What the stage is for
An edit passes through recognisable stages, and each one answers a different question. The rough cut answers the largest one: does this thing work at all.
At this point every scene exists, in order, at roughly the right length. Nothing is polished. Sound is whatever was recorded on the day, transitions are hard cuts, effects shots are placeholders or missing, and the grade is untouched. That absence is deliberate. Polish makes a sequence more persuasive, and a stage designed to find structural problems does not want to be persuasive.
The most valuable outcomes here are subtractive. A scene that can be removed entirely is the biggest possible improvement, and it is much easier to see before anyone has invested in making it beautiful. This is also why showing a rough cut to a small number of people is standard practice, and why their notes are worth more than notes on a finished version.
Reading one without being misled
Watching a rough cut requires some discipline, because the missing pieces distort the experience in predictable directions.
- Everything feels slow. Missing sound and missing effects both remove momentum. Sequences almost always tighten once they are complete.
- Performances look worse than they are. Reaction timings have not been cut yet, and reaction timing is most of performance in a finished film.
- Temp music oversells. A borrowed track can make a mediocre sequence play well, and everyone will respond to it. Use it deliberately and say so.
- Continuity errors read as bigger than they are. Many will disappear once the cut is tightened, so they are worth noting rather than fixing.
The question to hold on to is always structural: could this scene be shorter, later, earlier, or gone.
The stage that saves money on generated work
For anything built with generated footage, this stage stops being a formality and becomes the main cost control, because generation is charged per shot and the edit decides how many shots you actually need.
The wrong order is to generate finished-quality shots and then assemble them, which means paying for material that gets cut and locking yourself into decisions before you know whether they work. Sunk cost then quietly protects bad scenes.
The right order treats generation the way a live-action production treats a shoot: as something that happens after the structure is agreed. Build the rough cut from the cheapest placeholders available, whether that is fast low-resolution generations, stills held for the right duration, or even title cards with the shot description on them. What matters is that each placeholder occupies the correct duration, because duration is what you are testing.
Then cut it, watch it, and remove what does not work. Only after that should the surviving shots be generated at final quality, with prompts refined using what the edit taught you about how long each shot needs to run and what has to happen inside it.