I have been researching on various techniques and methodologies to tackle the problem of “noise”. I am aware that most artists would not have the access to an industry grade t"render farm". There have been ingeniously elegant solutions to this predicament like compositing :https://www.youtube.com/watch?v=as_Wrpmb5BI . The principle is beyond adequate, yet the instruction is quite… rough.
One feature that caught my eyes was “caustics”, the pretty light patterns produced by transparent and translucent objects. However, Cycles is notorious for horrible caustics. This wonderful person found a way to “reduce noisy caustics.” I applaud him.
It was there that I discovered about combining “low sample images” with “variable seeds” into a “single image”. The render would be the same for a “brute render”, but the “noise would be significantly reduced”.
I have not seen a feature integrated in Blender where a frame would automatically have multiple samples at variable values of seeds and then combined.
I am aware that images can be combined manually. “But what if I want 1000 different seeds for one image?”“Or if I sought for each frame in an animation to be rendered in such fashion?”. Manual labor would be ineffective costly, and rather tedious.
“And I think that Blender, and Cycles will benefit greatly if this feature was directly integrated into the “render panel” itself.” Both in quality, and user friendliness.
Most of you would be aware of the effect that variable seeds for each frame has in animation.
I have attempted to contact the developers effectively on this to no avail. So I was wondering if you guys could help me push this feature to the next update.
Thank You For Reading fellow Blends ;D ,
Developers probably want to include more sophisticated methods to get rid of fireflies and noise for the set samples. In case of caustics, probably by also retaining the actual values so that the result can be adjusted in post processing.
From user perspective, most useful thing about image stacking is that one doesn’t lose already rendered time when a frame was already rendered but needs more. Could also use multiple computers to render the same frame. It still requires costly render time whether you render one frame with high samples or 1000 frames with low samples, plus the overhead as disk space/operations and from calculating the mean for many images.
Your post is hard to read and follow with all the decorations in everything. The title doesn’t need any decorations, the hell is that exclamation mark supposed to indicate anyway? The quotation marks around every other word and sentence makes it hard to read and annoying enough that I wouldn’t normally even go to the next paragraph before closing the tab. Then there is the formal signing with your nick which is just extra crap in a post because everyone already knows who posted, unless I misunderstood and your actual name is Anonymouqs.
The reason I bothered to read and answer was because the provided visuals, links and paragraphs which makes it possible to even want to read and try understand what you have to say.
You are inventing the bicycle here. When you render 1000 samples for example, you get the equivalent of 10 images with 100 samples combined because each batch of samples has different pattern anyway. If sample pattern were to become repeating, image would not improve during render any more and this would be very noticeable and bad.
I think you misunderstand what a seed actually does. An image does not need thousand different seeds, it needs one seed to have each sample have unique location (within technical limits). Seed initializes the randomizer so that two sample patterns will not be identical, this is why it is necessary for an animation. But for single image, each sample already is unique.
EDIT: read the link about stacking clamped pngs and I think I got what you meant by rendering same image with different seeds. You can forget the previous rant, it is not very relevant to your actual wish and use case