How can I make both sides of this sprite be lit?

[west plank]

I'm a complete beginner when it comes to all things shader/shadergraph related, so I copied this shader graph from a tutorial. It works great for allowing 2d sprites to cast 3d shadows and to be affected by lights, but one problem is that the back of the sprite is unaffected by lighting changes. It still casts shadows, but the sprite doesn't change color at all when hit from the back.

Any help is greatly appreciated.

[crude thorn]

This is just how light calculations work in a lit material. They use the surface normal for calculations, but a single quad had only one normal. You would need to duplicate the faces and invert their normals so that both sides are lit properly.

While you might think that rendering the back face should work, it doesn't because it doesn't duplicate geometry(and thus normals).

It would require custom lighting calculationsif you really need it to work as is.

Though, perhaps a small hack might work: use the "is front face" node to determine if it's front or back face, and if it's back, then flip the normals.
Never tried it, but theoretically that might work.

[gilded musk]

I thought, yeah, how hard can this be, but i actually run into an issue doing this in the shader graph

When lighting the front, it looks fine, but when lighting the back, this happens
it looks like it's interpolating with some non-flipped value on the corner vertices (note that this is a plane consisting of 10x10 quads, so it's really only the corner vertices that are "wrong")

maybe there is something else that we cannot modify in the fragment shader, like some tangent data used to do the conversion between spaces

[crude thorn]

Is that in deferred rendering path?

[gilded musk]

forward+

actually on second thought, it's only the corner vertices that are "correct"

[crude thorn]

Perhaps switching to deferred and looking at the gbuffer would reveal some more info

[rancid anchor]

It looks like shadow bias
Shadows must be offset against the geometry normal so they don't appear on the surface that's casting the shadow itself
When flipping normals in fragment stage, shadow bias cannot consider the new normal

Confirm by disabling shadow casting on the mesh if it actually is related to that

Because you are using flat geometry only, you should be able to flip the normal to be always non-negative from camera's perspective without the issue that a surface could have both flipped and non-flipped normals

[gilded musk]

that makes sense, indeed it is self-shadowing

[rancid anchor]

Although I'm not sure if that alone solves the problem

Maybe it does and I'm just not thinking right

[gilded musk]

it does, this works fine

[rancid anchor]

Great!

[gilded musk]

actually, well, maybe not, now the backside is dark again of course, I forgot something 😆

you actually need to know the light direction as well

so instead of WorldPos - CameraPosition it needs the main light position
then it works for a single shadow caster

the main light position (or direction from the POV of the vertex) isn't available in shader graph by default though iirc

with the main light direction it works, but this is an extension taken from https://github.com/Cyanilux/URP_ShaderGraphCustomLighting

I think the easiest solution is to just disable double-sided rendering, duplicate the sprite object, and invert the UVs (or scale the duplicate negatively on X I guess)