used 8x8 cells at 1 bit with a colour attribute byte to bring it up to 1.125 bits per pixel. The spectrum used that to make things like http://fc05.deviantart.net/images3/i/2004/116/3/d/ZX_Spectrum_gallery_1997_2003.png http://fc05.deviantart.net/images3/i/2004/116/3/d/ZX_Spectru... You could quite easily support a 24x24 image in that form, but with only 3 cells across the colour restrictions would be a bit harsh. My preference would be to try for 21x21 using 3x3 cells. Being restricted to only 2 colours per cell is much less of a problem when the cell size is small. A quick and easy approach would be a 2 byte per cell structure where the first byte defined the foreground an background colours and the second byte defined the pixels. That gets you a palette of 16 colours overall, any two in each cell. It is marginally over the required limit (21x21 at 1.778bpp) though, so would require a little more cleverness to eek out that extra little bit. dropping back to an 8 colour palette would do it easily, but 1.778 is so close to 1.741 that those bits can surely squeezed out from somewhere.