The best video denoiser for Mac

Footage shot in a dark room, on a phone, or on a camera from 2009 arrives speckled — and most tools that promise to clean it up quietly sand the detail off along with the speckle. Crisp denoises video and photos on your Mac, offline. This page explains what the process can and cannot recover, so you can tell a real improvement from a smoothed-over one.

Three different things people call “noise”

They have different causes and they do not all respond to the same treatment, which is the main reason one tool looks brilliant on someone else's clip and useless on yours.

• Luminance grain. The fine salt-and-pepper speckle from a sensor pushed to a high ISO in low light. This is the kind denoisers handle best.
• Chroma noise. Coloured blotches, usually red and green, sitting in the shadows. Often easier to remove than grain, because colour detail is naturally coarser than brightness detail and you lose less by smoothing it.
• Compression mush. Blocking and mosquito artefacts baked in by a low bitrate — from a messaging app, an old upload, or a cheap recorder. This is not noise at all: the detail was discarded before you got the file, so nothing can restore it. A denoiser can only make the blocks less obvious.

What denoising costs you

Noise and fine texture live in the same high-frequency part of the image. Any filter strong enough to remove one will take some of the other, so the honest question is never “does it remove noise” but “how much texture does it take to get there”.

The failure mode is easy to spot once you know it: faces go waxy, pores and stubble vanish, hair turns into a solid shape, and brick or fabric flattens into a smear. On a small preview this reads as clean. At full size it reads as plastic. Push the strength up until you see that happen, then come back down — the setting just below the plastic point is usually right.

Temporal beats spatial, when you have the frames

A spatial denoiser looks at one frame and guesses which speckles are noise. A temporal one compares several neighbouring frames, and since real detail stays put while noise moves randomly, it can separate them far more reliably. That is why the same clip often cleans up better as video than a single exported still does as a photo. It is also why fast motion is the hard case: when the subject moves a lot between frames there is less agreement to exploit, and temporal methods fall back toward spatial behaviour.

Picture noise and audio noise are separate jobs

Cleaning the image does nothing for hiss, hum, or wind roar, and vice versa. They come from different hardware and need different passes. Crisp does both — there are separate walkthroughs for picture noise, background hiss and wind — but treat them as two tasks. A clip that looks clean and still sounds like a hurricane will read as unfinished.

The free option worth trying first

ffmpeg ships two competent denoisers, hqdn3d and nlmeans, and they cost nothing. On mild grain from a decent camera they are often enough, and they run fast. They are worth ten minutes before you evaluate anything else, because if they solve your problem you are done. Where they struggle is heavy low-light noise and old tape, where the difference between a general-purpose filter and a model trained on damaged footage becomes obvious.

How to judge a denoiser on your own footage

Vendor comparison sliders are chosen to flatter, so test on your worst clip and look at four specific things.

• Skin at 100% zoom. Pores and fine lines should survive. If the face looks airbrushed, the setting is too strong.
• Repeating fine texture. Brick, grass, knitwear. These are the first things to dissolve.
• Shadows. Most noise lives there, and so does most banding once it is gone.
• Playback, not stills. Play it at speed. Some denoisers leave a faint crawling shimmer that a paused frame will never show you.

Where Crisp fits

Denoising in Crisp is free, runs on-device up to 4K, and can be combined with upscaling in the same pass — useful, because denoising before enlarging avoids magnifying the speckle. Free exports carry a small “Made with Crisp” mark; a one-time $99 removes it and adds batch processing plus the generative Max-quality lane for badly damaged footage. That lane is slow by nature — budget roughly 20 minutes of processing per second of video — so it suits a short clip you are rescuing rather than a whole reel. No subscription, and nothing leaves your Mac.