Let’s get one thing straight: a rough cut is not a bad version of your video. It’s the honest version. Before colour, before captions, before anyone argues about a title font (and someone always argues about the title font), the rough cut answers one question. Does this sequence of generated clips actually tell the story we planned? In our workflow that question comes early. On purpose. We generate short clips in Premiere-ready batches, we assemble them in story order, and we watch the whole thing with no polish to hide behind. If the piece doesn’t work rough, no amount of finishing will save it. That’s not a theory. It’s just how it goes.
Polish Last, Structure First
Generated footage invites the wrong instinct. A single clip can look so clean on its own that we want to grade it, stabilise it, and add music before we even know whether it belongs in the video at all. We’ve all felt that pull. We resist it. Our Premiere stage starts with structure — because structure is the one thing finishing cannot repair. A beautiful clip in the wrong place is still in the wrong place. It doesn’t matter how beautiful it is. Wrong is wrong.
- Sequence first: build the delivery sequence before importing opinions — 1080×1920 at 30fps for vertical, 1920×1080 for horizontal, matching the plan from the storyboard.
- Selects only: only clips that passed review go into the rough-cut bin. Near-misses stay in the raw bin, out of sight. Out of sight means out of temptation.
- No effects yet: no Lumetri, no Warp Stabilizer, no transitions except straight cuts. If a join needs a dissolve to feel acceptable now, mark it — that’s a story problem, not a transition problem. Write it down and keep moving.
- Temp audio only: a scratch voice track or a simple music guide is enough to judge timing. The real mix comes after picture lock. Always after.
This discipline mirrors the wider assembly approach in our guide to editing AI clips in Premiere Pro, but the rough cut goes further: it deliberately removes every tool that could flatter weak footage. Every single one. If the story can’t survive without the flattering tools, the story isn’t ready.
The String-Out: Every Approved Clip In Order
We start with a string-out — every approved clip laid end to end in storyboard order, at full length, with no trims. It will run long. A 60-second plan might string out to two or three minutes. That’s fine. The string-out is not the video; it’s the evidence. Think of it as the crime scene before the verdict.
- Order by storyboard: follow the boards shot by shot, even where your instinct wants to shuffle. The plan earned its structure on paper. Trust the plan, at least for this pass.
- Use markers, not cuts: as you watch, drop markers where a clip starts late, ends early, drifts, or repeats the previous shot. Do not cut yet. Marking is fast. Cutting is a commitment. Commit later.
- Note the dead zones: AI clips often settle into a pleasant but static final second. Mark it. That second rarely survives. Pleasant isn’t the same as useful.
- Watch once, fully: no pausing to fix. The first full watch tells you where attention drops — and attention is the real currency. Everything else is decoration.
After that watch, we trim the string-out into a true rough cut. We keep heads and tails a little long at this stage — handles matter later for J and L cuts and for small timing repairs. A little insurance costs nothing here and saves real money later.
Radio Edit Thinking: Does The Story Play Blind?
Before judging pictures, we listen. Turn the monitor off — or minimise Premiere — and play the cut with its temp audio: scratch narration, clip audio where it exists, silence where it doesn’t. We call this radio edit thinking, borrowed from audio editing, where the story must survive without images. It’s a brutal test. That’s the point.
- Clear turns: can you hear where the setup ends and the complication begins, without seeing anything? If not, the structure is muddy.
- Speaker clarity: because we typically generate one speaker per generation, does each voice enter cleanly, or do lines step on each other? Stepped-on lines mean the edit is rushing.
- Beat gaps: long silent stretches usually mean a visual beat is carrying weight it can’t carry alone. Silence is only a feature when it’s doing work.
- Ending shape: does the last line land? Or does the piece simply… stop? A video that stops instead of ending feels unfinished no matter how pretty it is.
Where the blind listen fails, the fix is structural: reorder a beat, shorten a pause, move a line earlier. Those decisions belong here, while changes cost seconds. Move a clip now and it’s a drag. Move it after grading and it’s an afternoon. The same audio-first logic supports our approach to pairing voiceover with AI video later in the finish.
Trimming To Beats Before Touching Colour
With the order proven, we trim to beats. Every shot in the rough cut should start at the last useful moment and end at the first moment it stops adding information. That’s the rule. It’s a harsh rule and it works. Generated clips make this easy to get wrong because they often begin with a half-second of settling — the model finding the subject — and end with a gentle drift.
- Cut the settle: trim heads until the subject is already committed to the action. No warm-up. The viewer doesn’t need the warm-up.
- Cut on motion: leave a shot while something is still moving, and enter the next shot already moving. The join disappears. Like it was never there.
- Hold the reaction: the one place we keep extra frames is a face reacting. Reactions are where a generated sequence starts to feel directed. Give them room.
- Time the turns: in a 60-second piece, check that something changes — shot size, location, information — every few seconds. Flat stretches show up fast in a trim pass. And they bore fast, too.
Notice what’s still missing: colour, stabilisation, captions, titles, music. That order is the point — and it’s the same order we use when we build multi-shot 60-second stories from single-speaker generations. Rhythm first. Beauty second. Always in that order.
Locking Picture: The Point Of No More Generation
Picture lock is the most valuable moment in an AI video workflow. Why? Because generation is always tempting. One more variation. One better angle. One cleaner take. The tools make that loop feel productive. It’s not. The rough-cut process ends by closing that loop on purpose. Deliberately. With ceremony, even.
- Lock means lock: after this point, we do not generate new shots for this video. A genuine break — a corrupted clip, a continuity hole — is the only exception. And “genuine” is doing a lot of work in that sentence.
- Watch at full length: screen the locked cut once, start to finish, at delivery size, before signing off. No skipping. No “I know this part.” Watch it.
- Write the fix list: every remaining problem becomes a finishing task: grade, sound, captions, a stabilise here, a punch-in there. Nothing on the list should say “rethink the middle.” If it says that, it’s not locked.
- Duplicate the sequence: keep the locked version untouched and finish on a copy. If a finishing idea fails, the honest cut is still there. Waiting. Patient.
Locking early protects the schedule. It also protects morale. The team stops browsing generations and starts finishing a film. There’s a psychological shift — from “what if” to “let’s make this good.” It’s underrated.
A Worked Pass: From String-Out To Locked Cut
Okay, let me show you what this actually looks like on a real short. Say the storyboard calls for eight shots: an opening wide, a close detail, a speaker beat, a reaction, a second location, another detail, a turn, and a closing image. Seedance or Veo gives us maybe twelve usable generations for those eight slots, because a couple of shots needed a second attempt. Nothing unusual. That’s just how it goes.
We lay all twelve out in board order and watch. The piece runs long. It feels even longer in the middle — you know that feeling, where time stretches out because nothing’s changing? Two things jump out straight away. Shot four and shot five are saying the same thing visually, just from different angles. And the turn — the whole reason the video exists — arrives so late that we were checking our patience before it landed. Not our watches. Our patience.
So we do the unglamorous work. One of the duplicated shots goes. The turn moves two beats earlier. The opening wide loses its settling frames and starts already moving. Nothing has been graded. Nobody has touched a caption. Yet the cut suddenly has a pulse. Because the problem was never picture quality. It was order and weight. It was never the pixels. It was always the arrangement.
That is the part new editors tend to skip. They keep generating, hoping a better clip will fix a saggy middle. A better clip will not fix a saggy middle. Moving the turn will. Write that on something.
When The Rough Cut Says Generate Again — And When It Doesn’t
Honesty cuts both ways. Sometimes the rough cut exposes a genuine hole: a missing reaction, a location that never reads clearly, a line delivered in a shot where the mouth is barely visible. Then yes — go back and generate. That’s what the rough stage is for. No shame in it.
But learn to hear the difference between a hole and a preference. A hole stops the story making sense. A preference is just you fancying a nicer angle while the perfectly good one sits there doing its job. Regenerating for preferences is how a one-day edit becomes a four-day edit and still ships late. We’ve watched it happen. We’ve done it ourselves. Learn the difference.
- Ask whether a first-time viewer would notice the problem without you pointing at it. If not, write it on the finishing list and move on. Your taste is not the audience’s taste.
- Check whether the flaw can be trimmed out. A surprising number of “bad clips” are good clips with two bad seconds attached. Trim the bad seconds. Keep the good clip.
- Try a reorder before a regenerate. Shots read differently depending on their neighbours — and the neighbour is free. Regeneration costs a queue. Reorder costs a drag.
- If you do generate again, generate against the cut — same framing, same light direction, same screen position — so the new clip drops into the hole instead of starting a new argument. Fit the hole. Don’t make a new shape.
One more edge case: the clip that is technically flawed but emotionally right. A reaction that lands, even with a slightly soft final second, will beat a pristine replacement that feels empty. Every time. Keep the one that plays. Audiences remember how a moment felt long after they forget whether it was sharp.
Frequently Asked Questions
How Long Should A Rough Cut Be Compared To The Final Video?
Longer, almost always. We let the string-out run well past the target, then trim down to the beats that earn their place. Cutting down is easy and honest. Padding a thin cut up to length is where weak shots sneak back in — and they’ll bring friends.
Should I Add Music Before Or After The Rough Cut?
A quiet temp track is fine for judging pace, but don’t cut to a finished mix yet. Music flatters everything. It flatters bad structure too, which is exactly the problem. Get the story working with straight cuts and scratch sound first, then let the real track refine timing later.
What If A Generated Clip Is Great But Does Not Fit The Storyboard?
Save it for another video. Forcing a beautiful clip into the wrong slot bends the story around the footage, and viewers feel that bend even if they can’t name it. Your selects bin will thank you on the next project. Nothing is wasted if it’s stored.
How Do I Know The Picture Is Really Locked?
Watch it once, start to finish, without touching the keyboard. If you can sit through it as a viewer — following the turns, feeling the ending land — it’s locked. If your hand keeps reaching for the timeline, it’s not. Your hand knows before your brain does. Listen to your hand.
What Rough-Cut Honesty Saves You Later
The rough cut is where weak ideas are still cheap. Moving a beat costs a drag. Replacing a shot costs nothing because nothing has been graded, mixed, captioned, or timed to music yet. An hour later in the process, the same change means re-grading a replacement, re-timing captions, re-mixing audio, and re-checking exports. We’ve learned to pay the small price early. It’s just cheaper. In every sense.
It also saves your judgement. Polished footage lies politely — it really does, it whispers “I’m fine” while the structure crumbles underneath. A rough cut cannot lie. Watched plain, with straight cuts and temp sound, a sequence either holds attention or it doesn’t. That clarity is what makes the finishing stage enjoyable instead of desperate — we’re improving something that already works.
So build the string-out, listen blind, trim to beats, and lock the picture. Then, and only then, open the colour workspace. The mindset shift is simple: a rough cut is not an unfinished video waiting for polish. It’s the finished argument, waiting for presentation. Remember that, and the whole process gets lighter.