Blog · The cut
How to tighten a cut without losing what it says
· Updated · 5 min read
Short answer
Tighten from the largest unit down: repeated beats first, then lines, pauses, and the heads and tails of shots, with frame-level trims last. Turn the target into frames before you start. Stop when you reach it, or when the next cut costs meaning, a reaction or variety instead of time.
On this page
Work the other way and you polish frames you later throw out, and every shot drifts toward the same length.
How much are you actually taking?
Turn the target into frames and divide by the cuts. That one sum tells you whether trimming can do the job.
Say a 3:00 piece at 24 fps has to run 2:40, with 30 cuts in it. That’s 20 seconds, or 480 frames. Taken evenly from every cut, it’s 16 frames each: two-thirds of a second off every edit, through words and reactions alike. Trimming can’t pay for that, so most of the frames have to come from somewhere else.
Being long at this stage is normal. Thelma Schoonmaker told Film Comment that her cuts with Martin Scorsese always start long, and that the work is bringing them down until the pace is right.
In what order do you cut?
Largest first, because each step changes what the next one should do.
- Watch it through without touching it. Mark where you feel it drag. Mark, don’t fix: a fix made here will be undone by step 2.
- Take out repeats. Find two beats that say the same thing, keep the better one, and lift the other whole.
- Take out lines and clauses. Remove what the viewer already knows, and check the setups you may have just orphaned.
- Close pauses and fillers. Breaths, ums, and the gap before an answer that isn’t doing anything.
- Trim heads and tails. Cutaways that overstay, entrances, exits, a walk across the room.
- Adjust frames last. One- and two-frame rolls move a cut without changing the length. They change how a cut feels, and they’re wasted on anything that may still go.
Step 3 is where most scenes survive. Schoonmaker said of The Wolf of Wall Street that losing some lines was better than losing whole scenes. Jennifer Lame told SlashFilm she couldn’t recall a single scene cut out of Oppenheimer entirely: they cut scenes down, and not for running time, but to keep from going too far down any one path. Removing a repeat in step 2 isn’t in tension with that. A scene that says something nothing else says gets shorter, and one that doesn’t, goes.
Where do the frames come from?
Mostly from whole units. Here’s one way the 480 frames could add up on a 3:00 interview piece. The figures are an illustration, not a measurement.
| Source | Example | Frames at 24 fps | What it can cost |
|---|---|---|---|
| A repeated beat | One 9-second answer that an earlier answer already covers | 216 | A good moment you liked |
| Lines and clauses | Three lines of about 2 seconds each | 144 | A setup or a payoff |
| Pauses and fillers | 12 pauses, 6 frames off each | 72 | Thinking that looked like thinking |
| Heads and tails | 8 cutaways, 4 frames off each | 32 | Shots with no time to settle |
| Entrances and exits | 2 shots, 8 frames off each | 16 | Where people are in the room |
| Total | 480 |
Three-quarters of the budget, 360 frames, comes from the first two rows, before a single shot has been trimmed. The last two rows together give 48 frames, a tenth of it. If your plan is mostly the bottom of this table, you’re likely to end up with the same film, slightly out of breath.
Each row changes what’s left. Taking out a line often leaves a cut with no reason to be there, so check the two shots either side of it. Closing a pause in a talking head leaves a jump, which may want a cutaway or a J-cut to cover it.
How do you know a cut costs too much?
Rank what you’re giving up. Walter Murch’s rule of six, from In the Blink of an Eye, ranks what a cut has to do: emotion, story, rhythm, eye-trace, the two-dimensional plane of the screen, and three-dimensional space. A summary on PremiumBeat gives his weights as 51, 23, 10, 7, 5 and 4 percent. When a cut can’t have everything, you give up from the bottom of the list.
Tightening is mostly a rhythm decision, and rhythm is worth 10 of those 100 points. A trim that helps rhythm and costs a reaction has traded down, even if the scene now moves better.
So tight doesn’t mean short. In a diner scene in Baby Driver, a film built around exact timing, Paul Machliss told Steve Hullfish he stayed on Ansel Elgort for perhaps 20 seconds, because it was the right moment to slow down. A hold like that does what no trim can. Cut elsewhere to pay for it.
For the frame-level pass, one useful landing point: Murch told Hullfish that the moment to cut is generally a frame or two before an actor in character blinks, and that you should override it when your own sense of the shot’s rhythm says so.
When do you stop?
Stop at the number. These are our own tests, not a standard, for stopping before it or for knowing you’ve gone too far:
- You’ve hit the target. The next trim is a habit, not a decision.
- The shots have converged. If every shot now runs about the same length, you’ve removed variety along with time. Hullfish warns against a cut that stays the same from start to finish, and suggests watching edits with the sound off to see their pacing. Do that with yours.
- Putting frames back helps. Take the cut you trimmed last and return 6 frames. If that plays better, you’ve overshot, and probably not only there.
- A new viewer gets lost. You can’t test for this yourself, because you still remember the line you took out. Show it to someone who’s never seen it and ask what happened.
- You can’t tell the versions apart. If two trims look the same at full speed, the difference isn’t worth a decision.
Duplicate the timeline before each pass, so that going back costs nothing and you’re comparing against the real earlier cut, not your memory of it.
Can an agent do the tightening?
It can do the mechanical part. In nolanlabs, our editor, you can ask in plain words, like “tighten the second half by a few seconds.” The agent’s timeline.tighten op reads the words the cut plays, shortens pauses over 12 frames down to 6 by default, and takes out um and uh. The result comes back as a proposal: Ctrl+’ switches between it and the original, and Ctrl+Z undoes it. It reads words, not performance, so it can’t tell a dead pause from one the scene needs. Those you put back yourself, with a drag. That split, legwork to the tool and judgment to you, is the one the labor data points to as well.