Page 1 of 1
Eliminate Noise an render time
Posted: Tue Feb 27, 2007 3:46 pm
No matter what I seem to adjust before or at render time - I seem to end up with very Noisy Images. have looked at a lot of the Flame renders on the Deviant site and all of them look equally noisy. Any way to get nice crisp noiseless renders?
Check out my Nemo's Wheel image and you'll get some idea of what I'm talking about - Looks like rust eating inti the flame
Posted: Tue Feb 27, 2007 5:27 pm
You can also increase the anti-aliasing factor, but this will significantly increase rendering time (since 2:1 renders something with four times as much detail, 4:1 takes sixteen times, and if you're nuts enough, 8:1 would take sixty four times as much). Not only does this increase the time spent rendering, it also will thrash your memory as you go for higher resolution results.
It would be interesting to see how long it takes to render (as well as the memory footprint required) one of the "higher quality" Apo images - random searching often shows comments about how they take 3 hours for even smaller ones (and someone saying how one of his images took 13 days!!!), so maybe one shouldn't be too concerned if it takes 2 hours to get a decent result...
At some point I hope to add XGrid support to allow you to spread this misery across your local network, but that only helps you if you've got a render farm handy (or at least two or three decent machines).
You can also switch to another rendering algorithm (found down near the bottom of the global panel) - something like ZRGB will often produce better results with fewer iterations (and thus be far less painful with higher antialiasing), not to mention that its saturation/gamma model is a bit more "tweakable", though I find the colors to sometimes be a bit garish and harder to work with at higher resolutions.
I've also been investigating other rendering strategies and techniques, but nothing significantly better yet.
Posted: Tue Feb 27, 2007 9:06 pm
where do I set the a/a render values for still images - I found them aright for movies but I don't want to make a Movie?
Posted: Tue Feb 27, 2007 9:11 pm
never mind - found it.....
Posted: Tue Feb 27, 2007 10:11 pm
Maybe I'm not sure that noise is the correct term butafter another hour or two making adjustments etc I still have the same problem - so I have included an image circled with the problem I am referring to - just cant get it smooth... The rest of the tips were great by the way
This is a re-rendering in Flame of the original created with Apo at 2:1 - at 4:1 the problem is noticably worse
Posted: Wed Feb 28, 2007 11:52 am
The thing to be aware of is that images like this create those "spheres" by performing a random blur of the point - this is done by literally generating a random offset each time (which, on the average, is centered on the original point). So those areas aren't actually solid - they are just a random Gaussian blur (which only looks solidish when a whole lot of points are plotted). Since each iteration plots one extra point of the sphere like shape, and since there are hundreds of spheres (some larger and many smaller) you can see how it will take whole lot of iterations to make it appear more solid. To make matters worse, the different spheres have different probabilities of being touched (which is not directly related to their size, either, so it's not like the large spheres will get more points and smaller ones less).
So what you'll need to do is increase the number of iterations, and decrease the detail and density sliders to compensate (otherwise the "higher density" areas will get "blown out"). This will, of course, take longer, but that's how the game works...
You can also play around with the weights of each of the flames to see if you can shift the "emphasis" to those spheres you want more points to land on.
Posted: Wed Feb 28, 2007 12:30 pm
Thanks - I'll give it a try and just focus on these few. Lastnight I spent hours playing with each flame one by one and no matter what I did with iterations, density and detail and still it wouln't clean up withthe exception of rge centre sphere.
Just wondering why it has to be so complicated with Flame. Render time is pretty quick but I am not heppy with this as a final render (took around 12 minutes).
Apo on the other hand takes a fraction of the time to create the same image and renders perfectly without all that distortion and is way less complicated to understand and to use. I seriously think this program is more gearerdt toward physicists and applied mathematicians that the average artist, user or even programmer for that matter (if you will forgive me for saying do).
- Far too many switches and complicated as hell to figure out what they all do
The C&P tip for the transformations was very handy thanks - makes far more sense if you can lay out the coordinates geometrically rather than hit and miss with that odometer thingy!!!!
However where it does copy the Julia(n) value it does not copy the distance and power values etc....
The app still reminds me of that PC game Return to Mysterious Island than any kind ofFractal or graphics program I've seen to date. Seem to spend mopre time solving puzzles that getting close to a decent clean render.
But I have not given up hope yet!
Posted: Thu Mar 01, 2007 11:12 am
Zooreka wrote:However where it does copy the Julia(n) value it does not copy the distance and power values etc....
The currently released version only copies a single row at a time - the next release will include the ability to copy an entire "group" (as well as drag & drop the entire group) by holding down the shift key
Posted: Thu Mar 01, 2007 3:30 pm
(n) couraging. I tried all you suggested (I really only needed to try and affect Affine #2) - I played for a few hours and it seems the only way to get rid of these artefacts at rendertime is to plunge the flame into almost total darkness.
Even the example pic above is way darker than the original render with apo and apo's is as smooth as a baby' b*** in comparison. If it is at all possible to get a smooth,non-artefacted and and the same time decently lit Julian from Flame I sure have not been able to figure it out so far, even with all the advice given.
I had apo - 1 tutorial and half an hour or so and produced the original from that with one or two tweaks here and there.
Why is Flame taking me forever to even figure out the basics?
Update: Maybe you can check out my deviant gallery. I have 2 julia(n)'s side by side that are very similar. One created with Apo and the other with Flame using more or less the same parameters. No matter how much tweaking I do the lovely material like rendering you get on the afine 2 layer with Apo I simply cannot reproduce in Flame. I got a really nice Julian out of it nevertheless but I would like to see less darkness and more detail without all the fuzz. I had some fun trying to eliminate the same fuzz from the Plastic scripts images on there two - they're not 100% smooth but I can live with what I got for now
Posted: Fri Mar 02, 2007 6:40 pm
So Apo and flame will never generate the same image - they've got some deep differences in the way certain things are handled (in general, flame tries to give decent results faster, as opposed to better results significantly slower).
That being said, there are two more tips to play with - first, go through each flame and play with the size parameter - this will increase the "size" of each dot. Then add some blur to the result - doesn't have to be much (maybe +0.3 to +0.5) but that will help take some of that extra "bits" from the larger size and spread them around a little (and when rendering at a higher resolution, it won't end up being blurry).
Ultimately, however, you need to increase the number of iterations - to get data to randomly hit all those spots, you need to have more of them. However, the results of this become a bit trickier to adjust. To remedy this, the next beta (which I'll send you when its ready) includes an "HDR" rendering algorithm, designed for this higher dynamic range. The results that I've seen from playing with it are encouraging, though it won't be able to support the full transfer mode model that RGB rendering does (since standard composition algorithms works in a bounded RGB space, as opposed to an HDR one).
Posted: Sat Mar 03, 2007 2:53 pm
That would rock! Or maybe the possibiliy to export to Apo's format and I could use it as a mini render-farm type utility on the Flame 3 renderer. I hate windows but don'tmind rendering on iteven if itcrashes the desktop 9 times out 0f 10.
That said I love the quick renders that QF produces - particularly if they are using some of the plastic scripts. I got a result last nightthatalmost blew me away - even leaving in a bit of the black 'noise'- added atmosphere to the overall result. The overall product disn't even lool plastic - using the plastic bubble script and a few tweaks here and there. However I have other issues in that some of the scripts seem very buggy, don't run or don't execute properly! Make bubble Spiral immediately throws an error for one!
Also the export alpha was not what I had in mind when I mentioned rendering to transparency though this is useful when it works. I want to export the rgb image to a transparent background in png format and no can do! Also when rendering if I do not change the format from pnf to another and then pack to png the renderer keeps defaulting to tiff. I'd like just to be able to select png and hit save and see the save dialog say saving to ***.png and not ***.tiff
Overall I think Flame is a good product but still needs quite a bit of development - I may never get comfortable with the F form as opposed to the triangle though.
Another problem I have is that running the convert to final script leaves me with nothing more than a black image. It also seems this can only be applied to a single flame as opposed to multiple flames as in the case of Apo.
If you import an apo flame into QFlame however QFlame will recognize it for what it is and not leave you with a black image.
Gotta go to Peru nextweek but am hopeful I can spend some time with Quadrium and supply feedback on that in the meanwhile.