# Why does render resolution affect output quality?

The two images below are sections of the same image rendered at different resolutions.

• The blurry one to the left is rendered at 600x350 and then cropped to the section shown here. So there is no scaling or other processing of this image except cropping for the sake of clarity.
• The sharp one to the right is rendered at exactly the same settings except for the higher resolution of 1200x700 then down-scaled and cropped to the same size as the first one.

I'm fairly new to blender and 3D-rendering in general so do not understand why the first one is so blurry when the scaled one sharp. Can this be fixed and in that case how? I thought that the imgages would be rendered as sharp as possible for the given resolution. But in this case that is obviously not the case.

Is there some switch or slider or property that should be adjusted to get it right? Or is this behaviour to be expected?

I'm running blender 2.80.42.

• What did you use to downscale your image? For example Photoshop by default uses bicubic algorithm that applies sharp (as Martin mentioned) – Sergey Kritskiy Feb 20 at 11:08
• I used [imagemagick.org/] for the cropping an scaling as: convert src.png -resize 600x out.png – UlfR Feb 20 at 13:18

There are 2 reasons:

• the down-scaling can have a sharpening effect on its own - depends on the algorithm used
• the higher resolution image contains more traced samples per the same area = less noise. The less noise the less blurry image. Let's say single pixel in original size received 100 samples, the same area in double the resolution would receive 400 samples.

So the apples-to-apples comparison should be rendering the original resolution with 4x the sample count and down-scaling the higher resolution with a method that does not add any sharpening.

This behaviour is what I would have expected. If you render in higher resolution, you naturally get more information into the image. Scaling it down then makes it look sharper. Google "bicubic sharper".

The edges in both renders are not 100% sharp, they are always a gradient. Lets say the edge between the black part and the background is a gradient in the size of 4 pixels in both renders. When you scale the bigger one down to 50%, the gradient is also scaled down - now it has only a size of 2 pixels, which makes it look sharper compared to the smaller render. the interpolation adds some sharpnes too.

Edit: just found another thread on the same issue: Why does increasing resolution and then downsampling to desired size increase render quality?

• I'm not sure I understand or accept your answer. With this reasoning the image would appear sharper the smaller you make it, and that is not the case. If you scale it down infinitly to 0x0 or 1x1 it will not be infinitly sharp I think. The act of scaling down an image is actually removing information from it and by doing that it seems counter intuitive to get a sharper result. Please explain this a bit more... – UlfR Feb 20 at 9:30
• This is hard to explain... Sharpness means local contrast. In the example I gave, the scaled down gradient has only a size of 2 pixels - one being black, the other being white. This is the maximum sharpnes you can get. – Martin Ackerl Feb 20 at 10:31
• But maybe the effect you described comes more from the fact that your bigger render has just more information (total samples). Scaling down a bigger render with a higher total number of samples has the same effect as rendering a small image with a higher sample count. Example: picture size 200x200 pixel, 10 samples per pixel: 200x200x10=400000 samples in total ---- picture size 400x400 pixel, 10 spp: 400x400x10=1600000 samples in total. – Martin Ackerl Feb 20 at 10:41
• So if you render your smaller image with the same number of samples (40 spp), you should get more or less the same result (without the sharpening effect of the scaling) The math would be: 200x200x40=1600000 samples in total – Martin Ackerl Feb 20 at 10:41

The naive answer, ignoring clever downsampling algorithms and assuming simple averaging, is that your second image is not 'sharper'. But its colors are more accurate, so the contrasts over the boundaries between regions of color are higher.

The more point-samples that are taken from the original domain, (let's assume that has infinite resolution,) the more accurately each pixel will represent the average color in the area of the original domain that it covers.

To reduce the case to the absurd, suppose only one sample could be taken for the entire area represented by your cropped section. Then it would all be one color, no matter what the resolution of your final image. As 'blurry' as you can get. Now imagine 2,4,8.. and so on...