Noise reduction
Simulator
AI generated demo imageYour photo stays on your device. This page opens it in your browser and works on it there.
Higher ISO for a darker scene: the same exposure brightness, with more noise. The levels are roughly those of a current full-frame camera, seen across the whole frame.
How noise reduction works
Noise is random variation the sensor adds to every frame, worse in shadows and at high ISO. Noise reduction tries to separate it from real detail. It can look at neighbouring pixels in one frame, at the same pixel across several frames, or both, and each approach fails in its own way.
Luma and chroma noise
Luma noise is grain in brightness; chroma noise is blotchy speckles of colour. Chroma noise can be removed heavily with little visible cost, because eyes see colour detail poorly. Luma noise is harder to remove without softening the picture.
Spatial noise reduction
Averages each pixel with similar neighbours, keeping strong edges. Gentle settings clean flat areas; strong settings also average away skin pores, foliage and fabric, leaving the smooth, painted "waxy" look of many phone videos.
Noise-aware smoothing
Better tools first measure how strong the noise is at each brightness, then smooth only differences smaller than that. Texture that stands out from the noise is left alone, so rock, foliage and skin survive much stronger settings.
Temporal noise reduction
Averages the same pixel over several frames. Noise is different every frame, so it cancels, while static detail stays fully sharp. Averaging 4 frames halves the noise.
Ghosting
When something moves, simple frame averaging blends its old positions into the new frame, leaving smeared trails. Motion-adaptive and motion-compensated noise reduction detect movement and average less there, so moving objects stay clean but noisier.
In camera or in post
In-camera noise reduction is baked into the recording and works on compressed data. Noise reduction in post, such as DaVinci Resolve's, is slower but far more controllable, so many cinematographers keep camera NR low or off.
Leave some grain
Some noise reads as texture and helps hide banding and compression. Cleaning to zero often looks more artificial than a fine, even grain, so most tools can blend a little of the original back in.