uberzev
![]() |
Ugly part in red:
![]() My fix: ![]() There any logical reason for this issue to have lasted three versions? [snip] Nicer Stones.ffxml |
|||
Posted: July 5, 2012 2:05 am | ||||
Morgantao
![]() |
[QUOTE]There any logical reason for this issue to have lasted three versions?/QUOTE]
Nobody noticed? ![]() |
|||
Posted: July 6, 2012 4:28 am | ||||
uberzev
![]() |
![]() This problem may also seems to occur in 'Blocks' & 'Techno'. |
|||
Posted: July 6, 2012 4:50 am | ||||
uberzev
![]() |
Bump!
![]() |
|||
Posted: December 5, 2015 5:26 pm | ||||
Vladimir Golovin
Administrator |
This is a very old dilemma that goes back to when we designed the Noise components. The clipping is caused by our so-called 'amplitude boosting'. Noise functions, in their original form, produce an image with a fairly low dynamic range - i.e. low contrast gray, concentrating around 0.5 for Perlin and anywhere else for Worleys. When we released FF1.0 we wanted to increase the contrast of the output of all noises, so we did 'amplitude boosting' - we multiply the output of the original noise functions by specially selected values larger than 1, such as 1.27. Obviously, when we boost too much, clipping occurs, but if we try to reduce clipping, the boost multiplier becomes too low, to the point of being completely ineffectual. Finally, we decided to select the boost value for every component by running a lot of automated test renders, measuring the dynamic range, and selecting the boost value in such a way that clipping is below, uh, can't remember the exact threshold value, 5 percent of coverage or so. Obviously, we can't test-render all parameter space of the noise function, so one can still find regions of parameter space where clipping is more apparent that we would like. The problem is exacerbated by the fact that for many Worley-based noises we have to select not just the Boost value, but also an Offset value, to shift the output of the noise function into out 'visible part of the spectrum' between 0 and 1. These numbers are different for each Worley component and, within components, for each Formula (F1-F2 etc.) As for how to solve this properly, I still have no idea. Now that we have HDR output, an example of a solution would be to remove amplitude boosting completely, remove the Contrast parameter, and introduce Multiplier and Offset parameters that would allow manual control over the dynamic range of noises. Note that the output would be raw unbounded HDR grayscale values which cannot be used for interpolation between colors, so we'll also have to remove Color and Background parameters. But essentially, this "solution" would just shift the burden of choosing the correct Offset and Multiplier values from our shoulders to the shoulders of filter authors, while still leaving them vulnerable to clipping: they can't reasonably search for globally optimal values of Offset and Multplier in their case, so an occasional Variation change may still reveal clipping. TL;DR: A dilemma between technical correctness and everyday usefulness. |
|||
Posted: December 17, 2015 9:13 am |
Filter Forge has a thriving, vibrant, knowledgeable user community. Feel free to join us and have fun!
33,711 Registered Users
+18 new in 30 days!
153,531 Posts
+39 new in 30 days!
15,347 Topics
+72 new in year!
15 unregistered users.