Re: Speed adjustment considerations
Posted: Mon Dec 19, 2016 1:37 am
What you're doing is a good test, but it will be a bit different if you ran the gradient from corner to corner than if you do it horizontally or vertically. As a general rule, more complex patterns tend to get chewed up less and you have to wiggle the noise percentage a bit to get it right for some things.
When I settled on 1.6% (all those years ago) it was several versions of the DDS filter plug-in ago. It was probably in 2003 and I was working with DXT1 textures to approximate human skin for 3d characters, which tend to have large portions of the texture at nearly the same colour since people are not generally striped or polka dotted. DXT1 is a particularly unforgiving format, not that DXT3 and DXT5 are loads better. Honestly, some of my first attempts looked like they had been done by an orangutan with house paint.
On a complex texture, like a locomotive or building, there may be times where you need to do one portion with more noise than the rest of the texture since some part are a good deal more complex than others. This was often true working with textures for a human body shape. Especially the sides of the torso, arms, and legs which have very few features to break up the field of colour. And anything that was subtly shaded was very often wrecked on conversion to DDS. The part of the texture for the face was usually sufficiently complex enough to not need much if any noise added.
Now, if I need to do a large field of a single colour, I will set up the gradient (foreground, background) with two nearly identical colours, one shifted about 1% red, the other shifted 1% green, then do a fill with the render clouds filter. Then I add noise as needed to stop banding/colour-blocking. It usually takes more than one try to find the magic number.
Long story short, 1.6% is my goto number because it saves me an hour of tinkering
When I settled on 1.6% (all those years ago) it was several versions of the DDS filter plug-in ago. It was probably in 2003 and I was working with DXT1 textures to approximate human skin for 3d characters, which tend to have large portions of the texture at nearly the same colour since people are not generally striped or polka dotted. DXT1 is a particularly unforgiving format, not that DXT3 and DXT5 are loads better. Honestly, some of my first attempts looked like they had been done by an orangutan with house paint.
On a complex texture, like a locomotive or building, there may be times where you need to do one portion with more noise than the rest of the texture since some part are a good deal more complex than others. This was often true working with textures for a human body shape. Especially the sides of the torso, arms, and legs which have very few features to break up the field of colour. And anything that was subtly shaded was very often wrecked on conversion to DDS. The part of the texture for the face was usually sufficiently complex enough to not need much if any noise added.
Now, if I need to do a large field of a single colour, I will set up the gradient (foreground, background) with two nearly identical colours, one shifted about 1% red, the other shifted 1% green, then do a fill with the render clouds filter. Then I add noise as needed to stop banding/colour-blocking. It usually takes more than one try to find the magic number.
Long story short, 1.6% is my goto number because it saves me an hour of tinkering