Help to remove noise
beachlegs
Posts: 516
Can someone please help I am sure this might have been asked already but it is my first time using Daz Studio and iray and although I am loving it, I am seeing alot of noise around the head and eyes of my image after rendering what can I do to resolve this?
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
What graphics card do you have?
Are you rendering an enclosed space, or a character lit by HDRI?
Does your characters skin use translucency?
Daz is super easy to get started with, but getting the most out of it requires work. If you tell us more about your setup, we can help you.
Edit: Also, images help us understand what's going on. Could you post an image of the problem?
LAMH Could be the problem here. Have you followed Alessandros tutorial on using LAMH with iray?
You probably need to up the Convergence Ratio, Iterations Limit, or Render Time limits in the Render Settings. They default to 95%, 2000 iterations, and 2 hours (I believe.) If the render stops too soon, you'll see noise in shadowed areas (and sometimes non-shadowed ones.)
Increase the Convergence Ratio to 98% or 99% (NOT 100%, that can cause issues) first. If it still finishes too quickly and with too much noise try doubling the Iterations Limit. If it's rendering for 2 hours then stopping, increase the Render Time limit (it's in seconds, so don't be surprised by the large number in there, say to double what it is at currently.)
If none of those fix the issue, you may need to adjust your Tone Mapping settings or increase your lighting. Iray doesn't like really dim scenes, and you'll get lots of noise (just like you will with a real camera.)
When you render, it quickly reads the "colors" of the images and roughly displays "Fast shadows". This is what gives you the initial render, which is often far from accurate. IRAY uses the "Convergence" value, to determine how "complete" the lighting within each pixel has been filled. Though you see one pixel, Daz/IRAY is actually drawing that one pixel with about 1.5 pixels worth of data. (That is the 100% of the actual pixel, and 50% of the 8 surrounding pixels, displayed as 100%) This is in addition to the "reality of light/shade/color", which is constantly being "tuned", as a ray nears the pixel. A ray doesn't have to hit the pixel directly, it can hit a surrounding pixel-area, and give that pixel an idea of what color it should be. But, until a random ray actually hits that pixel, it is never truly "complete". Thus, the use of the convergence value.
If you set it to render pixel-filter 1.0 quality, you will have horribly jagged edges on all objects. This is why they use 1.5 which is actually 1.1 pixel reach, in a radius or 1.5 pixels volume, to soften the edges and surface details.
I always suggest that people render 2x larger than they actually desire. Up to 8x larger, if you have the RAM. The reason being is simple...
1: Most art programs have a neat tool for "Despeckle", or "Soften contrasting pixels". This looks for single pixels that are out-of-place, because one color, usually, is shifted high, while others are low, relative to the surrounding pixels. (Eg, Noise... A bright red pixel surrounded by dark colored pixels, or green, or white.)
2: Once you despeckle the image, even if by hand, with a dab of a simple soften or "Clone colors" brush, you can then "soften" the image to reduce the non-contrasting noise, to a more gradient blend. (This obviously also "blurs" your image a slight bit, just like the quality value does. But that is why you make it larger.)
3: With the speckles gone, and the noise smoothed-out... Now you can reduce the image back to the size you originally wanted. If you rendered 2x, you reduce it 50% to get it back to normal. If you rendered it 8x, you reduce it to 12.5%, to return it to 100% of what you wanted. (Bonus if you did 8x, because beyond the normal "soften", you can do a full "blur", or "Gaussian blur", which gives better blending and noise reduction.)
Now, 8x larger renders... Won't that take 8x longer? (Actually, a 2x render takes about 3x longer. It is 2x2 of your original 1x1 size, so that is 4x larger, by volume, which is 3x bigger than the 1 you normally would have rendered.) 8x8 = 64, so it would take you 63x longer to render, if you waited for that 95% convergence. However, you don't need that long. 5% convergence is fine. It will be noisy as hell, but that is what the other steps above are for. There will be ZERO noise and ZERO stray pixels after that process.
I am attaching a sample set of renders. (Default settings, I just rotated it to get some good low-light noise in there, on purpose.) #2 and #3 reduced to 1000x700.
1: The original unedited picture rendered to default settings, with a focus-box showing the noise under the eyes, where it usually shows. Height output 1000, width 700. (Time rendered, 11:55 min {715 sec}, 5,047 MB used. Iterations: 425, @ 61% done.)
2: The same picture rendered 2x, despeckled and softened, then reduced and sharpened-up again, to restore any lost details to the "soften" filter. Same focus-box for a close-up. Height output 2000, width 1400. (Time rendered, 11:55 min {715 sec}, 5,255 MB used. Iterations: 105, @ 9% done.)
3: The same picture rendered 8x, despeckled, blurred, then reduced and sharpened. Same focus-box for a close-up. Height output 8000, width 5600. (Time rendered, 11:55 min {715 sec}, 9,425 MB used. Iterations: 6, @ 0% done.)
This also shows that #2 got only 1/4 as far, in iterations, and #3 was about 70x longer to render. (It takes longer to setup the larger image, which is why it is not exactly 64x longer time, seen in the memory use.)
NOTE: Due to #3 not getting to the reflection layer, which is done around Iteration: 16+, there is some "glow" missing off the pupil on the eyes and just inside of the hairline. However, there is clearly a sharper and more detailed, less noisy image. Seen in the lips, eyelashes, hair, ripples on the shirt, and skin pores. Plus, ZERO "speckles and noise", in the darkest shade, where they all manifest.
I would normally NOT render this fast. This was just for an example, to show how equal times being delivered, the "better" choice is 4x-8x larger renders, for quality, which then get reduced to your desired size. As opposed to trying to cram perfection into a size that is guaranteed to be imperfect, at its best.
P.S. Fourth image is the eye, of the 8000x5600 resolution image, before I edited it. (Tons of noise, but, like I said, not an issuse after removing it with despeckle and blurring it, then reducing it.)
Fifth image is the #3 image, but I ran it to 6% or 42 Iterations. Rendered in about an hour and four minutes. Much better, and what I would normally render for output. Even the original 8000x5600 looked photo-ready. (Minus the hair, which is still a hurdle for daz quality.) The colors are more "true" and the HDR shine finally showed in the eyes, after that 16+ iteration. These models are just so unnaturally glossy and waxy still. That is another thing that causes the stray pixel-noise. Light rays beaming off the fake bumps, into the shadows, off the eyeballs and ground, into the under-side of the eye and eyelids, and eyelashes.