AI Rough Cuts & Selects

Let's name the real enemy of every low-budget edit. It isn't talent and it isn't software — it's the ratio between how much footage you shot and how much attention you actually have. Call it the selects problem: hours of material, and maybe one good evening of genuine, awake, taste-making focus before life intrudes. Traditionally you spent that precious attention on triage — watching everything once just to find out what you had. Which means your freshest eyes, the ones seeing the footage the way an audience will, got burned on the most mechanical task in the entire edit.

AI-assisted selects exist to fix exactly that ratio. The machine watches everything so your first real viewing happens at the level of candidates, not chaos. That's the whole pitch — not that the AI edits your film, but that it absorbs the triage.

From mountain to assembly

Here's the working sequence, whether you're using a dedicated assist or building it from the transcription workflow in the last chapter:

• Ingest and transcribe everything. Day one, per Chapter 5 — selects tools are only as good as the searchable footage underneath them.
• Let the AI flag usable takes. Complete lines, no interruptions, clean audio — the machine strips out the obvious dead material before you spend a minute of attention.
• Group by scene and line. Every reading of the same moment stacked side by side, so comparing take 2 against take 7 takes seconds instead of scrubbing.
• Surface the best-audio takes. Let the tool rank for clean dialogue — a genuinely mechanical judgment — while you stay focused on performance.
• Build the first assembly. Selects laid in story order: long, baggy, and watchable. A sketch of the film, hours after the footage landed.
• Keep a "kills" bin. Everything the AI rejected goes in a bin you skim before locking any scene — because the machine is wrong about your best take often enough to check.

That last item is a rule, not a suggestion. The AI ranks takes by measurable cleanliness — audio clarity, completeness, technical polish. Your best take is frequently the messy one: the overlap, the flubbed-then-recovered line, the reading where something real happened through the noise. Those live in the kills bin. Skim it. Every project, it hands you back at least one moment you'd have lost.

A sketch, not a cut

So how do you hold an AI rough cut in your head? The same way a novelist holds a first draft: it exists to be rewritten, and its whole value is that it exists at all. A blank timeline is paralyzing; a bad assembly is legible. You watch it and instantly know things — this scene starts too early, that take is wrong, the middle sags — that you could not have known while staring at bins. Reacting is faster and truer than generating from nothing. That's the psychology the AI assembly exploits, and it's legitimate.

The failure mode is treating the assembly as a cut that needs trimming rather than a sketch that needs rewriting. The difference is posture. Trimming accepts the machine's structure and shaves it; rewriting keeps your hand on every decision — reordering scenes, swapping takes from the kills bin, restoring the pauses the machine deemed dead air. Remember the taste rule from Chapter 4: the assembly is labor, and labor is exactly what we automate. Everything after the assembly is judgment, and the judgment starts the moment you press play. The machine got you to the starting line days early. Now the actual edit begins — and it's all yours.

Assembly in hand, structure emerging — the next job is making it all look like one film. Next chapter: AI color, and why the machine's pass is a first pass, never the grade.