Hello guys.
I’m doing renders where I use the Shadow catcher in composite mode. but I’m struggeling to get the “shadow” in the alpha channel good without any noise.
It doesn’t really seem to do any difference how long I let it render either.
The denoising works well in the Beauty pass but not in the alpha channel.
Adding the image on the white background you can still see the noisy background.
My end goal is to get an image with a shadow for website viewing without having to do a separate pass and then composite that together since I will render 10.000 images or more.
Also make sure to save and compose your passes linearly! Failing to do so will result in different outcome from what you see in the VFB and it will also be noisier.
Alpha is not to be denoised by design, it tells where the render is opaque or transparent, nothing noise related.
What actually happens is shadow calculation, that is the thing be improved. You can try regular things like increasing/decreasing GAvsAA balance, use different solvers, tweaking shadowcatcher settings.
I cannot quite tell what exactly to do without the scene.
I decided to use 0.1 shadow to make the shadow a bit smaller/less pronounced. but I did try 1.0 for the shadow ammount and it made it worse. (attached image)
I tried to change GI/AA to lower which makes it clean up faster (I think) but also it made the noise spread out more.
a obvious thing it to use more passes. (lower the noise level) but the product itself is noiseless very early. only 10 passes. while I need almoast 60-70 passes to clean up the alpha map.
which means a lot of rendering time and double/triple the cost of rendering. considering my cost now is 3dollars and I will render thousands.
I also just want to add when you do the composoting mode its ver very noisy. Obviusly this black part is not used. but not sure if that has something to do with it.
I only use 5 lights (they exclude the shadow catcher) and one top light that is the shadow catcher illuminator.
no HDRI or anything.
I can think of some tricks to make it clean up faster, including post production denoising for alpha, but increasing GIvsAA balance should be the best bet for you for cleaning up the noise specifically in alpha. Here is an example of 2 passes rendered, GIvsAA balance of 16 (default) vs 64.
Of course this increases the render time. So if your limit is time based, you shall find a sweet spot for the settings, which cleans up the noise in alpha the best, as you say that main model is cleaned up rather fast.
Another parameter, which I did not expect to do much, but it did a quite bit, was increasing Light Samples multiplier. This comes with some penalty on the main model. Maybe this is the compromise you want to have.
