Noise Texture Fac output fails when using exact unit inputs and exact scale values #87781

New Issue

Gorgious · 2021-04-24T19:28:06+02:00

Gorgious commented

2021-04-24 19:28:06 +02:00

System Information
Operating system: Windows-10-10.0.19041-SP0 64 Bits
Graphics card: NVIDIA GeForce RTX 3070/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 465.89

Blender Version
Broken: version: 2.92.0, branch: master, commit date: 2021-02-24 16:25, hash: 02948a2cab
Broken on the latest 2.90, 2.93 and 3.0.0 commits I could find.
Also broken on 2.80.75 but there was no W input back then.
Also broken on 2.79b

Short description of error
The Noise Texture Node exhibits weird values for its Fac output (and by extension*, the Red channel of it Color Output) when using default properties. The Green and Blue channels seem to work properly though.
The Fac output of the Noise Texture Node produces unexpected results on discrete unit inputs, and on exact unit scales. The W seed also has an effect when it's set to an increment of 0.5.

*As you may know the red channel of the color output of the noise texture contains the same values as the Fac output.

Exact steps for others to reproduce the error
Default Scene. Create a cube and scale it way up in edit mode.

Create a new material, Texture Coordinates node > Object Output > Separate XYZ > Z Output > Math node set to Ceil, Round, Floor, or Truncate > Noise Texture > Fac Output

Problem : If you keep the noise texture at the default scale of 5, the shader will produce a continuous value of 0.5 on the whole mesh

Expected result : Get a different value for each subsequent strata of the material, as exhibited if you plug in the color output or try out the green or blue channel of the color output.

Troubleshooting : If you change the scale of the noise texture node, the behaviour of the randomization seems to work again.

But, there is a band that consistently stays at exactly 0.5 value whatever the scale.
This may be intended behaviour since the input / output values at Z = 0 is the same whatever the scale. However the same behaviour isn't shown on the Green and Blue channels of the color output...

And, every time the scale is set to a precise unit, (1.000, 2.000, etc.) the continuous value of 0.5 appears again.

Also, there is a weird banding of exactly 0.5 output values when using scales of *.500, .250,.125, scale values :

Moreover, if the user chooses to use 4D input, and tweaks the W value to something other than 0 (I assume it is the default seed value even when using 1D), the random expected behaviours is again present.

But there are certain combinations that produces an unexpected number of values that are exactly 0.5. Example :

Also, if the W value is set to an increment of a half unit (0.5, 1.0, 1.5, etc.), this happens.

For Pair scale unit size (eg 46), continuous grey (0.5) value

For odd scale unit size (eg 45), unexpected number of exactly 0.5 value bands.

Maybe this is intended behaviour but it seems odd that the Fac outputs returns a single value when the Green and Blue channels act "truely" randomly in this case.

noise_fac_test.blend

**System Information** Operating system: Windows-10-10.0.19041-SP0 64 Bits Graphics card: NVIDIA GeForce RTX 3070/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 465.89 **Blender Version** Broken: version: 2.92.0, branch: master, commit date: 2021-02-24 16:25, hash: `02948a2cab` Broken on the latest 2.90, 2.93 and 3.0.0 commits I could find. Also broken on 2.80.75 but there was no W input back then. Also broken on 2.79b **Short description of error** The Noise Texture Node exhibits weird values for its Fac output (and by extension*, the Red channel of it Color Output) when using default properties. The Green and Blue channels seem to work properly though. The Fac output of the Noise Texture Node produces unexpected results on discrete unit inputs, and on exact unit scales. The W seed also has an effect when it's set to an increment of 0.5. *As you may know the red channel of the color output of the noise texture contains the same values as the Fac output. **Exact steps for others to reproduce the error** Default Scene. Create a cube and scale it way up in edit mode. Create a new material, Texture Coordinates node > Object Output > Separate XYZ > Z Output > Math node set to Ceil, Round, Floor, or Truncate > Noise Texture > Fac Output Problem : If you keep the noise texture at the default scale of 5, the shader will produce a continuous value of 0.5 on the whole mesh ![image.png](https://archive.blender.org/developer/F10035633/image.png) Expected result : Get a different value for each subsequent strata of the material, as exhibited if you plug in the color output or try out the green or blue channel of the color output. ![image.png](https://archive.blender.org/developer/F10035635/image.png) Troubleshooting : If you change the scale of the noise texture node, the behaviour of the randomization seems to work again. ![image.png](https://archive.blender.org/developer/F10035638/image.png) But, there is a band that consistently stays at exactly 0.5 value whatever the scale. This may be intended behaviour since the input / output values at Z = 0 is the same whatever the scale. However the same behaviour isn't shown on the Green and Blue channels of the color output... ![image.png](https://archive.blender.org/developer/F10035640/image.png) And, every time the scale is set to a precise unit, (1.000, 2.000, etc.) the continuous value of 0.5 appears again. ![image.png](https://archive.blender.org/developer/F10035642/image.png) Also, there is a weird banding of exactly 0.5 output values when using scales of *.500, *.250,*.125, scale values : ![image.png](https://archive.blender.org/developer/F10035729/image.png) ![image.png](https://archive.blender.org/developer/F10035720/image.png) Moreover, if the user chooses to use 4D input, and tweaks the W value to something other than 0 (I assume it is the default seed value even when using 1D), the random expected behaviours is again present. But there are certain combinations that produces an unexpected number of values that are exactly 0.5. Example : ![image.png](https://archive.blender.org/developer/F10035673/image.png) Also, if the W value is set to an increment of a half unit (0.5, 1.0, 1.5, etc.), this happens. For Pair scale unit size (eg 46), continuous grey (0.5) value ![image.png](https://archive.blender.org/developer/F10035676/image.png) For odd scale unit size (eg 45), unexpected number of exactly 0.5 value bands. ![image.png](https://archive.blender.org/developer/F10035678/image.png) Maybe this is intended behaviour but it seems odd that the Fac outputs returns a single value when the Green and Blue channels act "truely" randomly in this case. [noise_fac_test.blend](https://archive.blender.org/developer/F10035702/noise_fac_test.blend)

Gorgious commented

2021-04-24 19:28:06 +02:00

Added subscriber: @nathan.hild

Bertrand Coconnier commented

2021-06-12 01:37:37 +02:00

Added subscriber: @bcoconni

Bertrand Coconnier commented

2021-06-12 09:56:41 +02:00

I encountered this exact same problem. It took me an awful lot of time before realizing that only the green and blue components of the noise texture are "noisy", unlike the red component which holds a constant value. So I had to add a node "Separate RGB" after "Noise Texture" just to get random values using the texture green component rather than normally using "Fac". Thanks to @nathan.hild, I now know that setting the scale to a non integer value restores the "noisy" output from the "Fac".

Please fix that bug. Thanks.

I encountered this exact same problem. It took me an awful lot of time before realizing that only the green and blue components of the noise texture are "noisy", unlike the red component which holds a constant value. So I had to add a node "Separate RGB" after "Noise Texture" just to get random values using the texture green component rather than normally using "Fac". Thanks to @nathan.hild, I now know that setting the scale to a non integer value restores the "noisy" output from the "Fac". Please fix that bug. Thanks.

Philipp Oeser commented

2021-08-19 17:24:21 +02:00

Added subscriber: @lichtwerk

Philipp Oeser commented

2021-08-19 17:24:21 +02:00

Changed status from 'Needs Triage' to: 'Confirmed'

Philipp Oeser commented

2021-08-19 17:24:22 +02:00

Can confirm the behavior.
(havent checked code if there is a deeper reason for this, but will confirm for now)

Can confirm the behavior. (havent checked code if there is a deeper reason for this, but will confirm for now)

Philipp Oeser commented

2021-08-20 09:26:54 +02:00

Added subscriber: @CharlieJolly

Philipp Oeser commented

2021-08-20 09:26:54 +02:00

@CharlieJolly : do you know more?

Charlie Jolly commented

2021-08-20 10:36:59 +02:00

Added subscriber: @OmarEmaraDev

Charlie Jolly commented

2021-08-20 10:36:59 +02:00

This is similar to #83144, the noise is generated procedurally using cells, when the inputs are set to certain values (normally whole numbers), this can result in zero noise at the cell boundaries. It can be mitigated by offsetting the values. It is not ideal but I think should be regarded as a known issue.

Tagging @OmarEmaraDev in case he has any comments to add.

This is similar to #83144, the noise is generated procedurally using cells, when the inputs are set to certain values (normally whole numbers), this can result in zero noise at the cell boundaries. It can be mitigated by offsetting the values. It is not ideal but I think should be regarded as a known issue. Tagging @OmarEmaraDev in case he has any comments to add.

Omar Emara commented

2021-08-20 11:18:34 +02:00

Charlie's assessment is correct. The reason why the green and blue channels seem "more random" is that we add a random offset to their coordinates as a form of seed already. I will update the documentation of the noise node to note that behavior.

Philipp Oeser commented

2021-08-20 11:33:49 +02:00

I will update the documentation of the noise node to note that behavior.

Thx. I think once this is properly documented, this report could be closed then.

> I will update the documentation of the noise node to note that behavior. Thx. I think once this is properly documented, this report could be closed then.

Gorgious commented

2021-08-20 12:18:46 +02:00

Thanks for your time and all the clarification, I agree it's absolutely okay to have this kind of limitation if it's properly documented.

Omar Emara commented

2021-08-20 16:06:34 +02:00

Added documentation in rBM8331.

Philipp Oeser removed the

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Noise Texture Fac output fails when using exact unit inputs and exact scale values #87781