"Triplanar" volumetric shader

[uneven rivet]

I have been trying, without success, to make a fog volume shader which samples 3 different textures on different axes.

I rendered the standard xy flipbook head on in embergen, then the xz rotated to the side by 90 degrees, then a final yz rotated above 90 degrees.

I can't figure out a way to blend them together. Doing anything "useful" on the z has been eluding me. Using a 3d texture works just fine but I can't really get 2d textures to do what I want.

Another option I considered was just lerping the showcased texture based on the viewing angle, but then I would need to billboard the volume, which sounds kinda ridiculous and defeats the purpose of it.

Can I fix this?

[clever bane]

Well show what you have tried

[glass panther]

I'm not sure you can meaningfully do triplanar sampling for a volumetric effect

Triplanar shading requires you to have a position and a normal

what is the normal vector for fog?

it doesn't point in a specific directino

[uneven rivet]

Any variation of manipulating W of the uvw. Tried blending with a xz/yz of a 3d procedural noise, tried blending with all 3 floats of the procedural noise.

Even tried adding time to z from the position node and adding back to time for the position input and that also doesn't behave how I'd expect it.

Also, tried a bunch of stuff but honestly, the way I test things is I have a throwaway shader that I keep adding nodes to. Then if that doesn't work, I disconnect them and add more nodes etc.

If I show you the shader graphs, you'll get cancer

[uneven rivet]

I see

that's what happened when I tried to do triplanar sampling for the texture

well

is it techically triplanar if it's using 3 different textures? Maybe not

[glass panther]

I would define "triplanar sampling" as calculating UVs based on three orthogonal planes and then blending the resulting colors together based on how well those planes align with your normal vector

[uneven rivet]

I got this deranged result when I tried

[uneven rivet]

ye, I tried to show them based on camera angle

[glass panther]

Volumetric effects really demand a 3D texture

You need continuous values in every direction

Triplanar shading works because surfaces are 2D

You just need to handle the transitions between the three planes

[uneven rivet]

Right. Since I can't render 3d flipbooks (I don't think anyway), I would probably need to resort to billboarding the local volume component (lol) to account for moving through it

or, like, construct a cube out of the different samples of 2d flipbooks and add a volume inside but that sounds stupid

[clever bane]

You definitely can use three 2D textures projected on each axis, blended together to produce volumetric variance
But the pattern is revealed quite easily because any variation occurs per-axis

[uneven rivet]

What do you mean with the last part?

[glass panther]

how are you deciding where to sample the fog? are you doing raymarching?

I'm using a volumetric fog effect in a VRC world that's raymarched and that uses a 3D noise texture

[uneven rivet]

I'm just making a fog volume shader graph

[glass panther]

ah, right, there's a specific shader graph type for that

[uneven rivet]

3d noise textures are cool and they work, but not for my use case

[clever bane]

Well, rather to say there is no consistency along any one axis, because noise textures have smooth transitions between values which aren't respected in the third dimension using this method

It works but there's lattice-like patterns that can be noticed

When not viewing along any axis directly

[glass panther]

i wonder if using non-orthogonal planes would help

it'd probably just make it look bad from different angles

[uneven rivet]

ahhh yeah, I got that result at some point too

so is this method doomed?

[clever bane]

Well it does produce a result so that's up to you really

[glass panther]

That is better than I was imagining

but yeah, it's going to reveal that you don't have a single continuous 3D noise texture

[clever bane]

How good it is also depends on how much continuity the type of noise requires

[uneven rivet]

well tbf you used a texture which is non-representative for my use case.

I am using textures which are thick around the middle and black on the outside so I won't get much of this behavior, no?

[clever bane]

What is it for then?

[uneven rivet]

I decided to try the hand silt thing the following way:

I forwent the simulation thing for the hand thing. I decided on a compromise, but better than just plain flipbook particles.

I decided I would bake the smoke around the hand with collision with embergen (which I did) and make variations of that.

But because of that, I can't billboard them, I need a consistent shape, hence this "solution"

[clever bane]

Can't think of how you can get better than that with just three
But four textures sampled in a tetrahedral configuration might
A bit silly though, since at that point there's probably no advantage at all over 3D textures

[glass panther]

i do like that idea

quadplanar sampling

i found an example of someone doing six planes

[uneven rivet]

well, since I can't render 3d texture flipbooks, there is an advantage

[clever bane]

What does it look like?

[clever bane]

It has at least more than no potential, depends on the shape
This example using the same texture on all axes, which doesn't really correspond to an actual axial view of something, but it exhibits similar limitations

Anything that's occluded from the projections will start forming into the 3D gridlike patterns

If the textures also had depth information, maybe that'd enable some clever workarounds

But even without anything fancy this might work for simple use cases

Ultimately it's not possible to avoid losing data when stuff is overlapping

[uneven rivet]

what about unironically using 2 or 3 overlapping fog volumes

that somehow hasn't occured to me till now

[clever bane]

I'd be surprised if they can
But more surprised if that's anywhere near acceptable regarding performance

[uneven rivet]

they can what? That will works mechanically. Whether it will be acceptable in a multitude of different ways remains to be seen lol

[uneven rivet]

well, it kind of works

it's not perfect but things could be tweaked to make it look better

and it has a negligible effect on performance

the lighting and everything is trash in this project, but here's just to get an idea of what it does for me

shader is trash, scaling should be tweaked etc etc

[clever bane]

The combined result looks to be the same as what you'd get with just one volume if projecting from multiple axes

Now it appears you're projecting from just one

[uneven rivet]

that was supposed to show you the switch

when I enabled the other volume

[clever bane]

Right
It does look like the second volume can "carve" density out of the first
Which is interesting but probably doesn't make a difference here, since you can do that within the graph just as well

[uneven rivet]

I didn't actually manage to do that tho

doesn't work like projecting textures on planes as Fen said

[clever bane]

That sounded like it was related to a misunderstanding

I'm not in a position to test it, but I'm not aware of any reason why the same technique I used earlier in blender wouldn't work in shader graph

It should be as simple as sampling 3 textures, each using object space coordinates swizzled into xy, xz and zy respectively

Multiplied together

[uneven rivet]

I already tried exactly that but I get a barcode texture every time when I sample it on either xz or yz

I assume it has something to do with the slices

It doesn't really project it normally, it kind of divides it into slices where every slice is same on the y (for the xz one)