The best AI video clipping tool, measured on one file

Contents

The short answer

Best overall: Vizard — it found 30 clips in the test panel against OpusClip’s 32 and Submagic’s 31, at 85 cents a job versus OpusClip’s $2.03 on an identical meter. Its free plan, though, releases 10 of those 30 and charges for the remainder.

Best free: CapCut — the only clean free export among the four I ran, 1080p with no watermark, though on Auto, the setting it opens on, what came back was 8 long chapters spanning 73.6% of the source rather than short highlights.

Best flat fee on long recordings: Submagic — cheaper than OpusClip past about 26 source minutes and cheaper than Vizard past about 63, because the fee was quoted the same for a 35-minute source and a 46-minute one, though its trial refuses every export.

Before any of that, the thing worth knowing about searching for the best AI clipping tool: when I checked on 16 September 2026, there was not a single independent article on the first page of Google for it. Every organic result is a tool’s own website, a Reddit thread or a YouTube video. Not one independent article. Google’s AI Overview recommends four products and cites each one to that product’s own marketing page. So the recommendations a searcher currently gets are, without exception, tools describing themselves.

That is why this post exists, and why it is built on one file rather than a feature matrix.

How did I test these AI clipping tools?

I ran the same video through four tools on their default settings, between 10 and 15 September 2026, and counted what came back. One of the four needed a second attempt before it produced anything usable: Submagic’s first run returned nothing and burned the credits anyway. It is described where it happened.

OpusClip’s account was already badged Free Trial (Pro Plan) when I started, which is why its 32 clips are not free-tier output. The source was “An Honest Conversation on AI and Humanity” from the World Economic Forum’s 2026 annual meeting — 35 minutes 27 seconds of multi-speaker panel, published under a Creative Commons Attribution licence, which is why I can show the output rather than describe it.

Tool	Clips found	Handed over	Clip length	Cost per job
OpusClip	32	32 (on a Pro-trial account)	0:21–2:35	$2.03 (Pro annual)
Submagic	31	31 shown, none exportable	0:32–1:30	$1.53 (Pro annual)
Vizard	30	10, other 20 paywalled	0:32–0:56	$0.85 (Creator annual)
CapCut	8	8 (chapters, not highlights, on Auto)	1:26–3:58	$0

Two columns need reading carefully. A number in “Handed over” means the tool showed me that many, which is not the same as letting me keep them: Submagic displayed all 31 and its trial then refused every export, while OpusClip’s 32 came from an account running a Pro-plan free trial, not from the free tier proper.

Costs are each tool’s Pro annual rate for that one job, Vizard’s equivalent tier being Creator. Those are not the cheapest rates available: Submagic’s Business tier does the same job for $1.09 and its Starter tier for $2.13, which the Submagic pricing breakdown works through tier by tier. Several are promotional rates that move, and OpusClip’s annual price changed twice in a single day while I was testing it.

One identical file is the point, not a shortcut. Clipping tools are judged on whether they find the interesting moments, and “interesting” is a property of the footage as much as of the software. Run four tools on four different videos and you learn nothing comparable; run them on one, and the differences that remain are the tools. A multi-speaker panel is also close to the hardest case in normal use, because the interesting moments are exchanges rather than monologues.

Descript I have priced from its live pages and never run. Klap I have since signed into: its trial took two uploads and paywalled the generate step both times, which the Klap review documents. Nothing in this post describes either tool’s output, because I have not seen it.

Which AI video clipping tools are worth comparing?

Tool	Meter	Free tier	Entry paid	Measured here
Vizard	per source minute	60 credits/month, renews	$14.50/mo yearly	Yes
OpusClip	per source minute	60 credits/month, renews	$15/mo monthly only	Yes
CapCut	per source minute	60 source min/month, renews	$19.99/mo US, region-priced	Yes
Submagic	flat per job	none, 15-credit trial	$12/mo yearly	Yes
Descript	media hours + credits	1 hour/month, plus 100 AI credits once	$16/mo annual	Priced only
Klap	per output clip	trial only, cannot generate	$14/mo yearly, $29 monthly	No output seen

Three of these six charge for what you feed in, one charges for what comes back, one charges the same whatever you do, and Descript bills on a mix of media hours and credits. The meter is the column that decides your bill and the one no vendor leads with. If that is the axis you care about, the Submagic alternatives roundup organises this same field by billing meter and works through which unit suits which kind of source. This post ranks them by what they actually returned. And if the destination is YouTube specifically, clip length decides more than the meter does: the AI YouTube Shorts generator comparison sets what each tool returned against YouTube’s three-minute ceiling.

What does Google’s AI Overview leave out?

Google’s AI Overview for “ai video clipping tool” recommends four products and footnotes every one of those recommendations to that company’s own website. It is what most people asking this question will actually read, and no part of it rests on an independent test.

Checked on 16 September 2026, the four are OpusClip, Vizard, Choppity and Videotto. It calls OpusClip “Great for face-tracking, virality scoring…”, Vizard “highly consistent for interviews and podcasts, often praised for preserving context and complete thoughts”, Choppity “specialized in adding animated captions, auto-reframing…”, and Videotto “built for podcasters and creators who want automatic moment discovery”. Each of those four descriptions is footnoted to that same company’s own website.

I have tested two of the four. The OpusClip line is fair. The Vizard line is fair as far as it goes, and it omits the single most important thing about Vizard’s free plan: it releases ten of the thirty it locates and charges for the remainder. I have never run Choppity or Videotto and will not tell you anything about what they produce.

The problem is not Google’s honesty. It is that the omission is exactly the fact that would change a reader’s decision, and that a summary of marketing copy is what you get when there is no independent page on the results to summarise instead.

1. Vizard — most clips found per dollar

V

Vizard found 30 clips in the panel, one fewer than Submagic and two fewer than OpusClip. On raw yield the three are level, and anyone claiming a large gap between them is not measuring the same thing I measured.

Where it separates is price. Vizard and OpusClip bill on the same unit, a credit for every minute of footage you submit, and Vizard’s Creator plan gives 600 credits a month for $29 where OpusClip’s gives 300 for the same money.

Worked through to the job, that 35-minute panel cost 85 cents on Vizard against $2.03 on OpusClip, at Vizard’s Creator and OpusClip’s Pro annual rate. Same meter, same file, 2.4 times the price. Both annual rates were discounted when I read them, Vizard’s by a 50% badge and OpusClip’s off a $29 list, so the gap depends on which promotion is running: at undiscounted list the same comparison is 2 times, not 2.4.

The ten clips it delivered ran 0:32 to 0:56, the tightest of anything here, each with a numeric score plus a written rationale and transcript excerpts.

And the catch, which is serious. Thirty is the number Vizard’s own processing screen reported. Ten is the number that reached the results page, with the missing twenty sitting behind an upgrade banner. The analysis of all thirty was already paid for out of the same credits, so what is being sold back is delivery, not compute.

Nothing else in the test behaved this way. CapCut handed over all eight, OpusClip all 32 on its Pro trial, and Submagic displayed all 31 — and while Submagic then blocked every export, it never hid a clip it had found. It is also the thing the AI Overview’s summary of Vizard leaves out.

Worth saying plainly, since it is the pick at the top of this page: Vizard, OpusClip and Submagic are the links here that earn us a commission. Vizard and OpusClip are also the two tools I spent the most time measuring, and all three are scored on the same numbers as CapCut, which pays us nothing. Neither does Descript or Klap. The paywall finding above is the kind of thing an affiliate relationship is supposed to suppress, and it is in the section anyway.

Two more things the pricing page does not mention, both of which I hit during the run:

What I expected	What actually happened
”1 credit = 1 minute of video.”	Only on the default model. A second model in the dropdown quoted 44 credits for the same file, which Vizard reads as 35:28 — a 25% surcharge nothing announces. I ran the default, so I was never billed it
Credits would be spent when clips are generated	The balance dropped from 60 to 25 the moment the upload finished, before I clicked Get AI clips

Neither is fatal and both are the sort of thing you only learn by spending the allowance. The full Vizard review has the rest.

2. OpusClip — the highest count, and the widest toolkit

O

OpusClip returned the most candidates of anything I ran: 32 clips, one every 66 seconds of source, where Submagic returned 31, one per 69 seconds. If the job is “give me every candidate”, it wins by a nose.

It also gives you the most to work with afterwards — a virality score, four letter grades per clip, and XML export to Premiere Pro and DaVinci Resolve, which I did not find in any of the other three export panels. If your clips are a starting point for a real edit rather than a finished product, that export path matters more than the price difference.

The price difference is real, though. On an identical per-source-minute meter it costs 2.4 times what Vizard does, and its list price moves enough that I watched its pricing page move between $208.80 and $174 a year across four readings in two days. Look at that page the same day you intend to pay, and the OpusClip pricing breakdown tracks what each tier actually buys.

OpusClip, measured	Result
Clips returned	32, one per 66 seconds of source
Score range	99–80, mean 88.2
Cost per job	$2.03 at the Pro annual rate
Export path	Premiere Pro and DaVinci Resolve XML

One quirk worth budgeting for. Upscaling and downloading a clip is priced separately, per clip, and it varies a lot: the six clips I opened were priced at 35, 58, 58, 70, 128 and 128 credits. The cheapest of those equals what the entire 35-minute job cost to analyse. It tracks clip length only loosely — two clips at 0:49 and 0:40 both priced at 58 — and I did not buy one, so what the upscale actually delivers is something I cannot tell you.

The OpusClip review covers the scoring in depth, and if you already use OpusClip and are weighing a move, Opus Clip alternatives sets out what each cheaper tool takes away.

3. CapCut — the only unwatermarked free export I found

C

CapCut is the only tool here whose free output you could publish as-is. I exported all eight of its clips, pulled every file to disk and probed them: 1080×1920, 30fps, no watermark on any frame I sampled. Nothing else I tested in this category gives you that for nothing.

But look at what came back. Eight clips running 1:26 to 3:58, totalling 26 minutes 5 seconds — 73.6% of the entire source, laid out sequentially with no overlap between them. The arithmetic gives the shape away: nothing that returns three-quarters of a 35-minute panel is selecting highlights from it. That is chaptering, not highlight-picking. The other three hand back thirty-odd overlapping candidates ranked against each other; Auto hands back the video in pieces. For a webinar you want split into sections, that is exactly right. For finding the twenty seconds worth posting, it is not the same product.

It is worth being precise about whose behaviour that is. I later ran CapCut’s under-60s preset against Auto on a 12-minute cut of the same panel, and coverage fell from 88.3% to 27.4% with separate picks instead of a sequence. Its three clips still came back at 1:04, 1:05 and 1:08, so the preset changes the shape without clearing the minute it names. Chaptering belongs to Auto, which is the setting the screen arrives on. The CapCut walkthrough has that test in full.

CapCut, measured	Result
Clips returned	8, running 1:26–3:58
Share of source covered	73.6%, sequential and non-overlapping
Free export	1080×1920, 30fps, no watermark
Free wall	512MB of cloud storage
Exports before refusal	4 completed, 2 failed silently, 7th refused

Two workflow traps, one of which cost me real exports before I understood it. Minimise the export dialog while a render is running and the finished file is never handed back — it exists, but only on the Space page, reachable by right-click and Download. And the export resolution silently resets to 720p on every page reload, which I confirmed across three separate exports. On a quota metered in megabytes, re-exporting at the resolution you meant to use costs the space twice. The first trap is what filled my quota; the second I caught before it cost me a file. Neither is fatal and both are learnable in one session.

Its free wall is also not where you would expect. Not a watermark, not a resolution cap: 512MB of cloud storage. Renders land there first and sit there until you pull them down, and four 1080p exports exhausted it on my account — clips 1, 2, 3 and 5 — while clips 4 and 6 failed silently along the way and the seventh was refused outright. Clear one out and the quota frees up straight away.

The CapCut review has the export testing, and CapCut pricing covers why the paid figure depends on the region you buy from.

4. Submagic — explains itself, but the trial cannot export

S

Submagic found 31 clips and explained every one of them. Every clip arrived with a paragraph arguing why it had been selected. Across 31 candidates that turns triage from guesswork into reading, and only Vizard attempts anything similar.

Its meter is the reason to choose it. Submagic bills a flat 8 credits per job where the others meter by the minute — 8 on the 35-minute panel and 8 quoted again on a 46-minute source, the longest I put through it. On a 35-minute source that works out at $1.53, more than Vizard.

The arithmetic reverses earlier than it feels like it should. Submagic’s $1.53 flat fee divided by OpusClip’s 5.8 cents a source minute puts the crossover at about 26 minutes; against Vizard’s 2.42 cents it is about 63. Those are divisions of published rates rather than a test, so they need no caveat — but they do assume the flat fee holds, and the longest source I put through Submagic was 46 minutes. The Submagic alternatives roundup shows the workings. If your recordings are long, this is the cheapest of the metered tools I measured — though CapCut is still free up to sixty source minutes a month.

Submagic, measured	Result
Clips returned	31, one every 69 seconds of source
Exportable on trial	none
Score range	97–90, mean 93.0
Cost per job	$1.53 at the Pro annual rate
Failed run	8 credits, 0 clips, nothing refunded while I watched

It also produced the only outright failure in the test. Of my two Submagic runs, the one submitted as a YouTube URL returned nothing at all — no clips, no error, and nothing refunded while I watched — and still consumed 8 of that trial’s 15. Uploading that same footage from disk, on a second trial account, produced the full 31. The lesson is about the meter rather than the tool: when a job costs the same whether it works or not, a 15-credit allowance funds one attempt and no retry.

Two things to know first. It has no free plan — only a 15-credit trial that refuses every export, so you cannot evaluate the finished output without paying. And the explanations never disagree with themselves: I read all 31 and none of the 31 calls its clip weak.

Do AI virality scores mean anything?

Every one of these tools ranks its output, and running four of them on one file exposed something none of their marketing mentions: not one of them, on any clip it actually showed me, ever indicated that clip was weak.

Tool	Scale	Range across the clips it showed me	Spread
OpusClip	0–100	99 down to 80, across all 32	19
Submagic	0–100	every one of its 31 between 90 and 97	7
Vizard	0–10	10.0 down to 9.2 — but this is the top 10 of 30, and the other 20 were never seen	0.8, four tied at 10.0
CapCut	none	a qualitative tag on 4 of 8, nothing on the rest	—

Submagic used a seven-point band, 90 to 97, of a hundred-point scale. Vizard’s ten arrived inside eight-tenths of one point of each other, four of them tied at the maximum — though that range covers only the third of its output it let me see, so it is not comparable like-for-like with the two full sets above it. OpusClip has the widest spread of the three and still never dipped below 80.

Submagic’s written explanations make the pattern clearest, because it writes one for every clip. Two clips that both scored 90 — the floor of everything it showed me — are “perfectly calibrated for sharing” and “a powerful one-two punch”. That is the same vocabulary it reaches for at 97, which is why reading the explanations tells you nothing about the ranking.

CapCut is the interesting exception: it prints no number it cannot justify, tagging half its clips qualitatively and staying silent on the rest. That is more useful than a hundred-point scale that only ever uses its top quarter, even though it gives you less to sort by.

Which AI clipping tools are actually free?

Free is a bigger question than it looks, because the four tools here mean four different things by it, and the differences all land after the processing finishes rather than in the allowance.

Five of the eight related searches Google shows for this term are about free tools and one asks specifically about watermarks, which is more weight than a paragraph can carry. Free AI clip generator is where I answered it: every free tier in this category side by side, what each one does to the file at the end, and which of them was running on a genuine free plan rather than a trial.

What about Descript and Klap?

Descript I priced but never ran, and Klap I signed into without ever reaching a clip, and they belong in the comparison because each charges on a unit none of the measured four uses — Descript by media hour, Klap by finished clip. Neither belongs in the ranking, because I have not seen a single clip either of them produced — and ranking a tool on its pricing page is exactly what the results this post is arguing with already do.

Descript is an editor that has clipping in it rather than a clipper. The transcript is the editing surface, which makes it the right answer for one specific person: someone whose clips fall out of editing they were doing regardless. It runs media hours and AI credits as two separate budgets, and neither rolls over; the Descript pricing guide works through what each tier buys. Our Descript review covers that trade in full.

Klap bills per clip handed back rather than per minute fed in. Feed a three-hour recording to a per-minute tool and you have spent 180 credits before learning whether one clip is worth posting; Klap only charges once something returns. Klap’s pricing page states the principle directly: “Only pay for clips you generate.”

Klap plan	Clips a month	Billed yearly	Month to month	Per clip, yearly / monthly
Basic	100	$14/mo	$29/mo	14¢ / 29¢
Pro	300	$39/mo	$79/mo	13¢ / 26¢
Pro+	1,000	$94/mo	$189/mo	9.4¢ / 19¢

Which AI clipping tool should you pick?

• You want the best price on a normal-length video → Vizard. It found 30 where OpusClip found 32, at 85 cents a job, and its free tier renews monthly.
• You want every possible candidate, and an edit afterwards → OpusClip. Most clips found, and XML export to Premiere and Resolve.
• You will not pay anything → CapCut, provided chapters are what you want, which is what its default Auto setting produces. It is the only clean free export of the four I ran.
• Your recordings run past about 26 minutes → Submagic, whose flat fee beats OpusClip’s per-minute meter from there, and Vizard’s past about 63. It was quoted flat from 35 minutes to 46, the longest source I put in front of it.
• Most of what comes back gets binned → Klap is the only tool here billed on output rather than input, which is the right shape for that problem. Its trial paywalled me before it made anything, so treat that as a billing observation rather than a recommendation.
• Skip all of them if your footage is already short. Every tool measured here is a search engine for long recordings, and a two-minute clip gives it nothing to search. If captions rather than clipping are the job, the Submagic review covers the tool built for that.

The final word

Tool	Found	Handed over	Keepable without paying
OpusClip	32	32 (Pro trial)	none — free and trial exports are watermarked
Submagic	31	31 shown	none — the trial cannot export at all
Vizard	30	10	10, watermarked at 720p
CapCut	8 chapters	8	all 8, clean 1080p

Three of the four landed within two clips of each other on the same video, which is not what any of their marketing would lead you to expect. What actually separates them is how much of that work you are permitted to walk away with. It also showed the thing none of them will tell you: on every clip any of them put in front of me, not one was ever marked as weak. Submagic’s 31 all landed in a seven-point band near the top of a hundred-point scale, OpusClip’s 32 in a nineteen-point band, and four of the ten Vizard released tied at its maximum. No score here substitutes for watching the clips yourself.

If you want one answer, it is Vizard — within two clips of the tools that beat it on count, at 42% of what OpusClip charges, as long as you know the free plan releases a third of what it locates.