8
$\begingroup$

Will increasing the samples eliminate fireflies? I know that the question of reducing fireflies has been already asked, but i want a straight answer on whether it will help or not.

$\endgroup$
1
  • 3
    $\begingroup$ BlenderGuru has a nice article about how to get rid of fireflies here $\endgroup$
    – ranidspace
    Commented Jul 27, 2017 at 22:39

4 Answers 4

11
$\begingroup$

Just posting for empirical data:

enter image description here

These renders are at 2000 samples with a Sun Lamp pointed at an Icosphere with 100% glossy shader... just to demonstrate differences between firefly reduction techniques:

enter image description here

For noise reduction, see this answer: https://blender.stackexchange.com/a/84326/38953

$\endgroup$
2
  • 2
    $\begingroup$ Love the crystal ball example - that really shows the difference the number of samples makes. It also shows just how high it needs to be to produce nice caustics. $\endgroup$ Commented Jul 28, 2017 at 21:33
  • 2
    $\begingroup$ @RichSedman - Yes, it took my three GPUs 3.5 hours to render the million sample frame. $\endgroup$
    – bertmoog
    Commented Jul 28, 2017 at 22:00
5
$\begingroup$

Short answer: No.

Long Answer: As far as I know, not always; there is a point where more samples will do you no good for fireflies.

Increasing the number of samples will more often then not reduce noise in your scene, generally to a point where it is acceptable for the final result, at the expense of render time. The higher the sample count the less noise you should get.

Similarly, increasing samples will also more often then not reduce the amount of fireflies in your image. However there is an undefined point where increasing the amount of samples will no longer reduce the amount of fireflies.

Cycles being a physically based render will in theory converge to a physically correct and accurate representation of your scene.

Rays are cast, they bounce around the scene, and yield a result. More samples for the same pixel will average out the result and smooth any irregularities reducing noise and fireflies, the more samples are calculated the smoother the result.

However in some cases the scene converges poorly and the math averages slowly to a point where churning more and more samples will only converge to a same physically correct solution that still suffers from some bright spots (fireflies).

At this point more and more samples will no longer solve that problem and will only refine and confirm the already obtained result that includes some undesired bright spots, which are unacceptable from an artistic point of view, despite being accurate from a mathematical interpretation.

In this commentary by Brecht, the author of Cycles Render engine, the linked scene shows a common culprit, caustics, where he states that increasing samples will actually produce more fireflies.

Suggested solutions are changing seeds, clamping, post production, blurring samples, turning on Filter Glossy, noise reduction or any other of the well known methods will alleviate the problem.

$\endgroup$
4
  • 1
    $\begingroup$ I'm not convinced that there's a point where fireflies won't be resolved by more samples - artifacts from inacuracies in the calculation should not occur unless you're running to extremes... eg, very large scene far from the origin with very small details with small (or distant), bright light sources - generally more samples will resolve fireflies in all but the most extreme cases but may require an extremely large number of samples to become unnoticeable (which might be impractical). Are there any solid examples of more samples not resolving fireflies? $\endgroup$ Commented Jul 28, 2017 at 21:30
  • $\begingroup$ I have a faint recollection of reading something about it somewhere, written by Brecht Van Lommel himself, (if I recall correctly) the original creator of Cycles. Unfortunately in my admittedly superficial search I could not find the original again. I'll try searching a bit harder. $\endgroup$ Commented Jul 29, 2017 at 0:08
  • $\begingroup$ Maybe @brecht can barge in and demystify the issue. Please correct me If I'm wrong. $\endgroup$ Commented Jul 29, 2017 at 0:33
  • $\begingroup$ @RichSedman Late to the party, but found comment by Brecht about a scene that converges poorly where more samples will not reduce fireflies developer.blender.org/T52313#450602 $\endgroup$ Commented Sep 21, 2018 at 16:45
4
$\begingroup$

There are numerous ways of reducing fireflies and increasing the number of samples is the simplest, most generic, and also typically the most expensive (in CPU time). Usually you would want to try all other methods first and only resort to increasing the samples as a last resort.

$\endgroup$
0
-1
$\begingroup$

I am a beginner in Blender, and what I found to reduce noise has nothing to do with the number of samples, blurring, or clamp setting.

Below is an image with only 50 samples. You reduce noise by using mesh objects as emitters.

enter image description here

Here, I consider the 'size of the mesh emitter[4*4]' and 'strength of li8' which is 5, based on the room size, and I enclose the emitter with a diffuse object in order to reduce the bounces on the ceiling i.e to reduce the fireflies.

If you want the emitters on the ceiling, follow the above method with only 50 samples.

Another way to reduce fireflies is to move the emitter away from the ceiling and reduce the size of the emitter by double[2*2], with the same strength of li8 from above case which is 5. No enclosing is needed and only 50 samples.

If you want the emitters a little bit under the ceiling, follow the 2nd method.

Here, my uploaded image is based on the 1st method.

I hope my suggestion might help you.

If you have found out a good method to remove fireflies better than mine, please let me know.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .