Quality and denoising
All controls on this page are in Baker and affect the next bake. Sample counts improve the estimate of lighting; they do not add lightmap texels or probe positions.
BakerDenoising
Denoise when
Default: On Completion. Choices: None, On Completion.
On Completion filters the rendered lighting on the CPU after the rendering stage finishes. The bake is not finished simply because GPU rendering has stopped. None keeps the unfiltered result, which is useful for seeing whether noise originates in sampling or is being changed by the denoiser.
Denoiser
Default: Spectral Denoiser. Available when denoising is enabled.
| Choice | Purpose |
|---|---|
| Intel Open Image Denoise | Uses the included Intel denoising path to filter noisy lighting |
| Simple Firefly Remover | Targets isolated bright outliers rather than serving as the same general filter as the other two choices |
| Spectral Denoiser | Uses Spectral's denoising path and is the release default |
Compare the same bake settings when changing denoisers. A denoiser can suppress noise, but it cannot reconstruct a missing UV chart or lighting detail below the bake's resolution.
Denoise Each Chart Separately
Default: enabled.
Filters each UV chart separately instead of allowing unrelated packed charts to influence each other. A chart is one connected island in the lightmap UV layout. This helps prevent one island's lighting from bleeding into another and adds processing work.
Disabling it is a filtering tradeoff, not a way to fix chart padding. If seams remain, inspect the UVs and texel density as well as the filter.
Global Illumination
GI Samples
Default: 512. Accepted range: 32 to 65,536; slider up to 8,192.
Sets the number of lighting paths sampled per texel. More samples reduce random noise but increase bake time. This is the first sampling control to inspect when indirect illumination is grainy.
Higher samples do not resolve a blurry shadow caused by a low-resolution lightmap. First distinguish noise, which changes irregularly between pixels, from missing spatial detail.
Stationary Light Shadow Samples
Default: 128. Accepted range: 32 to 65,536; slider up to 8,192.
Controls sampling of the separately stored stationary-light shadow data. Increase it when that shadow data is noisy. It is independent of GI Samples, so increasing GI Samples alone is not a substitute.
Use Irradiance Caching
Default: enabled.
Reuses nearby indirect-lighting estimates instead of calculating every estimate independently. This can reduce work and noise for both lightmaps and probes, but fine changes in indirect lighting depend on the cache spacing and acceptance tests.
Disabling caching also disables the controls that depend on it. It is a useful comparison when diagnosing lost indirect detail, but can require more direct sampling work.
Irradiance Caching
These controls are available while Use Irradiance Caching is enabled.
Quality
Default: 128. Accepted range: 4 to 65,536; slider up to 8,192.
Samples used to form cache records. Higher values make the reused lighting estimates less noisy at greater bake cost. Increasing GI Samples and increasing cache Quality solve different parts of the calculation.
Size
Default: 32. Range: 4 to 1,024. Advanced control.
Controls the spatial size of the cache cells. Smaller values retain more local variation and require more cache work. Larger values reuse lighting across broader areas and can lose detail near small geometry.
This is not a lightmap texture resolution and is not the probe volume's Voxel Size. Changing it does not add probe positions.
probe volumeCorner Rejection
Default: 1. Range: 0 to 8. Advanced control.
Controls rejection of cache reuse near corners to reduce lighting leaking across geometry. Higher values make the corner rejection more restrictive. This trades reuse for a more local result; it does not repair open or incorrectly modeled geometry.
Aggressive Leak Prevention (Lightmaps Only)
Default: disabled. Advanced control.
Adds rejection of problematic cache cells inside geometry for the lightmap path. It can reduce leaking but may increase bake time or bright outliers. The source recommends pairing it with a cache Quality of at least 256 when needed.
It does not apply to probes. Do not use it as the explanation for a change in probe-volume overlap or density.
First Bounce Ray Guiding
Enable (Lightmaps Only)
Default: disabled. Requires irradiance caching.
Uses trial rays to learn which directions contribute useful first-bounce light, then guides sampling toward those directions. This can help when light enters a space through limited openings. It adds setup work and applies to lightmaps, not probes.
Trial Samples
Default: 128. Accepted range: 1 to 8,192; slider up to 1,024.
Number of trial samples used to learn the guiding distribution. Those trial samples are discarded rather than counted as the final lighting estimate. Raising this value improves the guidance estimate but does not directly add the same number of final GI samples.
Spectral Probes
Direct Lighting Contrast
Default: 0. Range: 0 to 1.
Controls how sharply directional direct lighting is represented in the baked probe data. Lower values reduce ringing around small, bright light sources. A value of 1 retains the original directional sharpness.
Ringing means unwanted bright or dark variation introduced by representing a sharp lighting signal with a limited directional basis. This is not the same as overall light brightness.
Indirect Lighting Contrast
Default: 1. Range: 0 to 1.
Controls directional contrast in indirect probe lighting. At 0 the result is a single average color. At 1 it retains the baked directionality. Lower values make the lighting more uniform around the shaded object, but remove directional detail.
Both contrast controls are baked into the probe result. Rebuild lighting to compare them.
System
Full Speed Workload Factor
Default: 32. Range: 1 to 1,024.
Controls how much path-tracing work is submitted in a GPU batch. It is a scheduling control, not a quality setting: it does not mean more samples in the finished result.
Larger batches may reduce overhead but keep the GPU busy for longer stretches. Reduce the factor if editor interaction becomes impractical during a bake. Raising it does not guarantee a faster bake on every GPU.
Noise calls for sampling or filtering changes. Lost surface detail calls for resolution or UV changes. Lighting leaks call for checking geometry, cache behavior, and probe placement. Raising every value together makes the cause harder to identify.