question of quality - rendering

have been thinking about this one a lot recently, and am totally torn between 2 mindsets, so I thought I would throw it open and see what everyone else thinks.

is it better, from a quality of final image standpoint, better to render your texture larger than needed and resize down, say do clothing in 1024 so small details and such stand out easier, then resize to 512, as thats the highest SL will display.

OR

render at 512 and totally avoid any resizing issues and have it come out as best as possible for that size?

ask this cause I read a lot here about making it bigger then resizing

When creating a texture, I actually start with a 2048x2048 image in PSP or GIMP. Once I get everything the way I like it, I scale it down to 512x512 and examine it at 4x zoom to see what detail is lost. If too much is lost, and it's intended for a very large prim; I'll undo the resize, go to 1024x1024 and upload it that way. If I'm not as worried about the detail, I'll downsize it once more to 256x256. (^_^)

For anything clothing layer... I've never done a thing over 512x512, simply because I've never seen an example of a clothing texture looking any higher-rez than that after baking. (^_^)

Heck... Even my 'for sale' boxes are 512x512. And my store is all 'blank' texture. Because of this, it's a pretty fast rez and I'd be hard pressed to see any issues with detail in the place. (^_^)y

I was meaning more for those that use something like maya or 3ds to bake the textures they've already created in photoshop os PSP.

I totally get why to do them larger to begin with when you're working on them by hand, makes it a lot easier to get in the smaller details and such ... grins

but with 3ds/maya they wrap the model with that texture, so a large input texture is good, but the output... does it need to be bigger than 512.

without going into a big discussion of how things like raytracing work, suffice to say that the textures you put on "should" be perfectly set on the model, with the raytracing taking care of any resizing by just not sampling so many pixels

I *always* make textures bigger and shrink them before importing. Yout graphics application will then interpolate to smooth the edges better than you could possibly draw them at a 1 to 1 ratio. So your edges will look far smoother if you start biggerm, then scale down the final art.

For baking, it's always best to go as big as possible. The more pixels are on the canvas, the more precisely everything can be traced. If you bake at a small size, you'll find that things like light bleed and other artifacts end up having a huge impact. But bake at a large size, and all anomalies will carry relatively less impact. Furthermore, shrinking the textures in post can often smooth out or altogether eliminate certain artifacts. From a quality-first perspective, it's an all around win-win to bake big, and then downsize afterward.

That said, there are some practical drawbacks. Most obviously, bigger textures take longer to render. Where a 512x512 might take a minute, something like a 4096x4096 could take an hour. Also, bigger textures require far more memory and processing power. The bigger the render, the greater the chance of crashing part way through, and the more time-consuming it is to start over. So, depending on the project, it's not always practical to go big. If time considerations are not a factor, then by all means, go as big as you possibly can. But if you're on a budget, sometimes the best course has to be simply "get it done".



On a related note, for quality bakes, the physical size of the model can make a difference. The larger the model, the more photons and rays can hit it at minimal settings. If the model has to remain small because of scene complexity, you can always increase render settings to get a good bake, of course. But in simple scenes, it's usually faster and easier just to upsize the model itself.