How do I know if my video is interlaced?

Pause on a frame with fast movement and zoom in. Horizontal comb teeth along moving edges — and only on moving edges — means interlaced. A still shot in the same clip will look perfectly clean, which is why it is easy to miss.

Does deinterlacing lose quality?

Yes, unavoidably. You either discard half the vertical detail or interpolate it back, and both are lossy. The alternative is leaving visible combing, which is worse.

Should I deinterlace before or after upscaling?

Before, always. An upscaler treats comb teeth as real detail and sharpens them into permanent artefacts.

Crisp › How-to › Deinterlace

How to deinterlace video on a Mac

If moving objects in your footage have horizontal comb teeth along their edges while everything stationary looks fine, the video is interlaced. It has to be straightened out before anything else touches it — every other operation treats those teeth as detail worth preserving. Crisp now does that step for you when you enhance or upscale; the rest of this page is what it is doing, and how to do it yourself if you would rather.

What interlacing is

Broadcast television could not transmit enough bandwidth for full frames at a smooth rate, so it sent half of each frame at a time: all the odd-numbered lines, then all the even ones, a fiftieth or sixtieth of a second later. On a CRT this worked, because the phosphors and your eye blended the two fields into one apparently complete image.

Modern displays do not work that way. They show whole frames, so the two fields, captured at genuinely different moments, get displayed simultaneously. Anything that moved between them appears twice, offset, interleaved a line at a time. That is the comb.

It is invisible on static shots, which is the trap. A talking-head interview can look completely fine until someone waves, and then a third of the frame turns into a zipper.

Where it comes from

Source	Interlaced?
VHS, Hi8, MiniDV camcorder tape	Almost always
DVD (especially TV recordings)	Usually
Broadcast capture, 1080i	Yes — the "i" says so
Old digital camcorders, 60i mode	Yes
Anything shot on a phone	No
Modern cameras, 1080p / 4K	No

The two ways to fix it, and what each costs

Blend or discard. Throw away one field and interpolate the missing lines back. You keep the original frame rate and lose real vertical resolution: a 1080i source becomes something closer to 540 lines of genuine detail, resampled back up. Simple, fast, slightly soft.

Bob, or field doubling. Treat each field as its own frame, so 50 fields per second become 50 frames per second. You keep the full temporal smoothness — motion looks genuinely fluid, which is what interlacing was buying in the first place — at the cost of doubling the frame rate, which changes how the footage cuts against anything else.

For home video and archive restoration, the first is usually right: it preserves the frame rate the rest of your edit expects. For sport or anything where the motion itself is the content, the second preserves something real that the first discards.

There is no third option that loses nothing. The information genuinely was captured at two different instants, and any single frame you produce is a compromise between them.

Why it has to come first

An AI upscaler has no concept of fields. Hand it a combed frame and it sees a regular high-contrast horizontal pattern, exactly what it has been trained to preserve and enhance, and it will render those teeth sharper and more permanent than they were. Denoisers make the same mistake in reverse, reading the regular pattern as structure and protecting it.

Once either has run, the combing is no longer a separable artefact. It is part of the picture.

The check that takes ten seconds

Scrub to a shot with real movement — a person walking, a car, a pan.
Pause and zoom in on a moving edge.
Teeth on the moving edge and clean stationary background means interlaced. Uniform softness everywhere with no teeth means it is progressive and merely soft, which is a different problem with a different fix.

If it started life on film, deinterlacing is the wrong fix

Most old footage people want restored was shot on film and transferred to video later. To fit 24 film frames into 30 video frames a second, the transfer splits some film frames across two video frames — 3:2 pulldown, or telecine. The result looks like interlacing, and it is not: the original progressive frames are all still in there, chopped up.

Deinterlace it and you throw away what you were trying to keep. yadif will happily process a telecined file, soften every frame it touches and leave you at 30fps. The right move is inverse telecine, which reassembles the film frames and gives you back the 23.976fps progressive footage the camera actually shot:

ffmpeg -i in.mp4 -vf "fieldmatch,decimate" -r 24000/1001 out.mp4

Two things worth knowing, both measured on a 1936 newsreel we restored for this site. The first is that the file lies about itself: the original scan honestly reported its fields, but once it had been cut and re-encoded — which is how most footage reaches you — the container claimed to be progressive while every single frame was still combed. Do not trust the flag; look at the frames. ffmpeg -i in.mp4 -vf idet -f null - counts them for you, and the last "Multi frame detection" line is the verdict.

The second is that the check stops working once anything has been resized. Scaling blends the two woven fields into each other, so the comb teeth turn into a soft ghost that looks like motion blur. On our clip the 640×480 original read as interlaced and the identical frames scaled to 1280×720 read as clean — with the damage still plainly visible. Check the original, before you touch it.

Inverse telecine also has a bonus: it removes about a fifth of the frames, because a fifth of them were duplicates. Anything you do afterwards — restoring, upscaling, exporting — has 20% less work to do.

What to deinterlace with

If you are using Crisp, nothing. It checks for combing when you enhance or upscale, and straightens it out first, before anything sharpens it. You do not have to spot it, and you do not have to run another app before this one.

It also picks between the two fixes above rather than applying one to everything, because the right answer depends on where the footage came from:

Shot on video — genuinely interlaced, two different instants per frame. Crisp deinterlaces it, the blend-or-discard route, and the frame rate stays where it was.
Shot on film — telecined, whole frames split across fields to fit 24 into 30. Crisp reverses the pulldown instead, so you get the frames that were actually photographed back at 24fps rather than blends of two of them.

The line under the progress bar tells you which one it chose — it reads Inverse telecine or Deinterlaced at the end of the route. On a DVD-shaped telecined clip here, 300 frames became 240 and the whole job went from 19.6s to 17.4s: reversing the pulldown removes a fifth of the frames, so everything after it has a fifth less to do. On a very short clip the pre-pass costs more than it saves — a three-second 480×360 file ran about 4% slower — but it is still the version made of real frames.

Two exceptions worth knowing. The Max preset restores the frames it is handed and does not straighten them first, so if you want Max on combed footage, run an ordinary Enhance over it first and take that result to Max. And HDR sources skip the check entirely. For either of those, or if you are not using Crisp at all, the two free tools are:

HandBrake — free and open source, with a window. Its Decomb filter detects combing per frame and only deinterlaces the frames that actually need it, which is the "blend or discard" route: same frame rate, slightly softer.
ffmpeg, if you are comfortable at a command line. -vf yadif is the standard blend/discard filter and -vf bwdif is a better-quality version of the same idea. For the bob route — field doubling, double the frame rate — use -vf yadif=1.

Download Crisp for Mac

Background: what interlacing actually is.

Background: why tape captures come out interlaced.