The thing about textures is rather simple:
all image files are compressed into jpeg2000 format by SL's algorythm, which not only compresses with loss, but also transforms whatever you are uploading in a ^2-sided file at 100 dpi.
Now there are algorithms and algorithms: the SL algorythm is thought to reduce bandwidth consumption in detriment of quality. The photoshop and other graphical applications algorithms tend to compress towards quality.
So what i generally do is:
- first of all, work LARGE - if you want your final texture file to be 512*512, work at 1024*1024
- use Photoshop (or other) instead of SL to make the compression. That is, don't upload a file 'as it turns out'. Before uploading, use Image Size command in Photoshop to do the following (i generally use one of the bicubic algorithms):
- transform the image into a correct 100 dpi image
- resize the image to a ^2 side image - because even if distorted, photoshop will always do it better than SL. Textures to be repeated (say, a wooden flooring) will also repeat better if the side are powers of two - you won't see so many of those flashing pixels, etc. Remember that you must always reduce the image size rather then enlarging: if your image is 350 px in width, say, it will look better if you reduce the width to 256 instead of enlarging it to 512.
Then i always upload tgas, seeing that tga is a lossless format. So obviously you lose less from converting TGA to jpeg 2000 than from jpeg (in which you have already lost a bit) to jpeg2000.
If you are uploading an image that is not meant to be repeated, and want to maintain strictly its proportions (say, a logo or a painting), then you can add a blank margin to the original file to 'fill in' the gap between the image's size and the next ^2 dimensions.
If you take these precautions, there isn't a significant loss. So you don't need to upload stuff without compression. In any case, if you NEED something with a high res - say it's a commercial poster for a client - then perhaps it pays to upload it 256*256 with no compression rather than 512*512 compressed.
But i can also say that i seldom upload stuff larger than 512 myself - unless say, you need a string of text, and then can upload a 1204*128 (which is smaller and better res as the text is concerned than a 512*512 of the same, in which i need to use smaller fonts and waste a lot of space in the background). So it's this kind of reasoning that should be made.
Sculpties is quite another thing, i think. The issue is that you do work with a very small image to start with, so you have few pixels to define the sculpty shape. What happened is that the quality loss made the sculpties even more clunky than they must be. The reason is that a compression algorithm works by creating useful 'shortcuts' when listing the pixels' characteristics. A lossless format lists the pixels as they ARE. But a compression algorithm can say 'ah, pixel A is 57 green, and B is 53, so i'll list instead that pixels A= B=green 55', thus saving size in the file. But that causes indefinitions and fuzziness in the sculpty, since vertexes get slightly displaced - don't forget that the values of RGB determine the location of the vertexes. So i would say that sculpty textures should be uploaded with no compression.