Mantaflow: Noise doesn't follow smoke #81017

New Issue

Iago Mota · 2020-09-21T03:43:36+02:00

Iago Mota commented

2020-09-21 03:43:36 +02:00

System Information
Operating system: Windows 10
Graphics card: Nvidia GeForce GTX 650 Ti
Processor: Intel® Core™ i7-3770

Blender Version
Broken: Blender 2.90
Worked: Blender 2.81

Short description of error
After enabling "Noise", the Wavelet turbulence doesn't follow the smoke. It seems to only mildly adapt to certain movements; but, unlike in 2.81 it doesn't get "dragged" by the velocity of the smoke.
NoiseDifference-C.mp4

Exact steps to reproduce the error
In any smoke simulation (preferably one where smoke is being constantly emitted), enable "Noise" and bake.

2.9 Project file: 2.9 Example.blend

2.81 Project file for comparison: 2.81 Example.blend

**System Information** Operating system: Windows 10 Graphics card: Nvidia GeForce GTX 650 Ti Processor: Intel® Core™ i7-3770 **Blender Version** Broken: Blender 2.90 Worked: Blender 2.81 **Short description of error** After enabling "Noise", the Wavelet turbulence doesn't follow the smoke. It seems to only mildly adapt to certain movements; but, unlike in 2.81 it doesn't get "dragged" by the velocity of the smoke. [NoiseDifference-C.mp4](https://archive.blender.org/developer/F8904658/NoiseDifference-C.mp4) **Exact steps to reproduce the error** In any smoke simulation (preferably one where smoke is being constantly emitted), enable "Noise" and bake. 2.9 Project file: [2.9 Example.blend](https://archive.blender.org/developer/F8904663/2.9_Example.blend) 2.81 Project file for comparison: [2.81 Example.blend](https://archive.blender.org/developer/F8904665/2.81_Example.blend)

Iago Mota commented

2020-09-21 03:43:36 +02:00

Added subscriber: @Iago-Diogo-de-Vasconcelos-Mota

Richard Antalik commented

2020-09-29 18:36:50 +02:00

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

Iago Mota commented

2020-10-06 16:01:51 +02:00

A small update here. I was experimenting with different resolutions; and it certainly improves in quality as the upres factor is increased, but the noise still fails to follow the movement of the original simulation.

2.90.1Resolutions.mp4

2.90.1 Example.blend

A small update here. I was experimenting with different resolutions; and it certainly improves in quality as the upres factor is increased, but the noise still fails to follow the movement of the original simulation. [2.90.1Resolutions.mp4](https://archive.blender.org/developer/F8961177/2.90.1Resolutions.mp4) [2.90.1 Example.blend](https://archive.blender.org/developer/F8961168/2.90.1_Example.blend)

Jacques Lucke commented

2020-10-13 16:59:59 +02:00

Added subscribers: @sebbas, @JacquesLucke

Jacques Lucke commented

2020-10-13 16:59:59 +02:00

@sebbas is this a bug or just working differently in mantaflow?

Sebastián Barschkis commented

2021-01-27 14:44:06 +01:00

Investigated a bit and have an update:

TLDR: This is an issue which the old smoke simulation system didn't have since it worked differently.

The adaptive time-stepping is likely the cause of this problem: The current setup only uses it for the base simulation. The wavelet simulation (any resolution) runs at (fixed) 1 simulation step per frame. So, while the base simulation can simulate multiple smaller steps in one frame, the wavelet simulation cannot. Wavelet smoke plumes can therefore take different shapes.

Turning off "Adaptive Time Steps" should make wavelet smoke "follow" the base smoke better. The overall quality of the simulation might be not as good though.

Next step for me is to expand the adaptive time-stepping loop so that noise also makes uses of it. This should fix the issue.

Investigated a bit and have an update: *TLDR: This is an issue which the old smoke simulation system didn't have since it worked differently.* The adaptive time-stepping is likely the cause of this problem: The current setup only uses it for the base simulation. The wavelet simulation (any resolution) runs at (fixed) 1 simulation step per frame. So, while the base simulation can simulate multiple smaller steps in one frame, the wavelet simulation cannot. Wavelet smoke plumes can therefore take different shapes. Turning off "Adaptive Time Steps" should make wavelet smoke "follow" the base smoke better. The overall quality of the simulation might be not as good though. Next step for me is to expand the adaptive time-stepping loop so that noise also makes uses of it. This should fix the issue.

Iago Mota commented

2021-02-01 23:07:40 +01:00

Unfortunately, I'm not noticing a strong difference with Adaptive Time Steps disabled (in 2.91 Beta). I'll explore this with other simulations, though, to see if the results are any different. (upres is at 2 for this one)

2.91ASubstepOffTST.mp4

Unfortunately, I'm not noticing a strong difference with Adaptive Time Steps disabled (in 2.91 Beta). I'll explore this with other simulations, though, to see if the results are any different. (upres is at 2 for this one) [2.91ASubstepOffTST.mp4](https://archive.blender.org/developer/F9608158/2.91ASubstepOffTST.mp4)

hal9000 commented

2021-07-02 02:04:45 +02:00

Added subscriber: @hal9000

Iago Mota commented

2022-01-23 20:09:01 +01:00

Alright, a bit of an update here. Mostly, I just think that this illustrates the issue in question a bit better than my previous examples. Looking back at my previous posts, there was quite a lot going on in them, plus in the last ones, the smoke was moving quite fast. And, regarding issues with Noise and fast moving smoke, I feel like I might've been entering #78738 territory, if these are indeed separate problems.

But, anyhow. I think this shows a bit more clearly how the Noise really doesn't seem to want to properly follow the movement of the smoke in Blender 3.0. Though, I understand this is likely a really challenging issue to solve. Even though this footage makes the problem more clear, you can still see the noise being pushed around and morphing here and there. So I don't think it's as simple as the movement of the noise not being affected by the base smoke sim at all.

B3NoiseBugC.mp4

Now, another thing I couldn't help but notice was the apparent difference in quality. While looking more noisy, the 3.0 version seems a bit less detailed? This could likely be a separate issue, or even that, despite both simulations having equal resolution settings (86 for the base divisions, and 2 high-resolution divisions), there is still a difference in total voxel count between the two? I apologize if it's a bit hard to see. I'll do a bit more testing on the issue to know whether this is a valid thing to point out!

Anyway, here's the project file of the 3.0 scene here.
Noise_Movement_Bug_Smokesim.blend

Alright, a bit of an update here. Mostly, I just think that this illustrates the issue in question a bit better than my previous examples. Looking back at my previous posts, there was quite a lot going on in them, plus in the last ones, the smoke was moving quite fast. And, regarding issues with Noise and fast moving smoke, I feel like I might've been entering #78738 territory, if these are indeed separate problems. But, anyhow. I think this shows a bit more clearly how the Noise really doesn't seem to want to properly follow the movement of the smoke in Blender 3.0. Though, I understand this is likely a really challenging issue to solve. Even though this footage makes the problem more clear, you can still see the noise being pushed around and morphing here and there. So I don't think it's as simple as the movement of the noise not being affected by the base smoke sim at all. [B3NoiseBugC.mp4](https://archive.blender.org/developer/F12819408/B3NoiseBugC.mp4) Now, another thing I couldn't help but notice was the apparent difference in quality. While looking more noisy, the 3.0 version seems a bit less detailed? This could likely be a separate issue, or even that, despite both simulations having equal resolution settings (86 for the base divisions, and 2 high-resolution divisions), there is still a difference in total voxel count between the two? I apologize if it's a bit hard to see. I'll do a bit more testing on the issue to know whether this is a valid thing to point out! Anyway, here's the project file of the 3.0 scene here. [Noise_Movement_Bug_Smokesim.blend](https://archive.blender.org/developer/F12819421/Noise_Movement_Bug_Smokesim.blend)

Philipp Oeser removed the

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Mantaflow: Noise doesn't follow smoke #81017