Learn
Short, plain explanations of what is actually happening to your footage. No jargon for its own sake, and no assumed background.
Start with the order of operations
If you only take one thing from this section, take this: the order you apply fixes in matters more than which tool you use. Restoration steps interact, and several of them are destructive to the ones that follow. Doing them in the wrong sequence produces results that no amount of re-processing will recover.
The order that works is deinterlace, then denoise, then upscale, then colour.
Deinterlace first. Old footage often stores each frame as two half-pictures captured a moment apart, which shows up as horizontal comb teeth on anything moving. An upscaler has no concept of this, so it reads the comb as fine detail and sharpens it into something far harder to remove. What interlacing is covers how to recognise it.
Denoise second. Noise and compression blocking are, to an upscaler, texture worth reconstructing. Magnify them first and you have committed to keeping them. Clean footage also compresses far better, which matters if the clip is headed for a platform that will re-encode it.
Upscale third, once the frame is clean and progressive. This is the step that reconstructs detail rather than merely enlarging pixels, and it works best on a source that is not fighting it.
Colour last. Grading decisions depend on what the picture looks like after everything else, and a grade applied to a noisy or combed frame will need redoing anyway.
The articles below go into each piece on its own terms.
The explainers
AI video upscaling redraws detail a resize can’t
Measured on real frames: a resize is the most exact, an AI model restored up to 96% of the fine detail by drawing it.
Why Topaz ended perpetual licences
Topaz Labs stopped selling perpetual licences on 3 October 2025. What they cost before, what replaced them, and what neither source says about a licence you already own.
An offline Mac video editor opens no connections while it edits
Measured: ten real jobs, no connection outside the Mac. How to check any editor, and when Crisp does go online.
Why your video looks worse after you upload it
Every platform re-encodes what you send it. What that does, and the two things that actually help.
What is interlacing?
Where those horizontal comb teeth come from, and why deinterlacing has to happen before anything else.
What is telecine?
Why footage your file calls progressive still combs every few frames, and how to spot the pattern.
What is frame interpolation?
How smooth motion is generated, where it fails, and when smoother is the wrong choice.
Bitrate vs resolution
Why a 4K file can look worse than a good 1080p one, and which number to protect.
What is a video codec?
H.264, HEVC, ProRes and AV1, and why confusing a codec with a container breaks playback.
What none of this can fix
Restoration reconstructs plausible detail from what survived. It cannot recover information that was never recorded, and being clear about that saves a lot of wasted rendering.
Focus is the big one. A shot that missed focus has no high-frequency information anywhere in the frame, so there is nothing for a model to sharpen towards. It will produce something that looks crisper at a glance and is essentially invented. The same applies to severe motion blur from too slow a shutter.
Blown highlights are gone in the same way. Once a region has clipped to pure white, every pixel in it holds the same value, and no amount of processing distinguishes a bright sky from a brighter one. Detail can be darkened but not restored.
Tape damage from a misaligned head, dropouts where the signal was never read, and heavy chroma bleed on old analogue transfers are all better fixed at the source if a source still exists. Re-capturing a tape on a properly aligned deck routinely beats anything software can do afterwards, and it is worth checking before committing hours to processing.
Where restoration genuinely earns its keep is soft-but-focused footage, low resolution, compression damage, sensor noise, low light, minor camera shake and the ordinary degradation of ageing digital files. That covers most of what people actually have.
Why AI upscaling can't rescue a shot that missed focus
What restoration genuinely recovers, and what it only invents.
What is variable frame rate?
Why audio drifts out of sync partway through a screen recording.
Film grain and digital noise are not the same thing
One people pay to add, the other they pay to remove. How to tell which you have.
Why phone video looks shaky
Handheld wobble and rolling shutter are different problems. Only one is fixable afterwards.
Why iPhone HDR video looks washed out
Measured: read as ordinary video, an iPhone HDR frame tops out at 74% white; an HDR10 one loses half its colour. Why, and the fix.
Why upscaling to 4K takes so long
Where the hours actually go, and the levers that genuinely shorten them.
What is rolling shutter?
Leaning lamp posts and wobbling propellers, and the honest answer about fixing them afterwards.
What is chroma subsampling?
Why your video stores colour at a quarter the detail of brightness, and when that shows.
What is a keyframe?
Most frames are not pictures. That explains sticky scrubbing and cuts landing off-target.
BT.601 vs BT.709: why greens and cyans shift
Measured: decoded with the wrong colour standard, pure green moves 15%, and Rec.2020 read as Rec.709 loses a quarter of its colour.
Why VHS tapes degrade
What is physically failing, and which faults software can and cannot recover.
What is aspect ratio?
Black bars, vertical crops, and the difference between cropping and actually reframing.
And the measurements behind all of it
These pages explain what is happening to your footage. If you would rather see it counted, The Video Upscaler Index has seven metrics computed from raw frames, the harness that produced them, and the results in full — including where Crisp comes off worse.