Random by tile - brightness variations and mushy seams

[zaar]

Hello!

I've got a scanned material of various plaster-like surfaces (multi angle light to normal and height map in substance). And the physical samples the client sent me are quite small, so I have to use random by tile and offset and rotate them to not get repetitions. But I get slight variations in brightness on the tiles. Is this a know issue, or simply a fact that when rotating the normal map and displacemet, the shading self-shadowing might differ ever so slightly?
I'm super certain I've got things setup correctly. bitmap->normal->randomizer->triplanar. Acctually there's a bit more going on that that, but it's not like I'm putting the corona normal map in the wrong place.

I've attach a screenshot of slate. The only wierd thing going on is I'm multiplying on a contrasted version of the height map on the color map to get more AO, just a quickfix that perhaps should've been managed in substance, but you know... time is a scarse resource.

Apart from the slight variation in brighness. I also get mushy seams. BUT that I know is just how things go when the blending the values of the tiles. Something I'd like the developers to consider is having a look at the OSL "Random Bitmap Tiling2" (by Zap Andersson I presume) that comes with max. It also randomizes by tile, but it also has a wiggliness and wiggliness scale that doesn't make linear seams. Either that or more advanced blending of normal and displacement that doesn't average the values, but rather does some greather than-thing or what ever the correct math would be for letting the heighest pixel win.

[TomG]

On the suggestions for new things in the randomizer (such as wiggliness), please post over on the Ideas Portal at https://chaoscorona.ideas.aha.io/ as that is where we track all user suggestions (and also makes it easy for other users to vote for the suggestions so we know which are the most in demand). Thanks!

[maru]

Could you share an example of those brightness differences?

[maru]

I made a scene with a material setup similar to yours and I couldn't reproduce it. We will need an image or, even better, a scene to understand what is going on.

[zaar]

I've attached a crop of the material in the final render here. Can't show anything else. It only shows the mushy lines. And I'm not getting the difference in brightness right now. I had a previous light setup that might have made the effect stronger, but I've cleared all those renders now.
I'm also suspecting that it might be some moiré or filtering effect of the maps (colour or normal/displacement) that gets affected by the resolution of the render. And when the tiles get rotated a certain direction, they get variations in brightness.

So currently the brightness variation isn't an urgent problem, I'm just curious if I did something wrong or not. Will dig around some more and see if I can find an old render in some draft or PSD file I haven't cleared out yet.
The seams I'm dealing with in post doing old school frequency separation in Ps and cloneing them out.

[maru]

My best guess about the brightness differences is that the normal and/or displacement map could be showing more of a single specific direction so that light interacts with it differently depending on the angle. Something like "drawing" with your finger on a plush material.

As to the mushy lines, are you using the "high quality blending" option in the Mapping Randomizer?

[zaar]

yep, high quality! What is the difference, like technically? As I wrote above I'd like an even higher quality bledning of height and normal values to prevent this. Or is it already trying to do that?

[zaar]

If I wasn't such a noob in Substance Designer I'd create even bigger textures there to avoid the seams. I suspect that someone who's got a year or two in SD behind them knows how to do what corona uvw randomiser (per tile) does, but blend normals and height in a proper way that doesn't average the values over the seam.

[romullus]

Did you try to reduce blending radius? I find that often very small values like 0,01-0,05 gives much better result.

[zaar]

Hmm, never though of going so low. I've gone from 0,25 to 0,15 or 0,1.
But for something so featureless as this, it's probably a good idea! Thanks for the suggestion.

[zaar]

Doing interactive rendering now I'm getting the effect again. It's very faint so if you are on a crappy monitor or under bad viewing conditions it might not be visible at all.
screenshot attached, object cropped out and also made a copy of it with solar curves on it so one can see what's happening better.

I'll try to disable rotation in that randomizer to see if it's the "plush" effect as you called it, maru.

[maru]

That's odd. If it's happening in IR only, it could be due to adaptive light solver / adaptive enviro sampler. Both can be disabled in the Development / Experimental Stuff rollout.

[zaar]

It goes away when I disable rotation in the randomiser. So maybe it's the shape of the displacement. But I it doesn't seem to be very anisotropic in it's nature, it's just grain. But it does seem like the effect is only there in low resolutions with rotation enabled.

[maru]

You can try one more thing: turn random rotation back on and change "normals filtering" to "none". Then check if there are brightness differences.
When set to the default "roughness modulation", this basically turns bump into roughness. My thinking is that when the bump texture is randomized, different tiles may end up with different roughness.

[zaar]

Interesting! I will have a look soon, but I just got some client feedback I need to dive into :) I'll get back on this when it's possible.