Sharper, brighter, bigger

Every “enhance” button promises the same thing: the footage you have, looking like the footage you wanted. The honest version is narrower. A few specific defects can be reduced, in a specific order, and everything else is a trade.

Defect one: not enough light

Dark footage has two problems at once. The picture is dim, and the sensor compensated with gain, which brought noise. Lifting the brightness without touching the noise makes the noise brighter too, so the first pass must be a light denoise, then a tone curve, then a small amount of contrast to keep the blacks black.

Overdoing the lift is the classic mistake. Shadows that are pushed three stops turn grey and plastic. One stop, maybe a stop and a half, is the practical range.

Defect two: softness

Sharpness is local contrast at edges. An unsharp mask increases it by subtracting a blurred copy of the image from itself, which is why it also amplifies grain: grain is local contrast. The amount matters more than the radius — 0.6 to 1.0 is usually enough; 2.0 produces halos that look worse than the original softness.

Defect three: resolution

Upscaling is resampling, not magic. A Lanczos kernel reconstructs edges from the pixels that exist, which is genuinely better than a naive stretch, but no algorithm can recover detail that was never captured. Where upscaling does help is delivery: a 720p source in a 1080p timeline, or a platform that refuses anything below a threshold.

Defect four: compression

Low-bitrate video shows blocks in flat areas and mosquito noise around edges. A small amount of temporal denoising softens both; a large amount smears motion. If the source is very compressed, the better answer is often to keep the frame size and raise the output bitrate rather than to sharpen.

The order that works

Denoise, correct the tone, scale, sharpen, encode. Each step assumes the previous one has already happened: sharpening before scaling wastes detail on pixels that are about to be resampled, and denoising after sharpening fights the edges it just created.

Then compare. Put the original and the result side by side at 100% and ask which one you would rather watch. If the answer is the original, the preset was too strong — and that is a perfectly good result.