If I understand your message correctly, what you are suggesting is that perhaps the pink/cyan parts of the texture have ‘steeper normals’ than the other sides, like the magenta and dark blue sides?
If so, I have made sure that the colors in the normal map are the correct shades. For example, take the following 6 pixels in this section of the normal map, which all fall on the edge of the tile where the normals would be pointing a little sideways:
The color values of these marked pixels are (starting at the top left, in clockwise order):
51, 201, 197
132, 237, 192
221, 181, 195
211, 60, 196
128, 24, 201
24, 90, 192
To convert a normal vector to a color, you add 1 to each color component and then divide by 2 (as per this article). So to convert a color back to a normal vector, you can multiply each color component with 2 and then subtract 1. Let’s do that for the 6 sampled colors and then print their size and Z-component:
local colors = {
Color3.fromRGB(51, 201, 197);
Color3.fromRGB(132, 237, 192);
Color3.fromRGB(221, 181, 195);
Color3.fromRGB(211, 60, 196);
Color3.fromRGB(128, 24, 201);
Color3.fromRGB(24, 90, 192);
}
for i = 1, #colors do
local nx, ny, nz = colors[i].R * 2 - 1, colors[i].G * 2 - 1, colors[i].B * 2 - 1
local normal = Vector3.new(nx, ny, nz)
print(normal.Magnitude, normal.Z)
end
In the code above I print the magnitude of each vector (which should be 1 for each vector) and the ‘Z’ component of the normal vector, which corresponds to how much the vector is pointing ‘up’. In my case I sampled pixels on the edges of a stone tile which should all have the same amount of steepness. Their X and Y components in the normal vector will be different, but their Z component should all be equal. That said, not every pixel is exactly as far on the edge so there will be a slight difference.
The code above prints:
0.99471116065979 0.545098066329956
0.9973667860031128 0.5058823823928833
0.9970583319664001 0.529411792755127
0.9989075660705566 0.5372549295425415
0.9956383109092712 0.5764706134796143
1.0006917715072632 0.5058823823928833
So each color represents a vector with a length between 0.99 and 1.01 (which makes sense, because there are some rounding issues when you work with floating point numbers. And the ‘steepness’ of each normal vector is between 0.505 and 0.576. So there is a very slight difference in how far ‘up’ these normal vectors are pointing, but that is mostly just because I sampled these pixels by hand, so some of my pixels fall a little more on the edge than others.
I generated this normal map texture with code, so the colors on the edges of every tile are calculated in the same way, so each tile should share this same characteristic where the normal vectors on the sides are all pointing the same amount of ‘up’.
Even if there were a slight difference of up to 7% in how much the normal vectors on one side are pointing up versus another side, I think it is fair to say that the reflection in the original pictures I posted indicate a far larger difference in ‘flatness’ versus ‘steepness’ than that 7% error.
