GLOSSARY
Noise Reduction
The process of removing unwanted grain and sensor noise from footage — and the easiest way to accidentally turn a sharp image into a plastic-smudged mess if you push the threshold too far.
What noise reduction actually does
Noise reduction algorithms fall into two families: spatial and temporal. Spatial NR analyzes a single frame, identifying noise patterns by their frequency characteristics and removing them while preserving edges. Wavelet-based methods (used by Neat Video and DaVinci Resolve Studio's built-in NR) decompose the image into frequency bands and threshold each band independently — high-frequency bands get heavier suppression because that's where noise lives, but also where fine detail (hair, fabric weave, skin texture) lives. Push the threshold too high and you'll obliterate pore detail before you even notice. Temporal NR compares multiple frames — typically 3 to 7 — and averages out values that change randomly (noise) while preserving values that change predictably (actual motion). The temporal radius setting determines how many frames are sampled. A radius of 3 gives you decent noise suppression with minimal ghosting; a radius of 7 gives you cleaner results but introduces noticeable trailing artifacts on fast motion. FFT-based approaches (used in some Photoshop plugins and scientific imaging) work in the frequency domain directly, identifying periodic noise patterns like scan lines or sensor banding. They're less useful for random grain but excellent for removing fixed-pattern noise from specific camera sensors — the Canon C300 Mark III's faint horizontal banding at high ISOs, for example.
When noise reduction destroys your footage
The biggest mistake: cranking NR to compensate for bad exposure. If you underexposed by 3 stops on an A7S III at ISO 12800 and try to fix it in post with Neat Video at 80% temporal suppression, you'll get a waxy, texture-free image that looks like a bad TikTok beauty filter. No amount of NR recovers lost shadow detail — it just smears the noise into uniform gray blobs. The honest threshold: on well-exposed footage, you can typically apply 15–25% spatial NR and a temporal radius of 2–3 without visible degradation. Go past 40% spatial and you'll see skin losing texture at 100% zoom. Go past a temporal radius of 5 and you'll see ghosting on hair and fast hand gestures. ProTip: always apply NR before color grading, not after. NR operates on luminance and chrominance channels differently — if you grade first, the NR will see your creative contrast adjustments as detail to preserve and fight your grade. In DaVinci Resolve, place the Spatial NR and Temporal NR tools at the start of your node tree, before any curves or wheels.
Noise Reduction FAQ
Is Neat Video worth it over DaVinci Resolve's built-in NR?
For heavy cleanup, yes. Neat Video's auto-profile feature builds a noise fingerprint from a flat region of your frame, and its temporal NR is more aggressive with less ghosting than Resolve's built-in at equivalent settings. For light NR on well-exposed footage, Resolve Studio's built-in is sufficient and saves a render step.
Should I use spatial or temporal noise reduction first?
Apply temporal NR first — it's less destructive because it uses multi-frame data. Follow with light spatial NR to catch residual noise the temporal pass missed. Never reverse this order; spatial NR destroys the frame-to-frame variance data that temporal NR relies on.
Can I recover detail that noise reduction has already removed?
No. Once detail is smoothed away, the information is gone. You can fake it with sharpening or add synthetic grain to mask the plastic look, but you cannot recover actual texture. This is why you should always work from the original footage, not from NR-processed proxies.
Need Professional Help With This?
Salvaging noisy footage without destroying detail is a precision job. We've recovered underexposed shots from ISO 25600+ that looked unrecoverable.
Get a Free Consultation