The Noise Texture node evaluates a fractal Perlin noise at the input texture coordinates.
Texture coordinate to evaluate the noise at; defaults to Generated texture coordinates if the socket is left unconnected.
Texture coordinate to evaluate the noise at.
Scale of the base noise octave.
Number of noise octaves. The fractional part of the input is multiplied by the magnitude of the highest octave. Higher number of octaves corresponds to a higher render time.
Blend between a smoother noise pattern, and rougher with sharper peaks.
Evaluate the noise in 1D space at the input W.
Evaluate the noise in 2D space at the input Vector. The Z component is ignored.
Evaluate the noise in 3D space at the input Vector.
Evaluate the noise in 4D space at the input Vector and the input W as the fourth dimension.
Higher dimensions corresponds to higher render time, so lower dimensions should be used unless higher dimensions are necessary.
Value of fractal noise.
Color with different fractal noise in each component.
While the noise is random in nature, it follows a certain pattern that might not evaluate to random values in some configurations. For instance, consider the following configuration where a grid of objects have a material that evaluates a noise texture at their locations. One might expect the objects to have random values since they have different locations, but this is not the case.
It seems all objects have a value of 0.5. To understand why this happens, let us look at the following plot of a 1D noise texture.
The horizontal line denotes a value of 0.5 and the vertical lines denotes whole numbers assuming a noise scale of 1. As can be seen, the noise always intersects the 0.5 line at whole numbers. Since the aforementioned objects were distributed on a grid and have whole number locations, they all evaluate to 0.5. Which explains the issue at hand.
Generally, any discrete evaluation of noise at integer multiples of the reciprocal of the noise scale will always evaluate to 0.5. It also follows that evaluations closer to that will have values close to 0.5. In such cases, it is almost always preferred to use the White Noise Texture.
Regardless, one can mitigate this issue in a number of ways:
Adjust the scale of the noise to avoid aligning the noise with the evaluation domain.
Add an arbitrary offset to the texture coordinates to break the alignment with the evaluation domain.
Evaluate the noise at a higher dimension and adjust the extra dimension until a satisfactory result is achieved.
Similarly, in other configurations, one might experience some banding patterns in the noise, where there are bands of high contrast areas followed by banding of low contrast areas. For instance, planar surfaces that are slightly tilted along one of the axis will have such a banding pattern.
This happens because the slight tilt along one of the axis causes values along the perpendicular axis to change very slowly making the grid structure of the noise more apparent. The easiest way to mitigate this issue to rotate the coordinates by an arbitrary amount.