- Observations on film art - http://www.davidbordwell.net/blog -

My name is David and I’m a frame-counter

DB here:

Elsewhere on this site and in Figures Traced in Light [1] (Chapter 1) and The Way Hollywood Tells It [2] (second essay), I sometimes say rude things about the rapid editing in today’s films. But I’m actually a fan of fast cutting–when it’s precise, functional, and thoughtful. Most directors today don’t think about their cuts. Godard: “The only great problem of cinema seems to be more and more, with each film, when and why to start a shot and when and why to end it.”

My interest in fast cutting goes back to the 1960s, when I was first reading Eisenstein and seeing fims by Godard, Truffaut, Hitchcock, and, yes, Richard Lester. At first I would wind sequences backward and forward on a 16mm projector, stopping to pull the film off the reels to hold up to the light. For my dissertation on French cinema of the 1920s, I had a chance to examine films by Abel Gance, Jean Epstein, Germaine Dulac, et cie on editing tables or rewinds. In quickly-cut passages, the patterns of shot-length showed how a director could control the visual pace.

Over the years, when I studied 35mm film prints regularly, I periodically returned to my love of rhythmically cut passages. I studied Dreyer’s Passion de Jeanne d’Arc, the work of the Soviet Montage directors, and Asian action movies, from 1920s Japanese swordplay items to 1990s Hong Kong martial-arts sagas. Somewhat surprisingly, I found that Ozu, as in everything else, was there too; a lot of his intermediate spaces (landscapes, objects in rooms) are cut arithmetically.

By examining 16mm and 35mm prints, I could measure shot lengths in frames. In a sound film, a 24-frame shot would last a second on the screen. Things were trickier in gauging the cutting of silent films, since they were shown at rates between 16 and 24 frames per second. But at least the number of frames per shot gave me a sense of rhythmic relationships. A sequence in Jean Epstein’s Chute de la Maison Usher (1928) displays an arithmetical unfolding that would yield a distinct tempo at any projection speed of the era.

Directors have been counting frames for a long time. Experimental filmmakers like Brakhage did. Ozu had a special stopwatch built to register feet and frames during filming. Hitchcock cared about frame counting too. In Film Art‘s chapter on editing (pp. 224-225 of the new edition), we show how the gas station fire in The Birds gains impact from its steadily shorter shot lengths. The first shot lasts 20 frames, the second 18, the third 16, and so on down to 8 frames.

In sum, frame counting is one valuable way to find how a director can govern the pace of our pickup. When shots come this fast, our involvement can quicken as well.

When 35mm plays us false

For silent films, counting frames can be a problem if you have a “stretch-printed” copy. Since silent films often ran at something less than 24 frames per second (sound speed), some later film versions have been printed to repeat frames at certain intervals, to stretch out the print to 24 fps. That makes it easier to add a synchronized score to the print.

In the 1970s the Soviets were big fans of this procedure, rereleasing many classics by Eisenstein, Pudovkin, and others in stretch-printed 35mm and 16mm copies. The Soviet Montage filmmakers definitely counted frames, calculating the shots to play off each other rhythmically, so adding extra frames spoiled things. Worse, the stretch-printed versions often played sluggishly at sound speed, making the action spasmodic and blurring the figures. The result completely wrecked the rhythmic effects the directors were after. Even today, stretch-printed copies of Eisenstein’s Strike are in circulation.

Videotape and the 3:2 pulldown

Videotape arrived. The first Betavision players measured something–a little tally counter clicked away–but you couldn’t be sure what was being counted. Then came digital displays that measured hours, minutes, and seconds. But what about frames? When I tried to count frames on videotape, I found that the figures didn’t match those I got from film frames.

To see why, we have to remember that television developed its own frame rates. The technical details can get pretty daunting; see here [3] for one account. I also recommend Jim Taylor’s DVD Demystified [4] and Kallenberger and Cvjetnicanin’s Film into Video [5]. Still, we non-engineers can get a general sense of what’s at stake.

There are two primary television standards, NTSC and PAL. NTSC (for National Television System Committee) is used primarily in North America and Japan. PAL (Phase Alternating Line) is common throughout Europe and Asia. Because NTSC video operates at a 60-Hertz refresh rate, the rate for individual frames is close to 30 per second. (Actually, it’s 29.97 fps.) The problem is this: Do you run 24-frame movies at 30 frames per second on television? This would make them look speeded up. Or do you pad out a second of film to fit a second of TV? American engineers took the second solution and created the so-called 3:2 pulldown.

In interlaced video, each “frame” is really two venetian-blind image fields presented as one image on your screen.(1) In the 3:2 pulldown, some film frames are selectively repeated to fill out the video playback format. The film is converted to video by making two fields from one film frame and then three from the next frame. Here is a useful diagram from a site [6] on video transfer.


The 3:2 pulldown preserves three of the original four film frames (video frames AA, CC, and DD above).

But you don’t see this exactly this result when you jog/shuttle through a videotape frame by frame because the VCR has been engineered to display the 3:2 pulldown slightly differently. Stepping through a videotape, we find that some frames will show phases of movement and some repeat a phase. Depending on your playback machine and the tape itself, you may get varying patterns. Usually I see four video freeze-frames decomposing a movement, then they’re followed by an extra frame in which nothing changes.(2) That process turns 4 film frames into 5 display frames. Adding 6 repeated frames to the original 24 frames yields 30 video frames a second. I find that you can get a rough sense of the number of frames in the original shot by ignoring the repeated frames in the video display, but the video frames showing decomposed movement may or may not correspond to the actual film frames.

Laserdiscs and the frame count

Those heavy, rainbow-flashing 12″ discs: Cinephiles loved ’em. They gave us a sharper picture, properly letterboxed, along with terrific sound. Aficionados today still prefer the rich digital sound of laserdiscs to the compression of many DVDs.

Most laserdiscs used the CLV (Constant Linear Velocity) format, which preserved and even exaggerated the effects of the 3:2 pulldown. Counting frames meant stepping through the stills and ignoring the repeated ones, as with videotape, but often the coupled frames (AB and BC above) shivered wildly. But in the high-end CAV (Constant Angular Velocity) format, laserdiscs were able to make each stored frame pristine. This is why people thought that laserdiscs would be used archivally, for storing paintings, photographs, manuscript pages.

A bonus of the CAV format was that in transferring motion pictures to disc, one film frame could be matched to a single video frame. In fact, each film frame was assigned its own number, so the counter display spit out digits with blinding speed. During normal playback, the 3:2 repeated frames were added in, but in STEP mode, CAV laserdiscs gave you the original film frames one by one–steady, crisp, and bright. Frame-counting gained a new piece of hardware.

DVD and frame counting

Laserdiscs used optical analog technology for the image track. But consumers wanted something easier to handle, on the model of the music CD, so in came digital image technology via MPEG-2 and the DVD.

Despite all the differences between digital video and film (color, resolution, artifacts, etc.), DVDs in the NTSC format gained one advantage over tape. Like CAV laserdiscs, they could preserve the 24-frames-per-second rate of film. Essentially MPEG-2 technology encodes the film frame by frame. Flags on the DVD instruct the player to regenerate extra frames during playback. For a more detailed explanation, go to Dan Ramer’s article here [7].

The result is that during freeze-frame viewing, a DVD can yield the original film frames. Watch the Universal DVD of The Birds, and the frame-counts correspond to the 20-18-etc. layout we discovered on a 16mm print. Your remote’s STEP button moves through the original frames, hiding the repeated ones that were shown during normal playback.

But not always. DVDs can sometimes play us false.

Before the DVD was a gleam in Warren Lieberfarb [8]‘s eye, I studied King Hu’s wonderful A Touch of Zen on an editing table. There is a lot of frame arithmetic in the editing. During one swordfight, the mysterious stranger leaps away from the blows of Miss Yang and lands in a medium shot. He steps back, tipping up his face to reveal that she’s bloodied his head. He stares in astonishment. The shot lasts only 20 frames.


Yang stands still, holding her sword at the ready. The shot lasts a mere 9 frames, which dynamizes her stillness.


The stranger turns and runs out of frame in another twenty-frame shot. The static shot of Miss Yang now becomes an abrupt punctuation, and the stranger’s panic registers on us more sharply by being caught up in a repetitive rhythmic pattern.


Yang plunges after him, in a twelve-frame shot.


All four shots add up to less than three seconds of screen time, but they’re completely legible. It seems to me that the shots accentuate a pause in the fight by inserting a striking micro-rhythm, and then they dissolve that by letting the fighters continue the combat. Would that today’s filmmakers thought out their cuts so precisely! (If you want to learn more about King Hu’s editing style, see my Planet Hong Kong and the essay on him in the forthcoming Poetics of Cinema.)

But if you examine these shots on the US DVD of A Touch of Zen, you’ll find that the number of frames is way off from the film. The first and third shots run 25 frames each, not 20 frames; the second shot runs 11 frames (instead of 9), and the fourth 15 (instead of 12). Why? Inspection reveals our old friend, the 3:2 pulldown, at work. Improper authoring has left in the frames that should be jettisoned in step playback.

An even more severe instance: The sequence from Epstein’s Fall of the House of Usher that I admired in my dissertation days. It has become hash on the US DVD. When I try to count frames, I find that apparently the silent original has been stretch-printed and subjected to the 3:2 pulldown. When I click through the frames on a DVD player, I find that many frames are repeated, and some shots have doubled in length.

The lesson: Not all DVDs are created equal, or equally accurate in their correspondence with the original film.

What about PAL?

NTSC interlaced video has a 60-Hertz refresh rate. But PAL has a 50-Hertz one. That means not 30 video frames per second but 25. What do you do in transferring a film? The obvious answer: You speed it up.

Films shot at 24 frames per second tend to run at 25 in PAL videos, both tapes and DVDs. As a result, many films on European video run shorter than they did in the theatre, or on American DVDs. For example, Preminger’s Where the Sidewalk Ends runs 94:41 on the Fox DVD, but the same film, known as Mark Dixon Detective, runs 90:49 on the French DVD release. That’s a difference of nearly four minutes. PAL video playback of a 24 fps movie runs about 4% shorter than the original. For this reason, some European films are shot at 25 frames per second, to facilitate transfer to PAL.

So counting frames on a PAL videotape or DVD won’t necessarily mislead you on the number of frames, since none is repeated. That’s the case with our Touch of Zen example. Both the UK and the French DVDs of King Hu’s film yield the correct frame-counts for the shots in question. We just need to remember that each frame is onscreen for a tiny bit less time. In other words, each shot will contain the right number of frames, but that shot will be briefer in screen duration .

Note to Sátántangó fans: Yes, that means a much shorter movie. The PAL DVD [9] available from England’s Artificial Eye claims to run 419 minutes; if you disregard opening and closing credits, the film runs a shade under 400 minutes. But the 35mm version I clocked here [10] runs 434 minutes sans credits. So if you want to add 34 minutes to your life, watch the import DVD.

Note to Asian cinema fans: For reasons I don’t understand, many Hong Kong DVDs yield very blurred and vibrating still-frame images, not corresponding to the frames of the original film. These discs may be using a downmarket authoring process, or there may be problems converting to and fro between PAL and NTSC.

Progressive scan

This all applies to interlaced video, the original television broadcast format. Now we have progressive-scan video, for plasma and LCD displays and computer monitors. A progressive display, still adhering to the 60-Herz standard, is rigged to preserve the right number of film frames without repeating any.

But it does so by in effect combining two film frames together, and the step function may skip over several frames, or create a blurred composite that isn’t there on any frame. This is especially apparent on media-player computer software.

When I run my Touch of Zen shots on my computer, I get wildly abbreviated frame-counts in Step mode. In WinDVD, for instance, the first shot of the UK DVD registers as 8 frames, not 20; the second as 3 frames, not 9; the third as 9 frames, not 20; and the fourth as 6, not 12. Same thing for Epstein’s Fall of the House of Usher. On a tabletop DVD player there are more frames than in the original film; on my player’s software, there are fewer.

For this reason, I’d avoid trying to count frames on a computer’s media player.


Frame-counting is a nifty tool for discovering some secrets of filmmaking, but when we work from a video copy, we need to keep in mind the constraints of the various formats. And whenever possible, check the film!

(1) To complicate things, the scanning of alternate raster lines never yields a single complete frame. Before the image is fully completed at the bottom, it’s being refreshed at the top! To this extent, the video frame doesn’t have the stable, singular identity of a film frame. It can be encoded to correspond to a film frame, but as an image display, even a frozen frame is always in the process of being filled in. It’s just that our eyes, not designed to track such small-scale processes, are fooled.

(2) A quick sampling of my old VHS tapes shows that occasionally a tape doesn’t exhibit effects of the 3:2 pulldown. My Facets VHS copy of Where Is the Friend’s Home? plays back on two machines without any repeated frames. Perhaps this comes from making the transfer from a PAL master tape?