Give it a YouTube link or a file and it comes back with finished Reels: 9:16, captions burned in, the subject held in frame instead of drifting off the edge, and a 0-100 grade on each one so your posting order is decided before you open Instagram.
Real ClipSpeedAI output — every clip below was scored, cut, reframed to 9:16 and captioned automatically from the video above it.
Source video
15:27
The recording is done. What follows is the part nobody enjoys, so it is the part we took.
Paste a YouTube URL or upload the file directly — a webinar recording, an interview, a livestream replay, a talking-head video that ran too long for the feed. Uploads on a paid plan can run to two hours. The free demo takes anything under thirty minutes, which covers most single-session recordings.
The model works through the entire runtime and marks the passages that stand on their own: a claim with a number behind it, a story that resolves, the sentence someone said twice because it mattered. Each one becomes a clip cut on sentence boundaries, reframed vertical with the speaker tracked, filler trimmed, and captions timed word by word.
Every Reel arrives with a score and a suggested title. Sort by score, pick your five, and either download them for the Instagram composer — where you set the cover frame and add audio — or connect the account and let the scheduler publish them across the week without you opening the app again.
Generic vertical output is not the same as a Reel that survives the feed. These are the differences that show.
Scrubbing a ninety-minute recording for the four good bits is the reason most people post one Reel a month rather than twelve. The model reads the whole thing and hands back a shortlist. You are reviewing candidates, not hunting.
Not a landscape clip with black bars stacked above and below it, and not a blurred copy of itself filling the gaps. The frame is rebuilt at Reels dimensions with the subject inside it, so the video fills a phone screen the way native vertical footage does.
Cut a two-person conversation down the middle of a 16:9 frame and you get a Reel of the wall between them. Tracking follows whoever is talking, and when two people trade lines quickly the split layout keeps both faces visible instead of whipping side to side.
Instagram stacks your username, the caption preview and the audio ticker along the bottom of a Reel, and the like and share controls run up the right side. Anything burned into those zones is partly hidden. Preview the clip before you export and switch style if the words sit too low — there are eleven to choose between.
Twelve Reels come back and three of them deserve a slot. The 0-100 score reads how the clip opens, how tightly it paces, and where its strongest beat falls, which turns an undifferentiated folder of exports into a ranked queue.
Pauses, restarts and verbal tics are cut automatically. On a Reel that is meant to land in under a minute, four recovered seconds is the difference between a tight opening line and a viewer thumbing past while someone clears their throat.
Your fonts, colours and logo applied across every export, so a profile visitor scrolling your Reels tab sees one account rather than a dozen unrelated experiments. Consistency does more for a follow than any individual clip does.
The same vertical file belongs on three platforms. Connect the accounts once and queue the batch in a single sitting, which is how a Tuesday afternoon of clipping turns into three weeks of scheduled posts.
Connect YouTube, Twitch or Kick and clips are produced while the broadcast is still running. A moment can be a Reel within minutes of happening, rather than after you have downloaded a VOD the following morning.
ClipSpeedAI runs as an MCP connector, so you can ask Claude for Reels from a link and get finished files back in the conversation. The same engine sits behind a developer API if you would rather script it.
Your YouTube audience and your Reels audience barely overlap. Republishing the strongest three minutes of a twenty-minute video reaches people who would never have clicked the original. The YouTube Shorts maker covers the same source from the other end.
A client call or a workshop recording holds several complete answers, each of which is a Reel that demonstrates competence rather than announcing it. Those clips convert better than anything written about your services.
Instagram is where podcast discovery happens for a lot of audiences now, and it wants vertical with captions. See the podcast clip generator for the episode-shaped version of this workflow.
Founder explanations, product walkthroughs, behind-the-counter footage — you already film more than you post. The bottleneck is editing, and removing it usually multiplies output rather than improving any single clip.
A self-contained explanation lifted from a lesson works as a Reel because it teaches something in forty seconds. It also functions as a free sample of the paid material.
Volume across a roster is the job. The API and the Brand Kit mean per-client styling can be set once and applied without an editor opening each file.
A Reel is displayed at 9:16 on a phone, but the whole rectangle is not usable. Instagram draws your username, the caption preview and the audio attribution over the lower portion, and the like, comment, share and menu controls run up the right edge. Anything you burn into those areas competes with the interface. This is the most common reason a Reel exported from a landscape source looks amateur even when the footage is good — the subtitles were placed against the bottom of the video, and half of them now sit behind an interface element.
There is a second crop to think about. Your profile grid does not show the full vertical frame; it shows a taller-than-wide slice taken around the middle. A face parked near the top or bottom edge of the frame can be cut out of the thumbnail entirely, which matters because the grid is what a visitor judges before deciding to follow. Tracked reframing tends to keep the subject nearer the middle by default, and that happens to be exactly what survives the grid crop.
The practical version: watch the preview on a phone, not on a laptop. If the words are legible with the interface over them, and the face reads in the middle third, the Reel is built correctly. Everything else is content.
Instagram lets you pick any frame from the Reel as its cover, or upload a separate image. We do not generate a cover image for you, and we do not pretend to — that is a design job with a brand attached to it, and an automatically chosen frame is usually someone mid-blink. What we do is make the choice easier, because clips start on a sentence boundary rather than in the middle of a word, so the opening second is generally a composed shot of someone about to speak instead of a stray gesture.
When you do choose one, the rules are boring and reliable. Pick a frame where the eyes are open and pointed roughly at the lens. Avoid the frames where a caption word is mid-animation, since a half-drawn word reads as a glitch at thumbnail size. If your niche rewards it, a cover with a face beats a cover with a product almost every time.
The other thing we deliberately leave alone is audio. We do not add trending sounds, because we do not have a licensed audio library and quietly attaching one would be dishonest about what the file contains. If a particular Reel needs a trending track, publish that one by hand through the Instagram composer and add it there. For the rest, the spoken audio and the captions carry it, and that is how most talking-head Reels perform anyway.
Every clip leaves as an MP4 with the captions burned into the pixels rather than carried alongside as a separate file. That is deliberate for this platform: there is no subtitle track to switch on inside Reels, sound-off viewing is ordinary in that feed, and text that depends on a player setting is text a silent viewer never receives. It also means the words survive a repost — someone downloading your Reel and putting it up elsewhere takes the captions with them whether they meant to or not.
The frame is 1080 × 1920. If the same moment also has to serve a square grid post or a landscape upload somewhere else, re-render it at 1:1 or 16:9 rather than letting another app crop the vertical file down — the framing is recalculated for the new shape, which is not the same thing as a vertical clip being trimmed at the sides by whatever container it landed in.
From there the file goes out one of two ways. The first is download: you carry it into the Instagram composer yourself, which is where the cover frame gets chosen, the caption body written, a trending sound attached if that Reel wants one, and the decision made about whether it also lands on the main grid. The second is the scheduler: connect the account, drop the clip on a date, and it publishes without you being awake for it.
Those two routes are not equivalent, and it is worth knowing which decisions live on which. What the scheduler hands Instagram is the exported video and its text. Anything the composer owns — the cover image, the trending audio — is unavailable on the automated path, so a sensible working pattern is to let the calendar carry the clips that need neither and hold back the one or two that do.
One constraint on the scheduled route belongs to Instagram rather than to us: it runs through their professional-account publishing tools, and a personal account cannot be published to programmatically by any service. Switching to a creator or business account takes about a minute in the Instagram settings and costs nothing. If you would rather not switch, downloading the files and uploading them by hand works from any account type and loses you nothing except the calendar.
Say you ran a 45-minute product walkthrough through it and got nine clips back, of which four are not going anywhere near your grid. Working out which of the four failures you are looking at is quicker than re-running the whole job, because three of them are a setting and only one of them is the footage.
The first is captions you cannot read on a phone. This is nearly always contrast rather than size — pale type over a bright kitchen, a window blowing out behind the speaker's shoulder. Switch to a caption style with a solid backing plate behind the words instead of reaching for larger text, because bigger letters in the same colour are still invisible against the same background.
The second is an ending that lands flat. Cuts are placed at sentence edges, and a sentence edge is not automatically a good place to stop. For example, an answer that trails out with "and yeah, that is pretty much it" finishes a sentence and kills a Reel in the same breath. Pull the out-point back to the previous full stop and re-render — both versions are legal cuts, and the shorter one almost always ends stronger.
The third is framing that hunts. A speaker who rocks in and out of shot, or a second person who wanders into frame halfway through, gives the tracker more to follow than you want it following. Two-person footage belongs in the split layout, and anything shot in a room where people move around belongs in a wider static layout. Tracking is the right default; it is not the right answer everywhere.
The fourth is a batch that comes back thin — three clips out of a long recording when you were expecting ten. That is usually the source, not the model. Sessions that are mostly logistics, screen-share and two people agreeing with each other contain very few passages that stand up alone, and inflating the count would only hand you clips that die in the feed.
The first is bars. A landscape clip dropped into a vertical container with black or blurred strips above and below it announces its origin before the first word is spoken, and it hands away more than half the screen on a device where screen space is the only currency you have.
The second is starting cold. A Reel that opens halfway through a sentence gives a viewer nothing to hold onto in the half-second they are deciding, and that half-second is the entire contest. Cutting on sentence boundaries fixes the mechanical version of this problem; choosing a moment whose first line makes sense on its own is the part still left to you.
The third is length as a habit rather than a decision. People trim to a duration because it feels tidy, which produces Reels that stop before the payoff or drag past it. Let the thought decide where the clip ends, then check whether the last three seconds are earning their place.
The fourth is inconsistency across the grid. Ten Reels in ten different caption styles with three different logo positions read as ten experiments rather than one account, and a profile visitor deciding whether to follow is judging the set, not the clip that brought them there.
The last one is subtler. If every Reel you post is an excerpt from something longer, the account starts to feel like a trailer for content that lives elsewhere. Mixing in something filmed vertically and made only for Instagram keeps the profile from reading as a feed of leftovers.
Reels reach is decided per post. A weak one costs you almost nothing, and a strong one can reach people who have never heard of you regardless of how small your following is. That shape rewards attempts. An account posting five Reels a week with two duds is in a materially better position than an account that publishes one polished Reel a month.
Understanding that has never been the problem. Sitting down to cut, reframe, caption and export twenty clips by hand is the problem, and it is a problem that gets worse the more you record. Automating the mechanical part changes the constraint from your evening to your judgement about what to post, which is the constraint you actually want to be operating under.
One caveat worth stating. Volume only compounds if the clips are individually watchable, and a clip that opens mid-thought or frames a shoulder instead of a face is not watchable. That is why the cutting and framing decisions on this page get more attention than the throughput does.
Automation is not the only road to a Reel, and knowing what the other roads cost is how you avoid overpaying for the wrong one. The cheapest alternative is also the one people skip: if a piece was always destined for the feed, film it vertically to begin with and the reframing question never arises. Nothing here beats footage that was composed for the frame it will be watched in.
Editing by hand in Premiere, Resolve or CapCut has the highest ceiling and the worst throughput, and both of those facts matter. Imagine you have four Reels to make this month and a firm opinion about every frame of them — hand editing wins outright, and no shortlist will out-taste you. At twenty a month it stops being a choice and starts being your evenings, which is the point at which people quietly go back to posting once a fortnight.
Hiring an editor moves the labour rather than removing it, and it prices per clip, so it gets more expensive exactly as your strategy asks for more volume. The split that works for most people is to route the ordinary weekly output through automation and spend the editor budget on launches, paid ads and anything where a human on the timeline visibly changes the outcome.
Other AI clippers cover the same ground and several are good at it. Take the common case of a weekly show that goes out live and then sits as a VOD: most tools can only start once the recording exists, whereas clipping during the broadcast means the moment is postable while people are still talking about it. If you never work from live sources that difference is worth nothing to you, and you should pick on framing and cut quality instead — the face-tracking crop and viral score pages go deeper on both than a comparison table can.
The last alternative is doing none of this and posting the long video only. That is a real option, and for a small number of accounts it is correct — a channel whose audience arrives through search and stays for depth may gain nothing from a Reels tab. Worth being honest with yourself about which one you are before you build a clipping habit you resent.
Observations from operating the pipeline in production — not general advice.
Each of the eleven caption styles pins its text block at a different height inside the 1080 × 1920 frame, and the renderer holds no model of where Instagram will later paint the username row, the caption preview and the audio ticker. Nothing warns you when a style collides with that strip, because from the engine's side nothing has gone wrong — the pixels sit exactly where the style specifies. In practice that makes the preview the step worth slowing down on, and swapping style the actual fix, rather than trying to nudge text after a clip has already rendered.
What goes over the wire on a scheduled Reel is the exported video and its text, and that is the whole payload. No cover image travels with it and no soundtrack does either, which is why a trending audio pick has to happen inside the app on that specific post. The consequence is a split queue rather than a broken one: the clips that need neither go on the calendar and publish themselves, and the two or three that want a particular cover frame or a particular track get pulled out and handled manually.
Cut points are read off a per-word transcript and placed at sentence edges, so the first frame of a Reel is whatever the speaker's face happened to be doing in the instant before their first word — not a frame anybody selected. That is not cover-frame generation and we do not present it as such. What it changes is where the usable frames cluster: they bunch into the opening second, before the hands have started moving, which is why scrubbing for a cover on these exports is a short job and the same task on a clip cut at an arbitrary timestamp is not.
The in-app editor is fine for something filmed vertically on a phone in the last hour. It is a poor place to work from an hour-long landscape recording: there is no way to search the footage for the good part, and reframing means manually dragging a crop box across a video you have to watch in full first.
CapCut will do a beautiful job of any clip you have already identified, and its templates are genuinely good. The gap is upstream — nothing in that workflow tells you which forty seconds of the recording to drop into the template, and that decision is where the hours go.
Per-clip pricing makes volume expensive precisely when volume is the strategy. A sensible split is to run the routine output through automation and save the editor for launches, ads and anything where a human eye on the timeline changes the result.
Most handle the same core job on uploaded video. Two things here are less common: clips are produced from a live broadcast in real time rather than after the recording is available, and the whole thing is drivable from Claude or your own code. If you only ever process finished files, judge it on framing and cut quality — run one of your own videos through it and look.
Not a preview on a laptop — a real Reel on your own profile, opened on a phone the way everybody else will meet it. If a line of caption is hiding behind the username row or the audio ticker you will see exactly which line, and the fix is a different style, not a re-edit. Then tap back to your grid and check the thumbnail crop while you are there.
⚡ Make My Reels — $1 trial
