How to remove noise from video without losing the subject

The short answer

To remove video noise, identify whether it is sensor grain, color noise or compression damage, sample a short difficult clip, apply restrained spatial and temporal cleanup, and inspect fine texture during motion. Denoise before upscaling when noise would otherwise be enlarged, but avoid smoothing away real detail the upscaler needs.

Identify the noise pattern

Luminance noise looks like brightness grain. Chroma noise appears as colored speckles, often in shadows. Compression noise tends to gather around edges or pulse in blocks. A fixed sensor pattern stays in similar image locations, while scene texture moves with the subject.

Knowing the pattern helps you avoid treating freckles, film grain, fabric and animation texture as defects.

Usually denoise before upscaling

An upscaler sees noise as part of the source and may enlarge or reinterpret it. Mild cleanup first can give the scaling stage a more stable signal. The exception is a model with its own integrated restoration: extra denoising may remove texture that model could have used.

Balance spatial and temporal cleanup

Spatial denoising compares nearby pixels within a frame. Temporal denoising compares neighboring frames and can preserve detail when motion is estimated correctly. Too much spatial cleanup looks plastic; too much temporal cleanup creates trails or ghosting behind movement.

Test dark areas, faces and fast movement. If the background is clean but eyelashes, lettering or grain structure disappear, the treatment is too strong.

Add contrast or sharpening only after cleanup

Denoising can reduce microcontrast, but compensating with aggressive sharpening often creates halos. Use a small final adjustment, inspect the exported file, and accept some natural grain when removing it would cost more real detail than it saves.

Frequently asked questions