Upload a session, a form breakdown, a recorded coaching call or a training podcast. The AI finds the parts where you actually teach something, cuts them on sentence boundaries, burns in captions loud enough to survive a gym floor, and grades every clip 0-100. You choose the layout, so a deadlift stays in frame from bar to eyeline.
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
Film once. The sorting and the cutting stop being your problem.
Drop in the file — a filmed session, a form breakdown you recorded on a tripod, a Zoom coaching call, a long-form training video from your channel. If it is already on YouTube, the link alone is enough. Paid plans accept up to two hours per upload, which covers a full client call or a complete workout walkthrough. The free demo takes anything under thirty minutes.
It reads what you say. The cue you give when a client's knee caves, the reason you programme tempo work, the thirty seconds where you explain why a lift feels heavy on day four of a block — those are the segments that lift out cleanly and get saved by strangers. Sets performed in silence are not, and nothing pretends otherwise.
This is the step to actually pay attention to. Demonstration footage usually wants the fit layout so the entire lift stays visible; a piece to camera wants the tracked vertical crop. Choose per clip, check the grade, and schedule the winners to Reels, TikTok and Shorts. Your logo and colours come from the Brand Kit.
Gym video has two properties that defeat generic clipping tools: the subject is a whole body, and the room is loud. Here is how each is handled.
A crop that zooms to a face is exactly wrong for a squat. Bar path, hip depth and foot position all live outside a head-and-shoulders box, and cropping them away turns a technique clip into a video of somebody grimacing. The fit layout holds the full frame inside a vertical canvas so the lift stays readable.
The clips that build a coaching audience are the ones where you name a mistake and fix it. Selection runs on the transcript, so what it hunts for is the sentence where you tell a lifter their lockout stalls because the bar drifts in front of the shoulder. The loudest ten seconds of a training session is usually a plate hitting the floor, and nobody saves that.
A ninety-minute coaching call can produce twenty clips and you have three slots this week. The grade ranks them on how the clip opens and where its strongest point falls, so the decision is made in thirty seconds instead of by rewatching everything.
Your audience is watching between sets with headphones in one ear or nothing at all. Word-by-word captions in eleven styles are burned into the frame, so a cue lands even with the sound off, and nothing needs re-syncing if you trim the clip afterwards.
Real training footage is mostly waiting. Ninety seconds between sets, the walk to the rack, the fiddling with a clip. Silence and filler words come out automatically, which is often the difference between a clip that holds and one that dies during the plate change.
When you are coaching a client on camera, tracking follows whoever is speaking and the split layout keeps both of you visible when the exchange matters. A fixed centre crop on a coach-and-client shot frames the floor between you, which is the single most common way gym footage gets ruined vertically.
Cuts land at sentence boundaries. A form clip that starts with "and that is why it rounds" is useless — the fix has already been said. Starting at the beginning of the thought is what makes a thirty-second clip teach anything.
Stream a Q&A or a technique session to YouTube, Twitch or Kick and clips are produced during the broadcast. The best answer of the hour goes out while people are still in the chat asking about it, rather than tomorrow when you get round to the replay.
Connect your accounts and schedule a fortnight of clips in one sitting. For an online coach, consistency in the feed does more for enquiries than any individual clip does, and consistency is a scheduling problem more than a creative one.
Sessions you already run become content without adding a filming day to a week that has no room in it.
Recorded check-in calls are the richest source you have, because every client question is a clip somebody else also wanted answered.
Coach-led technique content markets a facility better than a photo of the equipment does, and it recruits staff as well as members.
Explaining why a movement hurts is inherently short-form material. Clarity converts here far more reliably than intensity.
Pacing, fuelling and race-week questions come up constantly. Answer once on camera and the clip works for years.
Technique breakdowns depend on the whole body staying in frame, which is a layout decision before it is anything else.
A heavy single gets likes from other lifters. A clear explanation of why someone's back rounds at the bottom of a deadlift gets saved, sent to a training partner, and occasionally turns into a paying client. Those are different outcomes produced by different clips, and the second kind is far easier to generate at volume because you produce it accidentally every time you coach.
This is why the tool reads the transcript rather than the video. It is looking for teaching. If your footage is entirely silent demonstration, there is nothing for it to find, and the honest answer is that you should talk more while you film rather than expect software to invent commentary you never gave.
The corollary is that your least glamorous footage is often your most productive. A recorded client call, filmed on a laptop webcam with poor lighting, routinely produces more usable clips than a professionally shot gym session, because it is thirty minutes of you answering real questions.
Vertical video and barbell work are in tension. The lift is a tall movement, which sounds like it should suit 9:16, but the setup also needs width — the rack, the feet, the plates, the angle you filmed from. Crop to the athlete and you lose the reference points that make the technique point legible.
The general rule that works: use fit for anything you want people to study, and use the tracked crop for anything you want people to listen to. A clip where you are talking to camera between sets is a talking-head clip and should be framed like one. A clip where the point is the movement should keep the frame the camera captured.
Getting this wrong is the most common reason auto-generated fitness clips look amateur, and it is not a subtle failure. It is immediately obvious to any coach watching, which means it is obvious to your entire audience.
Split the job into four parts and it becomes obvious which half is worth automating. Finding the good ninety seconds inside a long take. Cutting them. Reframing them. Captioning them. Coaches almost always describe the second, third and fourth as the work, and it is the first that quietly eats the evening — a 90-minute coaching call takes 90 minutes to search before a single edit has been made.
That search is what an automatic pass replaces. It reads every sentence in the file, finds the passages that stand alone, places the edges on complete thoughts and hands back a ranked shortlist. Most people expect the output to be finished clips and are mildly disappointed; the honest framing is that you are being handed the twenty candidates you would otherwise have spent an evening locating, already captioned.
Hand editing keeps three jobs for good. Anything where the point is visual rather than spoken — a slow-motion comparison, a wrong-then-right sequence assembled from two takes, a montage cut to music. Anything that needs frames from two separate recordings in one clip. And the one piece a month you are putting advertising money behind, where an extra hour of human attention on the first two seconds genuinely pays.
The sequence matters more than the choice. Generate, sort by grade, overrule the order using the thing no engine knows — which question your clients keep asking — and then hand-finish the two that are going out. Running both processes in parallel just means doing the search twice.
Filming ten deliberate answers in one hour rather than mining an hour of footage. This is the strongest alternative on the list and it is badly underrated. Sit down, work through the ten questions clients ask you every week, and record each as a ninety-second piece to camera. The clips come out cleaner than anything mined from a session, because you knew where the sentence was going when you started it.
Posting the full session with chapters. Cheap, and useful to people who already follow you, but it reaches nobody new — a long upload is found, whereas a short clip is pushed. If your channel already gets search traffic, chapters are worth the twenty minutes; they are not a substitute for short-form.
Writing the cue as a text post or a carousel. Genuinely competitive for coaching content, because a written cue can be screenshotted and taken to the gym. What it cannot do is show the movement, and a correction described in words is much easier to misapply than one you can watch.
Going live weekly and clipping nothing. Streaming a technique Q&A to YouTube, Twitch or Kick produces the content while you talk, which for a coach with no editing tolerance at all is a defensible whole strategy. Clipping the stream as it runs is strictly better, but the live session alone still beats the plan where you intend to edit later and never do.
Worth knowing that the most common alternative in practice is none of these — it is a folder of unedited session footage that nobody has opened since the day it was filmed. That is the real baseline, and it is why a mediocre clip published this week outperforms a good one that stays theoretical.
Layout is the setting that decides whether a clip teaches anything, and it should be chosen per clip rather than once for the batch. Fit for a lift, so the bar, the feet and the plates all stay inside the frame the camera captured. Tracked vertical for a piece to camera. Split-screen or picture-in-picture for a coach-and-client exchange where both reactions are the content.
Aspect ratio follows where the clip is going: 9:16 for Reels, TikTok and Shorts, 1:1 for a feed post, 16:9 for a website or a course module. Vertical renders at 1080x1920, which is what the platforms re-encode towards anyway, so there is nothing to gain from exporting anything larger.
Caption style is a legibility choice on gym footage specifically. Chalk, white walls, mirrors and bright overhead lighting are exactly the background that swallows a thin outlined caption, and a cue nobody can read is a clip that teaches nothing. Pick one of the eleven styles with a solid plate behind the words, load your colours into the Brand Kit, and stop revisiting it — captions are burned into the picture, so changing your mind later is a re-render rather than a setting change.
Watch where the words sit relative to the movement. A caption block parked across the middle of the frame will cover the hip crease on a squat clip, which is the one part of the picture the clip exists to show. Say you are cutting a depth demonstration: the fix is to reframe slightly rather than to fight the caption, because the platform will also stack its own interface along the bottom of the video once it is posted.
Filler and silence removal happens automatically on every clip, which on training footage removes a great deal — rest periods, plate changes, the walk back to the rack. The case to check before publishing is tempo work, where a paused rep means the silence is the coaching point rather than dead air.
When your footage is silent by design. If everything you film is demonstration set to music, the transcript is empty, and selection, captions and boundary cutting all read the transcript. What is left is a cropper, and a phone editor already does that for free. Narrating future sessions changes this completely; nothing on our side does.
When the volume is not there. This earns its place when you need ten clips a fortnight and cannot make them. At one post a month, the search you are paying to automate takes you twenty minutes, and the honest advice is to spend the money on a microphone instead.
When you have not cleared the footage with the person in it. A client session is a private interaction before it is content, and the engine will process any file you hand it without knowing whose face is in the frame. Ask before you film, not after you have found the clip you want.
And when what you actually want is movement analysis. Coaches arrive expecting rep counting, bar-speed tracking or a form score, and none of that exists here — the biggest mistake you can make with this tool is assuming it watches the lift. It reads what you say about the lift. If you need joint tracking, that is a different category of software and you should go and buy it.
It does not count reps, track joints, measure bar speed or assess technique. There is no pose estimation in this product and no automated form scoring. It does not know a hinge from a squat. Anything that reads like analysis of a movement in your clips is analysis you supplied out loud.
It also does not generate demonstration footage. No synthetic athletes, no AI-created exercise videos, no artificial voiceover reading a script over stock clips. Everything in the export came from the file you uploaded and everything in the captions came from your own audio.
And it is not a programming tool or a client management system. It cuts video. If you were hoping for something that builds a training block, that is a completely different category of software.
Opening on the set-up. Three seconds of walking to the rack is three seconds you were never given. Begin on the sentence that names the problem, and let the movement arrive underneath it.
Cropping to the face on a technique clip. If the point being made is where the bar tracks relative to midfoot, a head-and-shoulders frame has deleted the evidence and left only the opinion.
Explaining the anatomy before giving the fix. Nobody scrolling past needs the origin and insertion of a muscle. Lead with the cue, follow with the reason, and only include the reason if the clip still has room for it.
Publishing silent demonstrations. Without audio there are no captions and nothing that marks the video as yours, so it looks identical to every other clip of somebody doing a set, and nothing in it suggests you understood why the set mattered.
And building clips around your heaviest lift. Your best set is usually your least useful clip, because it tells a stranger what you are capable of rather than what you could teach them to do.
Check the audio before you check anything else. Music laid over the whole recording leaves the transcript sparse. So does a camera ten feet away in a room with a fan running, and so does a client answering from off-mic. A sparse transcript is a short shortlist, and no setting compensates for speech that was never captured.
If every clip came from the first ten minutes, that is usually correct rather than broken. The briefing at the top of a session is the densest speech in the file and the working sets after it are mostly counting. Recording a short verbal summary at the end gives the selector a second dense patch to draw from.
Exercise vocabulary comes back mangled more often than ordinary speech does. RDL, GHR, tempo written as three-one-one and most of the mobility glossary sit outside what a general transcriber expects, so they arrive as something close but wrong. Captions are editable per clip and this is the one edit worth making every time, because a wrong term is precisely what a coach's audience notices first.
And if a clip ends a fraction before the rep does, that is the boundary rule working in the wrong direction: the cut is placed at the end of your sentence, and the movement outlasted the sentence. Drag the out point two seconds later and re-render. It takes longer to describe than to do.
Talk while you film. This is the single highest-leverage change available, and it costs nothing. A silent set gives the engine no transcript, no captions and no basis for selecting anything.
Mic yourself if the room is loud. Clanking plates and gym music sit right on top of speech, and no transcription system recovers a cue that was drowned out at source. A cheap lapel mic changes caption quality more than any setting in this tool ever will.
Film wider than feels right. You can always crop in afterwards; you can never widen a shot that clipped your feet off. Extra floor at the bottom of frame is cheap insurance.
Record your calls. Ask the client, get a yes, and keep the file. Coaching calls are the most under-used content source in the entire industry, and clipping them is what turns one hour of paid work into a month of marketing. A podcast clip generator handles the same job for long-form audio-led episodes.
Observations from operating the pipeline in production — not general advice.
On a piece to camera the tracked crop is the right answer. On a heavy set it is the reason the clip looks broken: a lifter's head travels most of a foot vertically through a squat, so the crop chases it downward, the bar leaves the top of frame at the bottom position and the plates never appear at all. Fit layout exists for this exact footage, because holding the frame the camera captured beats guessing which part of it mattered.
Moment selection reads the per-word transcript rather than the pixels, which is why a beautifully shot session with no commentary can come back with nothing in it. There is no pose model in the product and nothing watching the bar. In practice a laptop webcam recording of a client call outperforms a tripod-shot gym session most times we compare them, because it is thirty unbroken minutes of somebody talking.
On a podcast a cut that lands slightly early costs you a little context. On a form clip it costs the fix, because the correction is frequently a single sentence and the clip either contains it whole or teaches nothing. That is why cuts are placed on sentence boundaries instead of on a round number of seconds, and why a clip will run past its target length rather than stop halfway through the thought that justified it.
Worth doing where you can, and it removes the reframing question entirely. It does not remove the other three jobs — finding the good thirty seconds in a long take, cutting the dead air, and captioning it. Those are still where the time goes, and those are what this automates.
Genuinely fine for one clip. The problem is the tenth. Manually scrubbing an hour-long call to locate four teaching moments, then trimming and captioning each, is around two hours of work that most coaches will do twice before quietly abandoning the habit.
A good editor produces better individual clips than any automatic tool, and they will never know which of your coaching answers matters most, because they are not the coach. The workable split is automation for volume and a human for the pieces you are putting money behind.
Most were designed around podcast footage where a centred face is always the right answer. For technique content that assumption produces bad clips. Compare on layout control and on whether the moments picked are the ones you would have picked — one upload settles it. Try a session.
A single recorded session is the fairest test there is. You already know which two minutes of it were the good part — see whether it agrees.
⚡ Clip My Session — $1 trial
