There is an ongoing quest, for those of us texturing clothes and things in Second Life™, to upload better quality images. So, here is some new information on making better textures or aka diffuse layers. See Compression Depression.
Beq Janus, is one of the people working with the Firestorm team and has tweaked the mesh uploader and other parts of the viewer, been involved in tracking down bugs, and figuring out workarounds for various problems. Now she is on about a discovery made with the help of Whirly Fizzle, all starting with one of Hamlet Au’s New World Notes posts about Frenchbloke Vanmoer… who?
Whoever, down in the Debug Settings it is said there is some magic. Unfortunately, it isn’t there, often as is the case with magic. The magic appears surprising to some as they do not understand the trick and think the wand, hand waving, and abracadabras actually do something magical.
However, there is a lesson to be learned and another step in the quest for better SL image uploads.
Beq gives the geeky explanation. SO, read her article if you are into the tech. Otherwise, resize your images to 1024×1024, 512×512, 256×256… etc. Or combinations of those numbers* to prevent the viewer from crushing your image with its resampling algorithm. Adobe’s algorithm in Photoshop is FAR superior. But, many of us knew that. The new part is bi-cubic as recommended by Adobe for downsizing is NOT the best for use with Second Life. Bi-linear is.
We are not going to get a huge boost in quality with this new knowledge. However, it might be enough that we can use a 512 in place of a 1024-image.
So…
Is Adobe wrong? Not exactly. Adobe considers the wide range of sizes, quality and a varying amount of detail in ALL images. I suspect they recommend the process most likely to work best for the largest range of images. In Second Life we are dealing with a subset of images, a narrow range. It looks like in this specialized range of images bi-linear is better.
As Beq says, this is really a matter that is up to the creative doing the image. I suspect some may look better after a bi-cubic resampling, others if bi-linear is used, and still others just suck no matter what you do. The point is to be in doubt and test to find what works best for your image.
*We can mix sizes. Instead of a 1024×1024 image, we might use a 1024×512. Rather than use two 512 images for texturing we could combine the two into a 1024×512 or four 512 images into a 1024×1024. The advantage is fewer images to download.
Each download has some overhead; checking to see if the image is in the cache, waiting for the CDN to check its cache, and the three systems in the process to negotiate a download. The overhead is minimal but it helps if we eliminate it.