Noise Reduction in DaVinci Resolve — Complete Settings Guide

The advanced Noise Reduction controls in DaVinci Resolve are found in the Motion Effects palette of the Color page and require DaVinci Resolve Studio. They are not available in the free version. Two independent engines — Temporal NR and Spatial NR — can be used separately or together depending on the shot.

The most common mistake is applying noise reduction in the wrong position in the node tree, or pushing Luma and Chroma thresholds without understanding what each one does. To use NR properly, you need to understand the parameters, their rendering cost, and how noise interacts with every other correction in your grade.

Temporal NR — Parameters

Temporal NR analyzes multiple consecutive frames to separate noise from real detail. Because it works across time, it is especially effective in static areas of the image, but less reliable where motion occurs.

Number of Frames
The number of frames DaVinci Resolve uses for temporal analysis. The range is 0 to 5. Higher values give a deeper analysis, but significantly increase render time. A value of 1 is often enough for shots with a lot of motion and helps limit the artifacts that higher values can introduce. (Reference Manual, p. 3472)

Motion Est. Type
Determines how the engine detects motion. Faster is lighter but less accurate. Better is more computationally expensive, but excludes motion more cleanly, reducing the risk of ghosting. None disables motion estimation — use it only on strictly static shots. (Reference Manual, p. 3472)

Motion Range
Three settings: Small, Medium and Large. This parameter defines the amount of motion the engine should account for. Small is suited to subtle motion, Medium to moderate motion, and Large to fast motion or subjects with significant motion blur. (Reference Manual, p. 3473)

Luma Threshold / Chroma Threshold
Both thresholds are linked by default: adjusting one affects the other. The range is 0–100. Start with both linked and increase gradually until you begin to see noise reduction in static areas. Once you have found a suitable Luma level, unlink the sliders. You can almost always increase the Chroma Threshold significantly more than the Luma Threshold without visible softening: chroma noise, meaning color speckling, usually tolerates stronger correction than luma noise. The manual explicitly states that the Chroma Threshold can often be raised higher than the Luma Threshold with less visible artifacting. (Reference Manual, p. 3473)

Motion Threshold
Defines the boundary between what the engine considers static and what it considers in motion. The default value is 50. Lower values exclude more areas from temporal processing. Higher values apply Temporal NR to more of the image. If you see ghosting on moving subjects, lower this value. (Reference Manual, p. 3473)

NR Blend
Blends the processed result with the original image. At 0, NR is fully applied. At 100, the image moves back toward the untreated version. This is useful when a setting works but feels too strong: instead of changing all thresholds, use NR Blend to reduce the overall effect. (Reference Manual, p. 3473)

Spatial NR — Parameters

Spatial NR works on a single frame. It reduces high-frequency noise while trying to preserve edges. It is useful in areas that Temporal NR cannot handle properly, especially moving subjects or residual fine grain.

Mode
Four options: Faster, Better, Enhanced and UltraNR.

• Faster — lighter, acceptable at low values, but can produce artifacts at high thresholds.
• Better — higher quality, heavier processing. Does not allow Luma and Chroma to be separated.
• Enhanced — better detail preservation at high thresholds. Allows separate Luma and Chroma control. More computationally expensive. (Reference Manual, pp. 3473–3474)
• UltraNR — AI-based mode trained on real noise patterns. It analyzes a flat area of the image to detect the noise profile and automatically adjust thresholds. It is especially useful on very noisy images. (Reference Manual, pp. 3474–3475)

Radius
Small, Medium or Large. A larger radius can improve analysis in detail-rich areas, but greatly increases processing cost. Medium works for most images. Move to Large only if details look unstable, aliased or too chunky at high threshold values. (Reference Manual, p. 3474)

Luma Threshold / Chroma Threshold
Same principle as Temporal NR: start linked, then unlink if needed. If the image becomes too soft but color speckling remains, lower Luma and increase Chroma. Enhanced mode is the best option for this separated Luma/Chroma workflow. (Reference Manual, p. 3474)

NR Blend
Same function as in Temporal NR: it reduces the overall strength of the treatment without changing individual thresholds. (Reference Manual, p. 3474)

Evaluating NR strength in Difference mode

Before locking your settings, enable Highlight in the viewer, then switch to Difference mode. The grainy texture you see represents what NR is removing from the original image. If you start to see object edges, silhouettes or recognizable structure in that pattern, NR is removing real detail. Lower the thresholds until you see only noise, without recognizable structure. (Colorist Guide, p. 391)

UltraNR — Procedure

1. Under Spatial NR, select UltraNR mode.
2. Click Analyze.
3. A bounding box appears over a flat area of the frame, used to extract the noise pattern.
4. Avoid repositioning this box unless you have a clear reason: the algorithm usually chooses the useful area better than the eye.
5. If the result is not enough, adjust the Luma and Chroma thresholds manually.
6. To check the analyzed area, enable Show Patch.
7. If you apply UltraNR to several clips, run Analyze on each clip individually: the analysis is shot-specific. (Colorist Guide, pp. 392–393)

Temporal first, then Spatial

The recommended order is Temporal NR first, then Spatial NR. Temporal NR usually preserves detail better in static areas. If you get good cleanup with Temporal NR, even partially, you will need less Spatial NR afterward, which means less overall softening. (Reference Manual, p. 3478)

NR node position in the node tree

The manual specifies that if you apply noise reduction and color corrections in the same node, NR is processed first, then color corrections are applied. (Reference Manual, p. 3479)

That means NR node placement has a direct impact on the result.

NR node after the INPUT CST, before primaries
This is generally the most coherent position in a scene-referred workflow. NR works on properly mapped RGB data, allowing UltraNR to analyze a more readable signal and detect the noise pattern more accurately. The result may slightly soften detail for downstream nodes, but this is often the most stable position. (Colorist Guide, pp. 393–394)

NR node at the end of the node tree
Edge detail may appear sharper because NR works on an already corrected image. But if the grade has strongly amplified noise — through exposure, contrast, saturation or white balance — NR may struggle more to separate noise from useful detail. (Colorist Guide, p. 394)

Position to avoid: before the INPUT CST
Placed before the input color space conversion, NR analyzes a log-encoded signal instead of properly transformed RGB data. With UltraNR, this can compromise noise pattern analysis and produce overly aggressive reduction, with visible smearing or trailing. (Colorist Guide, p. 394)

Think of it like a plumbing leak: if noise is amplified by an aggressive correction upstream, placing NR too far downstream treats the symptom instead of the cause. Test NR node placement depending on the shot, especially for:

• underexposed shots where Lift/Gamma/Gain recovery amplifies sensor noise;
• heavy contrast operations that reveal noise in shadows;
• white balance corrections that shift the color of the noise;
• saturation boosts that make chroma noise more visible.

The manual also describes a Splitter/Combiner approach: you can route a single channel, for example the Blue channel, through a dedicated NR node, then recombine. It is precise, but more complex. (Reference Manual, pp. 3478–3479)

Testing 4 positions without rebuilding the tree

The Nodal Tree Engine includes 4 preconfigured NR positions in the node tree, with each position containing a disabled group of 3 nodes:

• Sharp: used to visualize or emphasize grain in order to understand what will be treated.
• Temporal NR: dedicated only to temporal noise reduction.
• Spatial NR: dedicated only to spatial noise reduction.

The 4 preconfigured positions are: right after the INPUT CST, between contrast and saturation, after saturation, and at the end of the node tree in the Signature group. Enable the position you want to test, compare with Ctrl+D / Cmd+D, then disable it before trying the next one.

Slow motion, Optical Flow and NR interaction

If you work with slow motion — for example a clip shot at 50 fps slowed down to 25% — DaVinci Resolve can use Optical Flow or Speed Warp to synthesize intermediate frames. On shots with strong lines, such as shelves, fences, blinds or architectural elements, this can introduce visible artifacts.

In these cases:

1. Set Retime Process to Optical Flow or Speed Warp, depending on your Resolve version and available project options.
2. If artifacts appear on lines, test the Resize Filter set to Smoother in the Inspector, under Retime and Scaling.
3. Test at 50% speed first before going further.
4. If synthesized frames generate artifacts in the noise pattern itself, place the NR node after the full grade rather than before it. Temporal NR can create ghosting on interpolated frames because they are not real captures.
5. If the result is still not correct, try a higher-quality motion estimation mode, such as Enhanced Better or Standard Better, depending on the available options.

Cache, performance and export

Noise reduction, especially Temporal NR with more than one frame, is one of the most resource-intensive operations in DaVinci Resolve. (Reference Manual, p. 220)

The 4 cache levels in DaVinci Resolve:

1. Fusion Output Caching — useful for Fusion compositions or heavy codecs.
2. Node Caching — useful for intensive nodes, including NR.
3. Color Output Caching — caches the full node tree of a clip.
4. Sequence Caching — applies to transitions, titles and generators on the Edit timeline. (Colorist Guide, pp. 395–396)

Smart Cache vs User Cache:

• Smart Cache: Resolve automatically caches intensive elements. Recommended for most workflows.
• User Cache: you manually choose the clips or nodes to cache. Useful when you only want to target clips with NR. Procedure: right-click the relevant node, then Node Cache > On. (Colorist Guide, p. 400)

Recommended cache format
Use a high-quality cache format, ideally 444 or 4444 when the workflow requires it, to preserve image quality after NR. (Colorist Guide, p. 402)

Periodic cleanup
Use Playback > Delete Render Cache > Unused to free up space without deleting active caches. (Colorist Guide, p. 403)

Export
NR is rendered at full quality on export, regardless of timeline playback performance. If you deliver multiple versions, render the high-quality master first, then create web or proxy versions from that master to avoid running the NR pass multiple times.

Common mistakes

Linking Luma and Chroma too aggressively
Start linked, find the Luma floor, then unlink and push Chroma further. Noisy shots often contain more chroma noise than luma noise. If the Luma threshold is difficult to perceive, first identify the Chroma threshold, then lower Luma to compare the results.

Using 5 frames of Temporal NR on footage with a lot of motion
Start at 1. Move to 2 or 3 only if 1 is insufficient and the shot has limited motion. Five frames on moving footage can produce ghosting.

Evaluating NR while zoomed to 200%
Evaluate at actual output size on a proper display. What looks insufficient at 200% may be perfectly acceptable at 100%.

Applying NR in the same node as color corrections
NR is processed first inside the node. If you want NR to work on the already corrected image, place it in a separate downstream node.

Placing NR before the INPUT CST
On a log-encoded signal, UltraNR can misread the noise pattern and produce smearing. Generally place NR after the input color space conversion. (Colorist Guide, p. 394)

Disabling the NR node too early
If downstream nodes use HSL qualifiers, Power Windows or selections based on luminance or chrominance, the denoised signal directly affects the quality of those selections. Keep the NR node active and use cache to compensate. (Colorist Guide, p. 394)

Ignoring NR outside the Color page
NR is also available through Resolve FX > Revival > Noise Reduction, with similar parameters, but it still requires DaVinci Resolve Studio. (Colorist Guide, p. 393)

FAQ

Is Noise Reduction available in DaVinci Resolve free?

No. Advanced noise reduction on the Color page — Temporal NR and Spatial NR in the Motion Effects palette — requires DaVinci Resolve Studio. The free version does not provide access to these controls.

Limited workarounds exist in the free version, for example through Fusion or basic blur/denoise-style effects, but they are not the same tool, not the same quality, and not the same level of control as Temporal NR / Spatial NR in Resolve Studio.

What is the difference between Temporal NR and Spatial NR?

Temporal NR works across multiple frames and is most effective in static areas. Spatial NR works on a single frame and can reduce noise everywhere, including moving subjects. Using both together generally gives the best result. (Reference Manual, p. 3478)

Can I apply NR to only one color channel?

Yes. With Splitter/Combiner nodes, you can route a single channel through a dedicated NR node, then recombine. You can also limit processed components from the node options. (Reference Manual, pp. 3478–3479)

Should I disable the NR node after finding the right settings?

Not automatically. If downstream nodes use qualifiers or selections based on luminance or chrominance, the denoised signal directly affects the quality of those selections. Keep the node active and enable cache to manage performance. (Colorist Guide, p. 394)

Sources: DaVinci Resolve 20.3 Reference Manual, pp. 3472–3479 — DaVinci Resolve 20 Colorist Guide, pp. 391–394, 400, 402–403
Luca Giussani — Blackmagic Design Certified Trainer — lab.dcvisuals.fr