Another quick question:
Which is better to use for the pvrtc format, premultiplied alpha, or not premultiplied alpha?
Using premultiplied alpha for textures has several advantages for any compression format, and is generally recommended.
But what are the advantages exactly? Faster? or better quality?
I experimented using the pvr texture tool on a sprite sheet. The best quality I could get was if I bleeded the orginal texture without premultiplying. If I pre multiply the alpha there are more artifacts.
Generally, premulitpled alpha is preferred, because when interpolating values or when filtering, the calculated values will be undesired: https://developer.nvidia.com/content/alpha-blending-pre-or-not-pre
Depending on the source image, the results of texture compression can vary, and experimenting to find the best solution for each case is sometimes necessary.
Here’s another excellent write-up on pre-multiplied alpha. Some of the benefits highlighted specifically relate to image compression quality:
(NOTE: The forums are munging this URL, so just webseardch “premultiplied alpha tom forsyth” and the link will come right up.
