The question may sound a bit strange, but I was a bit surprised by some fact (Cycles). If I render a small pic in resolution bigger than it has to be, and then scale it back in Gimp, the resulted picture looks much better than if I rendered it in requred resolution. Is this a normal thing?
Thanks alot. So it’s the normal method to make small pics looking more like bigger ones? I mean is there any automatic option in Cycles to reproduce this effect for small pics?
Render with more samples. The result will be similar, but not necessarily identical as downsampling involves also the resampling filter that can sharpen (or soften) the image in addition to simple supersampling.
Thank you kesonmis! I got it.
Guys I got another problem after scaling big image back, all edges on small image became looking hackly, jagged. Is there a way to make small pic without jaggies? Thanks alot for any help
Here is how the jaggies look like
Attachments
Can’t see any jaggies, only pixels. If you want smoother edges and more detail use bigger resolution (do not downscale).
This screenshot was taken from render 420x234, 1000%. So actually 4200x2340. 5500 samples. Pixel width 1.35. It lasts about 14 hours. Do you really think I should increase it even more? Then, I think I can’t avoid downscaling as I should get in result pic 420x234, foe web. Really don’t know how to solve it.
An image that small (420x234 is small) there is nothing you can do to avoid low detail due to low pixel count. It is not vector graphics that you can zoom in and not lose detail. With raster images, if you zoom above 1:1 pixel mapping from monitor, you will ALWAYS see blocking due to pixel width being the smallest possible unit of detail. If you have to see this image bigger than 420x234 pixels is at 100% zoom, use bigger resolution.
An image that small (420x234 is small) there is nothing you can do to avoid low detail due to low pixel count. It is not vector graphics that you can zoom in and not lose detail. With raster images, if you zoom above 1:1 pixel mapping from monitor, you will ALWAYS see blocking due to pixel width being the smallest possible unit of detail. If you have to see this image bigger than 420x234 pixels is at 100% zoom, use bigger resolution.
The image you sent me has as much detail as such a small image can possibly have. The size of it before downscaling is of little importance in this matter. If you NEED more quality, do NOT downscale so much. You can not have high detail and small resolution at the same time with raster graphics, these two things are opposites of each other.
Thank you very much for the detailed explanation. I’ve understood. The reason why I had some doubts - I saw a render in exactly the same small resolution, but it seemed to me not as blocked as mine.
Also, when downscaling, I (IMO) am pretty sure it’s always better to do it by a factor of 2. So that means render at 2x, 4x or 8x the resolution, not 10x in this case. Try your render at 800%
Why do you think all the anti aliasing settings for games go 1, 2, 4, 8, 16x AA?
Generally when downscaling an image, I use a bicubic sharper (in photoshop) which generally preserves more details.
Thank you so much for the hints! I’ll follow them. Could you please also recommend Pixel width value for this task, or 1.35 which I use is okay? Thank you in advance.
Pixels do no have width unless you set them relative to some real world measure for printing etc. When output is on screen, one pixel is one pixel wide. How wide is it on some monitors screen plane (in millimeters for example) depends on monitor size and resolution.
Maybe you mean pixel aspect ratio? The relation between width and height of pixel on non-square pixel devices?
Sorry for the mess I meant this Pixel filter parameter:
Attachments
If your downscaling, don’t put it too small. The smaller the value, the sharper your image is, but the more jagged an edge can appear.
If I recall correctly, that looks at how wide a samples contribution is in relation to pixels.
Thank you. I’ve already done render with 1.35 and will do another one with 1.5. Then I compare the results.