Script to rough cut: what an assistant editor should do (and what it shouldn't)
Every editor knows the gap between a finished script and a first rough cut. The script is locked, the voiceover exists, and nothing is on the timeline yet. Somebody has to turn a document into a sequence.
On bigger jobs, a lot of that work belongs to the assistant editor. On solo jobs, it’s you at midnight. Now that software keeps promising to “edit for you”, it’s worth being precise about which parts of the script-to-edit stage a machine should take, and which parts are the actual job.
What an assistant editor does on a real job
The title sounds junior. The work is what keeps an edit from collapsing. On a typical project, the assistant editor handles:
- Ingest and organisation. Copying media off cards, backing it up, making proxies if needed, and building bins so anyone can find anything. Naming conventions live or die here.
- Sync. Lining up picture with separately recorded sound, and building multicam sequences when there’s more than one camera.
- Logging and selects. Watching the footage, marking usable takes and moments, and pulling the best material into selects sequences.
- String-outs. Laying all the material for a scene or topic end to end, so the editor can see everything they have in one pass.
- Assembly against the script. Putting scenes on the timeline in script order, using the scripted takes, so there’s a first version to react to.
- Conform and relink. Reconnecting the edit to full-resolution media and fixing broken links when files move or proxies are swapped out.
- Turnover. Preparing exports and project files for sound, colour, graphics and VFX, with notes so nobody guesses.
Notice how much of that list is mechanical. It needs care and consistency, but it doesn’t need taste. That’s the opening for automation, and also the trap.
From script to edit: the mechanical layer
There’s a specific band of work between a finished script and a first assembly that is pure translation. The decisions were already made in the script. Someone just has to move them onto a timeline.
On a narration-led video, that band looks like this:
- Lay the scratch VO per scene. Cut the voiceover into scene-length pieces and place them in order.
- Drop labelled placeholders where footage is missing. A slate that reads “S04: aerial of the harbour, 0:08” holds the timing and tells everyone what’s still owed.
- Pull footage candidates. For each scene’s visual plan, find freely licensed options and put them in a bin next to the placeholder.
- Build the timeline in scene order with durations. Markers or clip names that match the script’s scene numbers, so script and timeline speak the same language.
None of these steps requires judgment that wasn’t already exercised in the script. If the script says scene 4 is eight seconds of harbour aerial, building that slot is clerical. I’ve watched good editors lose an evening to it, and I’d hand all four steps to software tomorrow.
What must stay human in the rough cut
The rough cut is where the editing actually starts, and the important parts of it don’t automate well.
Choosing the take. The script says what’s said. It doesn’t say which of six reads has the right weight on the last word. That’s listening, not matching.
Rhythm. A cut that lands a frame early feels nervous. Two frames late feels sleepy. Rhythm is felt across the whole sequence, not computed per clip.
Emotional timing. How long to hold on a face after bad news, or when to let the music breathe before the next line. These calls change how the viewer feels, which is the point of the video.
What to cut. Every rough cut runs long. Deciding which good scene goes, so the video works as a whole, is an editorial argument. Software can tell you a section is long. It can’t tell you it’s the wrong section.
Story. Sometimes the edit reveals that scene 7 should be the opening. That isn’t in the script. It’s discovered on the timeline, by someone who understands what the video is trying to say.
The rule I use: automate anything that executes a decision already made. Keep anything that makes a new one.
Where today’s AI editing tools fit
Adobe has been shipping AI features into its video tools, and they’re worth knowing about. In January 2026 it added AI-assisted object selection and masking to Premiere. In September 2026, ahead of IBC, it announced generative tools in Premiere that can generate video to fill gaps in a timeline, plus generated sound effects and music. Separately, its browser-based Firefly video editor (in beta) has a Quick Cut feature that assembles a first cut from footage you upload.
These are footage-first. They start from media and help you shape it. That’s useful when the footage is the story, like event coverage or interviews. On a script-first video, the plan already exists before any footage does, and the blueprint for the assembly is the script itself.
Script-first assembly: where Syntale is heading
This is the direction I’m building Syntale toward. A script split into scenes, each with a duration and a visual plan, is already the blueprint for the assembly. Turning it into a timeline with scratch VO, labelled placeholders and footage candidates is the mechanical layer above. It’s on the roadmap, not shipped.
Today Syntale covers the half that comes first: a scene-split script with durations and visual plans, shared for comments and approval, and exported as a PDF the editor can cut from. More on that in script handoff to a video editor and scene-based scriptwriting, and for a full narration-channel example, the faceless YouTube workflow. Start for free.
FAQ
What is a rough cut?
A rough cut is the first full edit of a video that tells the story in the intended order, with the main takes chosen. Pacing, sound, graphics and colour are still unfinished, and it usually runs long.
What’s the difference between an assembly and a rough cut?
An assembly puts all the scripted material on the timeline in script order, often with little trimming. A rough cut is the editor’s first real pass: takes chosen, scenes tightened, and structure tested.
Can AI make a rough cut?
AI tools can now produce a first assembly from footage or a plan, which saves real time. Choosing takes, rhythm, emotional timing and what to cut are still editorial decisions, so treat any AI output as a starting assembly, not a finished rough cut.