How to Edit a Video by Just Describing It

For most of video's history, editing has meant learning software. Tracks, keyframes, ripple deletes, render queues: a wall of tools between you and the cut you actually want.

Describe-to-edit inverts that. Instead of operating an editor, you tell an editor what you want, in the words you would use with a person. This guide covers how it works, how to phrase things so it goes well, and (the part usually left out) where the approach stops being the right tool.

What describe-to-edit actually means

You upload footage and, instead of dragging clips, you type what you want. "Cut the dead air." "Add captions." "Put music under my voice." The instruction is interpreted, the edit is made, and you see the result. If it is wrong you say so ("make the captions bigger", "use a calmer track"), and it adjusts.

There is no timeline to manage and no menu to memorise. That sounds limiting until you notice how much of editing is a series of decisions you already know how to describe, executed through an interface that makes describing them impossible.

The decisions were never the hard part. The clicking was.

Start with the shape, then refine

The fastest route to a good cut is top-down: structure first, texture last. A reliable order:

1. 01Tighten the pacing. "Remove the filler words and dead air." On an unscripted recording this alone can take minutes off the runtime, and it makes everything after it easier.
2. 02Set the format. "Make this vertical for TikTok" or "keep it 16:9 for YouTube." Format affects framing, caption style and target length, so decide it before layering anything.
3. 03Fix the structure. Reorder, cut whole sections, move the strongest point earlier. Do this before polishing anything, or you will polish something you later delete.
4. 04Add the layers. Captions, music, b-roll, emphasis.
5. 05Trim to length. "Get this under two minutes."

Each step is a sentence. You are not building the edit, you are directing it.

The most common way to waste time here is jumping to step 4 immediately because you noticed something you disliked. Structure first. Always.

How to phrase instructions that work

The quality of the result depends heavily on how you ask. Four habits do most of the work.

Describe the outcome, not the mechanism

You do not need to know how captions are generated or how audio is ducked. Say what you want to see and hear: "captions that appear word by word", "music that sits quietly under my voice".

Instructions phrased as operations ("add a transition", "apply captions") get you a technically correct result that often misses the intent. Instructions phrased as outcomes ("make the jump between these two points less abrupt") give something to aim at.

Anchor to what was said or shown

Timecodes are hard to eyeball and you have to find them first. But you always know what you said.

"Zoom in when I mention the discount." "Cut the part where I fumble the intro." "Add b-roll while I explain the pricing." Each gives an anchor that is easy to locate and unambiguous.

One thing at a time when it is fiddly

For structural work, batching is fine: "cut the dead air and add captions." When fine-tuning, single instructions are better, because you can see the effect of each and course-correct.

Five small instructions almost always beat one elaborate one. An elaborate prompt gives you a single shot and no diagnosis when it misses.

Use comparisons

"Bigger." "Quieter." "Snappier." "Warmer." These work precisely because they are relative to what you are already looking at, which means you do not need absolute vocabulary for things you only perceive comparatively.

Point instead of describing

Some edits are genuinely easier to show than to say, and this is where pure language hits its limit.

If there is a distracting object in the corner of the frame, "remove the thing in the bottom right" is ambiguous the moment there are two things in the bottom right. Describing which one takes longer than the edit is worth.

So do not describe it, point at it. Pause on the frame, circle the object, say "remove this." It is the difference between giving directions over the phone and pointing across the room. More on how that works.

Iterate like a conversation

The real mindset shift is treating editing as dialogue rather than command entry. You are not trying to write the perfect prompt. You make a change, react, make the next one.

This is faster than it sounds because you skip learning and operating tools. A change that would be five minutes of clicking is a five-word sentence. And because you can undo with words ("actually, put that clip back") experimenting costs nearly nothing, which means you experiment more, which is how you end up with a better video.

A worked example

A ten-minute talking-head recording for YouTube, plus a short for Reels. The whole session:

• "Remove the filler words and dead air."
• "Add clean captions, centred, not too big."
• "Add subtle background music that ducks when I talk."
• "Zoom in a little whenever I make a key point."
• "That's the main video, now make a punchy 30-second vertical version for Reels from the best moments."

Five sentences, two finished videos, no timeline. The editing still happened. You were not the one clicking.

Where this approach is not the right tool

Worth being honest about, because every approach has a shape.

Frame-exact work. If you need a cut on a specific frame to hit a specific beat, a timeline is the correct instrument. You can get close by describing it ("trim half a second off the front"), but if precision to the frame is the requirement, use the tool built for precision.

Complex compositing. Layered effects, motion graphics with dependent animations, keyed elements over generated backgrounds. Describable in principle, faster in a dedicated tool in practice.

When you genuinely enjoy editing. Some people like the craft, and the craft is real. If the timeline is where you do your thinking, an instruction layer removes the thing you like.

When you do not know what you want. Describing an edit requires having an opinion about the result. Scrubbing a timeline is partly a way of discovering what the video should be. If you are exploring rather than executing, the conversation can feel oddly demanding.

Where it shines

The inverse of the above: the content most people actually make. Talking-head videos, podcasts, tutorials, vlogs, product clips, social shorts. These lean on the same handful of moves (trimming, captioning, scoring, reframing, emphasising), and every one is easy to describe.

If you have been putting off editing because the software felt like a second job, describing the edit removes the job and leaves the decisions.

The decisions were always the interesting part. For phrasings that work reliably, the prompt collection is the practical companion to this piece.