Bring in a lecture, a webinar recording or a finished module. The AI listens to the whole session, lifts out the explanations that stand on their own, cuts each one on a sentence boundary, burns in word-by-word captions, and grades every clip 0-100. You get short lessons you can post publicly, embed in a module, or send to a cohort — without opening an editor.
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
You do the first step. The engine does the rest and hands back a shortlist.
Paste the URL if the lecture is on YouTube, or upload the file straight from the folder your recordings land in. Zoom exports, screen recordings, camera footage of a seminar room and webinar replays all work. Paid plans take up to two hours per upload, which covers most single sessions; the free demo handles anything under thirty minutes so you can test the output on one module before committing.
Teaching is not evenly interesting, and it is not supposed to be. Most of a lecture is scaffolding: setup, admin, a worked example that only makes sense after twenty minutes of context. The engine reads the full transcript looking for the passages that survive being lifted out — a definition given cleanly, a misconception corrected, an analogy that lands, a worked answer to a question a student actually asked.
For slide-driven teaching this is the step that matters. A screenshare of a 16:9 deck cropped to a 9:16 phone frame loses the outer thirds, which is exactly where your bullet points and axis labels tend to live. The screenshare layout keeps the deck whole and puts your camera feed beneath it, so a viewer can still read the slide. Flip layouts per clip if a particular moment is better as a straight talking head.
Sort by the 0-100 grade and take the top handful for social. The rest are not waste: a two-minute explanation of one concept is a legitimate asset inside the course itself — a recap tile at the end of a module, an answer you paste into the student forum for the fourth person who asked, a warm-up you send the night before a live call. Download them clean or schedule them out to TikTok, Reels and Shorts.
Lecture recordings break automatic editors in specific, repeatable ways. These are the parts that deal with that.
Chopping a lecture into equal ninety-second blocks produces nothing usable, because the boundaries fall wherever the clock says. The model selects around a complete idea instead, so a clip contains the whole answer rather than the middle of one.
A deck is horizontal and a phone is vertical, and something has to give. The screenshare layout stacks the slide above your webcam feed rather than cropping into the slide, which is why a chart with labels on the left edge stays legible instead of being sliced off.
A long module can yield fifteen candidate lessons and you have room to post three this week. The grade ranks them on how the clip opens, how it paces and where its strongest beat sits, which turns a folder of exports into an obvious publishing order.
Educational short-form is watched on a commute with the sound off more often than anyone likes to admit. Words highlight in time with your voice, burned into the frame in one of eleven styles, so nothing has to be re-uploaded as a sidecar file or re-synced after a trim.
Live teaching is full of thinking time — the pause while you find the right slide, the false start, the sip of water. Those are fine in a lecture and fatal in a ninety-second clip. They come out automatically, which usually recovers several seconds and tightens the pacing more than any other single change.
The specific failure this prevents is a clip that begins halfway through the sentence defining the term it is about. Boundaries land at the start and end of a thought, so a learner is not dropped into the back half of a clause with no idea what the pronoun refers to.
Interview-format lessons and guest lectures put two faces on screen. Tracking follows whoever is talking, and split-screen keeps both people visible when the exchange itself is the content, instead of centring on the empty desk between them.
Connect a YouTube channel and clips are cut during the broadcast rather than after the replay processes. For a live cohort call or an open teaching session, that means the answer to the best question of the hour is postable before the session has even ended.
The Brand Kit carries your fonts, colours and logo across every clip, so a lesson posted in March looks like it belongs with one posted in August. For anyone selling a course, that visual continuity is doing quiet credibility work in the feed.
If you have sixty recorded modules sitting in a drive, clicking through them one at a time is its own project. There is a developer API behind the same engine, plus an MCP connector so you can ask Claude to work through a list. Start at the API docs.
A recorded seminar reaches the students who attended. A four-minute clip of the clearest ten minutes in it reaches next year's applicants.
Free micro-lessons are the most honest advert a paid course has, because the buyer is judging your actual teaching rather than a sales page.
A compliance or onboarding session recorded once becomes a set of two-minute refreshers that new starters will genuinely watch.
The Q&A at the end is usually the strongest material in the recording, and it is always the part nobody bothers to cut. See the webinar clip generator.
One clean walkthrough of a question type that trips everybody up gets sent between students for years. That is the clip worth finding.
Code walkthroughs and software demos live or die on whether the screen is readable, which is a layout problem before it is an editing one.
The first job is acquisition. A clip goes out on a public feed to someone who has never heard of you, and its only task is to demonstrate that you can explain something clearly. Nothing sells teaching like teaching. A prospective student watching you resolve a confusion they personally have is a stronger pitch than any testimonial, and it costs you nothing beyond the recording you already made.
The second job is retention, and it happens inside the course. Completion rates on long video modules are grim, and the usual reason is not that the material is bad — it is that a learner who missed one idea has no cheap way to go back and get it. A library of short, single-concept clips gives them that. Pinned in the module, dropped into a reply on the forum, sent the day before a live call.
Both jobs come out of the same source footage, but you sort them differently. The public ones are chosen by grade and hook strength. The internal ones are chosen by curriculum coverage — you want one clip per idea that reliably confuses people, and you do not care at all whether it opens with a hook.
Conversational footage is easy to cut because the participants keep resetting each other. Somebody asks a question, somebody answers it, and the answer is naturally self-contained. Teaching does not work that way. A lecture builds, and the payoff at minute fifty depends on a definition given at minute four. Lift the payoff out on its own and it can be genuinely incomprehensible.
That shapes what a good clip looks like here. The best candidates are the moments where you restate — the summary, the analogy, the answer to a student who asked you to say it again differently. Those passages were built to be understood without the preceding forty minutes, because that is precisely why you said them. The engine is looking for that quality, not for a peak in the waveform.
The second complication is visual. Most teaching recordings are a slide deck with a small camera feed, and the informational content is in the deck. Any tool that treats the frame as a face to be centred will crop the thing the clip is about. The layout decision is not cosmetic here; it decides whether the clip means anything.
Posting the introduction. The opening minutes of a recorded session are housekeeping — where the reading list lives, what is due on Friday, an apology for the microphone. It is the easiest segment to reach and the least interesting thing you said all hour, and it is a large part of why educator accounts so often look dull from outside.
Leaving the clip too long. Ninety seconds of a genuinely good explanation outperforms four minutes of the same explanation with its preamble still attached. If a segment needs a run-up before it makes sense, the run-up is probably the clip and the thing after it is not.
Assuming the viewer shares your vocabulary. Someone who never enrolled is watching, so a term you defined in week two arrives as noise. The passages that travel furthest tend to be the ones where you happened to define your terms as you went, which is worth noticing about your own teaching.
And reading the grade as a verdict on the material. It describes how a clip behaves in a feed. A slow, careful walk through a genuinely hard idea will often land in the sixties and still be the most valuable thing you publish that month.
Upload the same day you record, while you still remember how the room reacted. That memory decays fast, and by Friday you will not recall that the analogy in the middle third was the one that visibly landed.
Review in grade order but overrule it without hesitation. Take the top eight, discard the three that lean on context you never gave on camera, and five realistic candidates remain. That ratio is fairly stable across teaching footage.
Separate public from internal before doing anything else, because they need different titles. A public clip is titled for a stranger who has no idea what the module covers. An internal one is titled for a learner who already does, and writing both from the same instinct produces two mediocre titles.
Then schedule instead of posting. Two clips a week drawn from one recording beats five on Monday followed by a fortnight of nothing, and putting them on a calendar is what makes that happen without you having to decide again each morning.
Layout is the setting that changes most on this kind of footage, and it is worth choosing per clip rather than once for a whole batch. Screenshare when the deck carries the information, a plain vertical crop when you have stepped away from the slides and the clip is just you talking, split-screen or picture-in-picture when a guest is answering and their reaction is half the point. One module recorded in a single sitting will usually want two of those three across its clips.
Aspect ratio follows the destination rather than taste. 9:16 for Shorts, Reels and TikTok; 1:1 when the clip is going into a feed post or an email; 16:9 when it is being embedded back onto a course page, where a vertical video sitting in a desktop layout reads as a mistake. Vertical renders at 1080x1920, which is what every short-form platform re-encodes towards in any case, so exporting anything larger buys you nothing.
Filler and silence removal runs on every clip rather than being something you opt into. Here's why that matters more on a lecture than on a conversation: a monologue holds far more removable dead time, so the same pass that shaves a second off an interview can visibly change the tempo of a lecturer. The one case to watch is the deliberate pause before a punchline in the material — the pass cannot tell timing from hesitation, so if a beat you wanted has gone, widen the in and out points and re-render.
Caption style is a legibility decision on technical subjects, not a branding one. The heavier styles with a solid backing plate stay readable over a busy slide; the thin outlined ones disappear against a white deck. Choose once, put your fonts and colours into the Brand Kit so the choice carries across every export, and stop deciding it clip by clip.
Length is fitted to the thought rather than to a number you set, and you can drag the in and out points afterwards and re-render as often as you like. On teaching footage the most common adjustment by a wide margin is pushing the out point one sentence later so the clip closes on the summary instead of on the aside that followed it.
The slide is cropped and the clip no longer makes sense. That one rendered on a face-centred layout rather than screenshare. Switch it and re-render. The arithmetic is unforgiving: a 16:9 deck placed inside a 9:16 frame keeps slightly under a third of its original width, and what disappears is the outer edge on both sides — which is where axis labels, legends and left-aligned bullets almost always sit. Nothing downstream can rebuild a label that was never in the exported frame.
The clip opens on a pronoun. Cutting on a sentence boundary guarantees you start at the beginning of a sentence, not that the sentence stands up alone. "So that gives us the second condition" is a perfectly legal opening line and a useless one. Pull the in point back until the noun the sentence leans on is inside the clip, which on teaching footage usually means one more sentence rather than a few seconds of room tone.
The Q&A produced fewer clips than the lecture did. The usual cause is that the questions were asked off-mic. Your answer is captured and the thing it answers is not, so the passage reads as incomplete to the selector and to a viewer alike. Repeating the question back before you answer it is a four-second habit that converts the strongest part of most sessions into material that can actually be lifted out.
A segment where you read the slides returned nothing. Bullets spoken verbatim transcribe into a list, and a list has no argument in it — there is no claim, no turn and no resolution for the model to select around. The clippable passages are the ones where you leave the slide and explain why the bullet is true. If a whole module comes back thin, that is usually what the transcript shows.
The framing wanders on a lecture-theatre recording. A camera fixed at the back of a hall hands the tracker a small, distant face against a large static background, and below a certain apparent face size no tracker holds steady. A static crop is frequently the better call for that footage. The durable fix is a closer camera rather than a different setting.
The division of labour is not "the AI edits and you approve". The engine does search and mechanics: reading a two-hour transcript end to end, finding the passages that survive being lifted out, placing boundaries on complete thoughts, reframing, captioning and grading. That is the part that eats an evening and produces no judgement at the end of it.
Most people assume the expensive part of clipping is the cutting. On teaching footage it is the watching. A 90-minute seminar costs 90 minutes to review before an editor is even open, and you will do that once, enthusiastically, and then never again — which is why so many educators who intend to clip their back catalogue still have an unclipped back catalogue.
Editing by hand still wins in three places. A flagship promo where the opening two seconds deserve an hour of your attention. A clip that needs material from two separate sessions cut together. And anything whose meaning lives on a whiteboard or in a physical demonstration, because selection reasons over the transcript and the transcript does not contain your hands.
The realistic pattern is a graded shortlist first and a human pass second. Fifteen candidates come back captioned and ranked, you overrule that order using what you know about your own cohort, and your editing time goes on the three clips being published rather than on the search that found them.
Chaptering the full recording. Cheap, and it does a different job — it helps people who already found the video rather than reaching people who never will. If a lecture is already picking up search traffic on YouTube, an hour spent on chapters returns more than an hour spent on clips.
Publishing the transcript as an article. Teaching prose indexes well and answers the exact question somebody typed into a search box. It is slower to produce and it will not travel in a feed, but it keeps earning for years where a clip has a half-life measured in days. Running both out of one recording is barely more work than doing either alone.
An audiogram — a waveform over a static slide. Free to produce and reliably ignored, because a short-form feed is a visual medium and a still image loses to a talking head every time. Worth knowing that this is the format most institutions default to, and it is a large part of why so many university accounts post consistently and reach nobody.
Re-recording the idea to camera in ninety seconds. Frequently the best option and almost nobody takes it, because it feels like extra work when a recording already exists. If one concept keeps producing weak clips across several sessions, that is the signal to stop clipping it and record it deliberately once.
And a human editor for the two or three videos a year that carry a launch. Automation earns its keep on the long tail; the flagship is worth a person.
When the session itself was flat. Everyone has recordings where the room never warmed up and the explanation never quite arrived, and clipping does not repair a delivery problem — it distributes one. The grade is honest about this: if the strongest clip out of ninety minutes lands in the fifties, the ceiling was set in the room rather than in the edit.
When the material is sequential the whole way through. A derivation that builds for forty minutes without a single restatement contains nothing that stands alone, and forcing a clip out of the middle of it produces something that reads as broken to a stranger. Sessions like that want chapters, not clips.
When the content is confidential. Internal training, anything with a client name spoken out loud, a cohort call where a student described a real problem — a clip is a public artefact and a teaching recording usually is not. The trap is that the most engaging ninety seconds in a cohort call is very often the exact passage you cannot post.
And when you have not decided where the clips are going. Twenty exports in a folder are not a distribution plan, and by the time that becomes obvious the effort has already been spent. Say you record weekly and clip monthly: decide the destination before the clipping session, even if the destination is one pinned post inside the module.
It will not rescue a recording with no speech. If a segment is silent screen capture with music over it, there is no sentence for the cutter to end on, and no transcript for captions to be built from. Narrate over your screen recordings and the whole pipeline improves.
It will not follow a whiteboard. If you spend two minutes writing an equation while saying "and this term here cancels", the words are captured but the referent is not — that clip will read as gibberish to anyone who was not watching your hand. Those moments are better re-recorded than clipped.
And it does not generate anything. There is no AI narration, no synthetic footage, no invented B-roll. Every frame in the output came out of your recording, and every word in the captions came out of your mouth. For anyone whose credibility rests on being accurate, that constraint is a feature.
Observations from operating the pipeline in production — not general advice.
The strongest candidate in a lecture is rarely the first time you explain something, because that version is standing on the forty minutes in front of it. The clippable one is the restatement — the summary, the analogy, the answer to a student who asked you to say it another way. Those passages were built to be understood without the run-up, which is precisely the property a clip needs. Selection reads the transcript rather than the audio, so a quiet recap can outrank the liveliest moment in the room.
Two people talking regulate each other; one person presenting to a deck does not. What comes out of teaching audio is slide-hunting, the pause while a question is parsed and the breath before a correction. The same removal pass over a two-person interview recovers far less, which is why stripping filler and silence changes how a lecturer sounds more than any other single setting on the page. The content is identical and the delivery is not.
General speech comes back clean. The errors concentrate in the vocabulary a specialist uses without thinking about it: surnames, acronyms spelled out loud, units, and notation read as words instead of symbols. In practice the same handful of terms fails repeatedly across one module rather than scattering at random, so checking those specific words on the clips you intend to publish is far cheaper than proofreading every caption track you generate.
One moment often deserves two renders. A public clip has to survive a scroll, so its front edge belongs on the sharpest sentence with the context arriving behind it. An internal clip is watched by someone who already decided to watch it, and it works better opening on the question it answers, because a learner going back for an idea they missed needs the framing more than the hook. Re-rendering the same moment on a different boundary costs nothing.
You can absolutely do this by hand, and for a flagship promo you probably should. The problem is scale: reviewing a ninety-minute recording to find the good ten minutes is where the evening goes, and the editing itself is the quick part. Automating the search is worth more than automating the cut.
Sensible for a launch, expensive as a habit. An editor also cannot tell you which explanation in your own module is the one students need most — that judgement is yours, and it is much easier to make when fifteen candidates are already sitting in front of you graded and captioned.
Kajabi, Teachable and the rest let you trim a video and nothing more. No vertical reframe, no burned-in captions, no ranking. They are storage and delivery tools, which is a different product from a thing that decides what to cut.
Most handle a finished upload competently. The differences that matter for teaching footage are the screenshare layout and real-time clipping of a live class, which few of them offer. If you only ever clip recorded modules, judge on whether the moment selection matches your own — run one lecture through and see. Try it here.
Send one module through and count the keepers. If two explanations come back clean enough to post in public, you have an acquisition channel hiding inside footage you already paid to record.
⚡ Clip My Lecture — $1 trial
