What an "AI video editor" actually does

"AI video editor" is one phrase covering at least four unrelated products. They have different jobs, different failure modes and wildly different prices, and the marketing pages look identical — same dark hero, same floating timeline mockup, same promise about doing it in minutes.

So people sign up for one thing believing it's another, hit the wall about twenty minutes in, and conclude AI video doesn't work. Usually it worked fine; it just wasn't the tool for the job.

Here's the taxonomy, and the one question that sorts it.

Four things wearing one name

1. A generator with a Save button

Type a prompt, get a clip, download it. Genuinely magic for exactly one shot.

What it can't do is the second shot. There's no cut, no trim, nothing to put a title on. As soon as your idea needs a sequence — and almost every real idea does — you're stitching clips somewhere else. The generating is the easy part; the assembling is the work.

2. A template filler

You pick a template, drop in your logo, your text and a few clips, and out comes something that looks professionally made. Terrific value when your video is a shape that already exists: a listing tour, a quote card, a simple product promo.

The limit is structural. The template's timing is the template's timing. When you want the beat to land half a second later, there's frequently no way to say so — the thing that makes it fast for the common case is the thing that makes it a dead end for yours.

3. A repurposer

Feed it a long video — podcast, webinar, stream — and it hands back short vertical clips with captions burnt in. Excellent at scale, and mostly a selection engine: its actual skill is picking which sixty seconds are interesting, then reframing them.

It's not built for making something from nothing, and its cropping decisions are somebody else's taste. Great for volume, wrong for the one clip that matters.

4. A real editor with AI tools inside it

A timeline, lanes, cuts, keyframes — plus AI where AI is genuinely better than a human with a mouse: transcription, matting, reconstruction, generation. The tradeoff is honest: there's more to learn than typing a prompt, and nothing is one-click.

That's the whole spectrum, from "one prompt, one clip" to "you decide every frame". None of them is the correct answer in general. But only one of them is the correct answer for you, and the fastest way to find out is:

The test

Can I move this cut eight frames later?

If yes, you have an editor. If the answer is "regenerate", "pick a different template", or "there is no cut to move" — you have one of the other three, and no amount of AI in it will change that.

Ask it before you sign up, not after you've imported an hour of footage. A second, meaner version: can I put my own file on the timeline next to the generated one? Some generators quietly can't.

Which AI is actually useful inside an editor

"AI-powered" on an editor's landing page can mean anything from a genuinely load-bearing feature to a chatbot in a sidebar. Four categories earn their keep, and they have one thing in common: each replaces a task that was tedious and mechanical, not one that needed judgement.

Transcription-driven editing. Once the machine knows every word and its timestamp, word-level captions stop being an afternoon of hand-timing, and cutting every "um" and silence becomes a checklist you tick through. This is the single biggest real time saving in the list, and it's boring, which is why it doesn't make the hero section.

Matting and masking. Pulling a subject off its background without a green screen. Was rotoscoping — hours per shot, done by hand, per frame.

Reconstruction. Upscaling and frame interpolation: inventing plausible pixels between the ones you have. Bounded — it recovers plausible detail, not lost detail — but real.

Generation. Making a shot that doesn't exist: the establishing shot nobody filmed, the b-roll you can't buy, the product angle you didn't shoot.

Notice what's absent: nothing on that list decides where the cut goes. Pacing, emphasis, what to leave out — that's the part that is actually editing, and no current tool has taste. Anything claiming to do the whole edit is really doing #2, with extra steps.

What "free" means here

The word behaves differently on either side of one line, and the line isn't about generosity.

Editing is cheap. Cutting, titles, transitions, colour, mixing, silence removal — that's your own computer doing arithmetic. A tool that charges per cut is charging for software, not for compute.

Generation isn't. Every generated second is a GPU running somewhere, billed by the minute to whoever built the tool. Nobody gives that away without limit, and "unlimited free AI video" is always a sentence with a footnote — a queue, a watermark, a resolution cap, a trial length, or your work training the next model.

Which is why "free AI video editor" is a reasonable thing to look for and "free unlimited AI video generator" isn't. The useful question about the paid half is only ever what does one finished minute cost me — in tokens or credits or whatever the unit is, including the failed takes, because there will be failed takes.

Choosing, in one paragraph

If you need one clip and nothing more, use a generator; a timeline is overhead you don't need. If your video has a standard shape and you make a lot of them, a template tool will beat everything else on speed. If you're mining long recordings for shorts, use a repurposer and accept its taste. And if the video has to be specific — your timing, your cuts, your footage mixed with generated shots — you need a timeline, and the AI is there to spare you the tedious parts, not to do the editing.