Cycles Denioise Issue on Animation Sequence

I actually did not notice this at first. Then I started rendering some sequences.

Has anyone done any testing with animation?

I am not sure how to explain this clearly so I will just link to the video:


Showing the botchy artifacts. They look like a post filter over the image which is fine. But it does not seem to be doing any interpolation on a frame by frame basis. And this makes the blotchy areas appear to be a film over the top of the image that does not move.

Maybe there is a setting I am not using correctly. But this is not usable for animation as it is.

Even if I pump up the samples which is clearly the issue here. This lover sample rate (32) shows the issue of the denoise filter as a post effect.

As long as it does what it is doing it is not usable for animation even at higher sample rates. It will always just look like a static filter effect that does not appear to be moving. From what I can only guess.

Maybe someone has examples at a higher sample rate?

I will try another test as well with less frames at a higher sample rate.

To quote from the manual:
“For animation denoising can be used, however it still requires high sample counts for good results. With low sample counts low frequency (blurry) noise can be visible in animation, even if it is not immediately apparent in still images.”

Yes. I get that part. That is not the point. Did you read and understand my post?

There is a difference between blurry noise and static blurry noise. Do you understand the difference?

Could the static blurry noise be related to having a fixed seed? Does animating that help if it is fixed?

Did you animate the noise seed? If your noise was static - the denoised blotch artifacts will also be since the denoising algorithm will be denoising the same noise pattern from frame to frame.

Yep. I think you hit the nail on the head. I did not even realize Cycles was not doing this by default. I thought it was related to denoising. Handy little button to turn on random noise.

Rendering again now. But I already checked a few frames, and I am sure this is it.

Thanks! :slight_smile:

By the way, anyone want a clear demonstration of this you can go here:

Old tutorial here:

Explains why this happens.

Now there is a button right next to Seed to turn this on automatically. I suppose if you wanted you could play with setting up your own custom seed animation the old way. But it seems like the new bottom does the trick.

To me you are still using little sample rate for what you need with denoiser in animation, you should increase more samples. If you do not, you will get patches no matter you use static or random noise pattern (the difference will be that you will get static or moving patches/noise depending on what you choose). In another thread we have concluded that for using denoiser for animation might be better to use static noise as long as sample rates are high enough. For example, you assume that without denoiser you need 10000 samples to get a clean image. Then using denoiser you do not go much less than 8000 or 6000 samples. This is a good time saver anyway.
In short, denoiser is to clean residual noise.

Absolutely. But for this thread I was only concerned about identifying animated or static seed. I come from other render packages where animated is the default. So my experience is limited to those apps. And I just did not realize Cycles was static by default. In fact even though I have rendered in cycles before, for some reason I just did not notice it. So it really confused me. I assumed that it was the denoiser!

Higher sample rates are a given. But it is not the issue I was addressing.

By the way I would not recommend static for animations. Doing high sample rates might “hide” it. But really, technically, this is not what you want for animation. It will come back to haunt you later. Guaranteed. In post or compositing or especially if you get it on a large screen. If there is any speed increase at all it would not be worth it.

Trust me on this.

At any rate. As far as this issue is concerned. It is solved.

Thanks for the help!