Klap

Short-form video rewards frequency more than quality, which is an uncomfortable fact for anyone who makes long-form. Posting daily is the mechanism, and nobody produces original daily content indefinitely.

The obvious answer is to cut what you already have. The obstacle is that it takes an hour to find, trim, reframe and caption three clips from one recording, which is exactly as long as making something new.

Klap compresses that hour, and the part that actually determines whether the output is usable is what it does to the frame.

Klap: Long Videos Into Vertical Clips

• A file or a URL in, vertical clips out for TikTok, Reels and Shorts.
• AI Reframe 2 analyses scenes and applies a layout — split screen, screencast, gaming — rather than cropping centrally.
• Roughly five clips per minute of source, which is the site's own stated rule of thumb.
• 8.5M clips by 3.5M creators claimed, with AI captions generated alongside.
• ZIGG SAS, trading as Klap, built by two named founders, with an API and a Discord.

How Many Clips Come Out Of One Video?

Klap states a rule of thumb: a one-minute video produces about five clips, and that turning long videos into many vertical clips is the point of the product.

That ratio is worth reading carefully, because it is a claim about candidates rather than about usable output.

Five clips per minute of source implies significant overlap — the same passage cut at different lengths and start points — which is the correct behaviour for a candidate generator and a poor basis for posting everything.

The number that matters is how many you would actually publish, and in practice that is far lower. A forty-minute podcast might yield six genuinely good moments regardless of how many clips a tool can extract from it.

That is not a criticism of the tool so much as of how these products get used. Volume is available and cheap; the discipline of publishing only the good ones is what separates a channel that grows from one that posts a lot.

A forty-minute podcast has about six genuinely good moments in it, regardless of how many clips a tool can extract.

AI Reframe And The Layout Problem

Klap describes AI Reframe 2 as analysing your scenes and applying the perfect layout — split screen, screencasts, gaming and more — to resize video for any platform.

This is the feature that decides whether clipped output is watchable, and it is the one worth evaluating above all others.

The problem is geometric. A 16:9 recording cropped to 9:16 loses roughly two thirds of the frame, and a centre crop is almost always the wrong two thirds — an interview becomes a shot of the space between two people.

Scene-aware layout selection is the right answer because different footage needs different treatments. Two people talking wants split screen; a screen recording wants the screen large with the presenter inset; gameplay wants the action centred with the face small.

Getting that choice right per scene rather than per video is the hard part, since a single recording often contains all three situations. It is the same job Kapwing approaches with Speaker Focus, and the quality of the automatic choice is what you should test first.

Virality Scores And What They Mean

Klap states that it will instantly turn each clip into viral-worthy content and score its potential to go viral, adding that you should trust the algorithm.

Scoring is the weakest claim on the page, and it deserves scepticism rather than dismissal.

What a model can plausibly detect is real: whether a passage is self-contained, whether it opens with something arresting, whether the speaker's energy rises, whether there is a clear payoff. Those correlate with clips that perform.

What no model can predict is what actually determines virality — timing, what else was published that hour, whose feed it landed in, and the specific unpredictability of recommendation systems. Two identical clips posted a week apart perform completely differently.

So read a score as a proxy for structural quality rather than a forecast. Used that way it is genuinely useful for ranking a batch of candidates; treated as a prediction, it produces false confidence and a lot of posts nobody watched.

The same caution applies to any automated selection. Clip pickers, thumbnail generators and headline scorers all rank candidates plausibly and cannot see what an audience will actually do — which is why tools such as Quso.ai are best judged on how well their shortlist matches yours rather than on the confidence of the ranking.

• Test the reframing before anything else — it decides whether output is watchable.
• Publish the good clips, not all of them; volume is available and cheap.
• Read virality scores as structural quality, not as a forecast.
• Fix every clip's start point — automatic ones almost always begin too early.
• Read the auto captions; names and jargon are reliably wrong.

Captions, Posting And The Whole Loop

Alongside clipping, Klap states that it generates AI captions to post with the videos, and that you can share new videos across social platforms once they are made.

Captions are non-negotiable rather than a nice extra. A large share of short-form is watched without sound, and an uncaptioned clip is simply not viewed by that audience.

Auto-captions get the common words right and the important words wrong. Names, products, technical terms and anything domain-specific are exactly what a general speech model has least reason to know, and they are exactly what your clip is about.

The posting integration closes a loop that is more valuable than it sounds. Clips that get exported and sit in a folder do not get posted, and the friction between finished and published is where most repurposing efforts quietly die.

The caution about automated posting is the usual one: publishing to several platforms simultaneously means a mistake reaches all of them at once, so a review step before anything goes out is worth keeping.

The other half of the equation is having enough source material to clip in the first place, which is a production problem rather than a repurposing one. Generators such as Hypernatural sit on that side, and a channel short of long-form has the opposite constraint from one drowning in it.

What Clipping Cannot Manufacture

There is a limit to repurposing that no tool addresses, and it is worth being direct about it.

A clip cannot be more interesting than the moment it came from. If a recording has no arresting passages, clipping produces short boring videos rather than long ones, faster.

Context is the second casualty. A remark that works because of ten minutes of setup often reads as banal or wrong when extracted, and automated selection cannot detect that the setup was the point.

Format matters too. Some content is structurally unsuited to vertical short-form — anything requiring sustained argument, anything visual that needs width, anything where the value accumulates — and forcing it produces clips that technically exist.

The honest use is as an amplifier for material that already contains good moments. Long-form generators such as Hypernatural address the opposite problem, which is having nothing to cut from in the first place.

Getting Clips Worth Posting

The failure pattern with clipping tools is generating forty clips, posting thirty of them, and concluding that short-form does not work.

This sequence treats the output as candidates and spends the saved time on selection.

1. Run one representative video and judge the reframing, not the clip count.
2. Watch every candidate on a phone, muted, which is how they will be seen.
3. Cut the list to the ones that would stop you scrolling — usually far fewer than generated.
4. Move every start point later; automatic clips almost always begin too early.
5. Read the captions and fix names, products and jargon by hand.
6. Keep a review step before anything posts to several platforms at once.