How to make a low-quality video look better

There is no enhance button, because "low quality" is at least six different faults with six different repairs. Naming yours takes thirty seconds and saves you from sharpening noise.

The footage looks bad and you want it to look good. Every tool offers to "enhance" it, usually with a sparkle icon, and the result is often differently bad — sharper, but crunchier; brighter, but grey.

That's because "low quality" isn't one fault. It's at least six, they have different repairs, and several of the repairs actively fight each other. Thirty seconds of diagnosis is worth more than any slider.

Name the actual problem first

Pause on a still frame, full-screen, and decide which of these you're looking at. It's usually two of them.

  • Soft — in focus, but detail is smeared. Low resolution, or a bad lens, or a heavy denoise pass upstream.
  • Blocky — flat mush in dark areas, halos around edges, gradients in bands. That's compression, and it's the most common one.
  • Noisy — grain crawling frame to frame, worst in shadows. Shot at high ISO in poor light.
  • Dark or flat — nothing wrong with the pixels, everything wrong with exposure and contrast.
  • Juddery — motion stutters. Low frame rate, or a frame-rate conversion done badly.
  • Shaky — handheld, no stabilisation.

Note which two you have, because the order of repairs depends on it. And note the one that isn't on the list: nothing here is a resolution problem. Which brings us to the trap.

Resolution is rarely the problem

The instinct is to reach for a 4K upscale, because "4K" is the word we've all learned means quality. But upscaling a compressed 720p clip to 4K gives you exactly the same visible detail in a much larger file — and often a worse-looking image, because the upscaler faithfully enlarges the compression artefacts along with everything else.

Upscaling reconstructs plausible detail, not lost detail. It's genuinely good at soft-but-clean footage: old camcorder tapes, small AI generations, screen recordings. It's bad at blocky footage, because it has no way to know which edges were in the scene and which the codec invented.

If you take one thing from this: fix the picture, then enlarge it. Never the reverse.

The order

Repairs interact. This sequence exists because each step wants a clean input from the one before.

1. Trim first. Ruthlessly. The single most effective quality improvement available is deleting the worst 20% of the footage. Nothing else on this list has that much impact for that little effort.

2. Exposure and contrast. Lift the shadows, set a real black point, pull the highlights back if they're clipping. Most "bad" footage is really just flat footage, and a grade does more for perceived quality than any AI pass. Do it before upscaling so the upscaler sees the corrected image.

Watch the shadows while you lift them — that's where the noise and the blocking live, and lifting the blacks is how you find out how much of it there is.

3. Colour. Fix the white balance, then decide on a look. If this shot has to sit next to others, matching them matters more than making this one pretty: mismatched shots read as amateur far more loudly than soft ones.

4. Now upscale2× or 4× — and only if the source is soft-but-clean rather than blocky. Check the result at 100%, not fitted to the window, and specifically check faces and text: those are where reconstruction goes uncanny first.

5. Frame interpolation, for judder only. Interpolating invents intermediate frames, which smooths stuttery motion and enables clean slow motion. Applied to footage that wasn't juddery, it just adds a soap-opera sheen and occasional warping around fast edges. Don't apply it prophylactically.

6. Sharpening: don't, mostly. Any sharpening you add gets multiplied by whatever the platform's encoder does. A good upscale has already handled the sharpening you needed. Manual sharpening on noisy footage sharpens the noise.

7. Export high, once. Every re-encode costs something. Render once, at the highest quality you'll tolerate the file size of, and let the upload be the only other compression pass. And check you're not rendering 1080p from 4K sources — soft exports are, more often than anything else, a resolution setting nobody looked at.

What genuinely cannot be recovered

Compression artefacts. The information isn't attenuated, it's gone — the codec threw it away and wrote something cheaper. Upscalers will happily enlarge the blocks. Grading will reveal them. Get a better source file if one exists: re-download at higher quality, or dig out the camera original. Ten minutes of looking beats an hour of processing.

Motion blur. A blurred frame has no sharp version hiding inside it.

Clipped highlights. Blown-out white is white. Pulling it down gives you grey.

Camera shake. Stabilisation is a capture-time fix or a specialist tool; no grade or upscale touches it. Sometimes the honest answer is to use less of the shot.

Focus. Out of focus stays out of focus. Every "deblur" that looks impressive in a demo is inventing a plausible face, not recovering yours.

The audio does more than the pixels

This is the part that gets skipped, and it's the biggest single lever on whether footage reads as "cheap".

People forgive a soft picture almost instantly. They do not forgive hollow, echoey, wildly-varying-volume audio — it reads as unprofessional in a way that soft video reads as merely old. Normalise the loudness, run the denoise pass at render time, and put something quiet underneath if the room is dead. Then watch it again and notice how much better the picture looks.

Next time, at capture

Most of this article is damage control for decisions made in the first five seconds of filming. In rough order of impact: more light (fixes noise, softness and colour at once), stop moving (a phone on any stable object beats handheld), get closer to the sound, film at the resolution you'll deliver, and don't zoom digitally — that's a crop pretending to be a lens.

The order that works

  1. Diagnose on a still frame: soft, blocky, noisy, flat, juddery, shaky.
  2. Trim the worst 20%.
  3. Exposure and contrast, then colour.
  4. Upscale — only if soft-but-clean.
  5. Interpolate — only if juddery.
  6. Skip sharpening.
  7. Fix the audio, which was probably the real problem.
  8. Export once, at full resolution.