Hi everyone,

I’m running into a severe noise issue in Karma, related to using emissive fire / candle flame as a light source.

First, a quick clarification about my background: I’m still learning and I’m not a professional TD, so it’s possible I’m missing some important setup details. That said, I tried to test this as carefully as possible and compare different scenarios in a controlled way.

Key clarification

SSS (and the render overall) behaves normally when the scene is lit only by standard light sources (area / point / dome lights).
However, as soon as a candle flame or emissive fire actually contributes light to the scene, the render quality degrades significantly.
This affects not only AOVs, but the Beauty pass as well.

What exactly happens

When the flame:
is visible but does NOT emit light → everything looks clean and stable
does emit light (emissive flame / fire lighting the scene) → severe noise appears
The following become extremely noisy:

Beauty

SSS AOV

Glossy AOV

Reflection AOV

The noise does not converge, even with very high sample counts.

Setup

Houdini 21.0.440
Karma CPU and XPU
ACEScg
Resolution: 1920×1080
Scene: candle with wax (SSS) + fire / candle flame

Tested with:
Karma GPU (XPU)
Karma CPU

Render comparisons attached

I’m attaching the following renders for direct comparison:

Beauty render at 512 GPU samples, lit by standard light sources only
→ normal result, SSS behaves as expected

Beauty render at 2048 GPU samples, where the candle flame is the only light source
→ extremely noisy result (Beauty, SSS, Glossy, Reflection)

This suggests the issue is specifically related to emissive fire lighting, not overall sample count.

Observed behavior

SSS looks correct when lit by regular lights
Once fire / emissive flame lights the scene:
Beauty becomes heavily noisy
SSS / Glossy / Reflection AOVs become almost unusable
Increasing samples does not meaningfully improve convergence
Same behavior on both CPU and XPU

What I already tried

Very high Path Traced Samples (CPU & XPU)
Increasing Diffuse / Reflection / Refraction / SSS limits
Treating volumes as light sources
Different pixel filters
Different emissive intensities
Tests with and without depth AOVs
CPU vs XPU comparison
None of these approaches significantly improved the noise.

Questions

Is heavy noise in Beauty and SSS expected behavior when using emissive fire or candle flames as light sources in Karma?
Are emissive volumes sampled differently compared to standard lights?
Is there a recommended workflow for SSS materials lit by emissive fire?
Is denoising considered mandatory for this type of setup?
Are there known limitations or issues in Karma 21 related to emissive fire lighting?

Any insight would be greatly appreciated.

Render tests are available via the Google Drive link [drive.google.com]

MikhalovMaksym
Is heavy noise in Beauty and SSS expected behavior when using emissive fire or candle flames as light sources in Karma?

MikhalovMaksym
Are emissive volumes sampled differently compared to standard lights?

Yes, it is expected to be noisier when illuminated just from emission

Since emissive volumes and surfaces illuminate scene just when randomly hit by indirect samples

MikhalovMaksym
Is there a recommended workflow for SSS materials lit by emissive fire?

If you are using 21 Id suggest using Geometry Light to turn your fire volume into a properly sampled light

Thanks a lot for the explanation, that makes sense now.
I didn’t fully realize that emissive volumes are only contributing via indirect hits, which explains the noise I’m seeing.
I really appreciate the clarification.
I’ll try the Geometry Light approach you suggested and will report back after some tests.

tamte

MikhalovMaksym
Is heavy noise in Beauty and SSS expected behavior when using emissive fire or candle flames as light sources in Karma?

MikhalovMaksym
Are emissive volumes sampled differently compared to standard lights?

Yes, it is expected to be noisier when illuminated just from emission

Since emissive volumes and surfaces illuminate scene just when randomly hit by indirect samples

MikhalovMaksym
Is there a recommended workflow for SSS materials lit by emissive fire?

If you are using 21 Id suggest using Geometry Light to turn your fire volume into a properly sampled light

Thanks for the suggestion.
I tried using a Geometry Light driven by the fire volume as you described. While it definitely helps compared to pure emissive lighting, the SSS component is still very noisy in my case, especially with a small candle-scale source.

After some testing, I ended up using a different approach: I keep the fire purely visual and use a regular Area Light as the actual light source for the scene. To preserve the natural flickering and gradual fading of the flame, I drive the Area Light intensity from the fire simulation itself (for example using reduced temperature values), so the light still flickers and dies out exactly like the fire.

This gives me much more stable SSS while keeping the temporal behavior of the flame believable.

One question still remains open for me though: the Refraction pass appears to be very noisy regardless of lighting setup or sampling settings. Even with stable lighting and higher samples, refraction seems to remain noisy. I’m not sure if this is expected behavior in this type of setup or if there’s something specific I might be missing there.

I eventually solved the noisy gloss issue with SSS, but honestly — it took a lot more time and experiments than I expected.

I’m sharing this mainly for anyone who might run into the same problem.

What I found:

Gloss becomes very noisy when combined with SSS, especially if there are volumes (fire/smoke) and an Area Light in the scene.

Simply increasing samples doesn’t really fix it — the noise is not random, it’s structural.

What finally worked for me:

I render the fire in a separate pass, without any SSS geometry in the scene.

SSS is rendered separately (matte).

I use the same Area Light, but:

drive its intensity/exposure from a SOP-based driver

read a detail attribute from the simulation

feed that value directly into the light parameters

The light reacts to the fire animation, but the volume itself is not present in the SSS render.

Result:

gloss noise is gone

SSS becomes stable

the lighting still feels alive

no need for extreme sample counts or unreasonable render times

From my experience, the main issue is the combination of SSS + glossy + volume, not just “insufficient sampling”.
After going through many tests, it became clear that proper pass separation solves more than endlessly tweaking quality settings.