How to Make Clips From Long Videos Automatically

You recorded a two hour podcast or a forty minute tutorial, and somewhere inside it are eight or nine moments that would do well as Shorts. Finding them by scrubbing the timeline takes longer than the recording did, and by the time you have cropped, captioned and exported each one, the episode is a week old. Learning how to make clips from long videos automatically is mostly about changing what you search: instead of hunting through video, you hunt through the transcript, and let the timestamps do the cutting for you.

This guide covers the manual method first, so it works even without any specific tool, then the faster transcript driven version.

Step 1: Get a timestamped transcript of the full video

Everything downstream depends on this. A transcript with timestamps turns a two hour video into a document you can skim in ten minutes, and every promising line comes with an in point and an out point attached.

The free manual route: upload the video as unlisted to YouTube, wait for automatic captions to generate, open the video, click the three dots below the player, and choose "Show transcript". Toggle timestamps on in the transcript panel, then select all and paste into a document. The accuracy is usually rough on names and jargon, but it is good enough to locate moments.

If you want a clean transcript you can actually read and search, Tapescribe takes a link or an uploaded file and returns a timestamped transcript plus SRT and VTT files in one pass, which matters later when you need captions burned into each clip.

Step 2: Find the clip-worthy moments in the transcript

Read the transcript, not the video. You are scanning for self-contained units: a moment that makes sense to somebody who has no idea what came before it.

Mark anything that matches these patterns:

• A question asked, then answered in under ninety seconds
• A strong opinion stated plainly, especially one that contradicts common advice
• A specific number, result or before-and-after
• A short story with a beginning and an end
• A step by step explanation that stands alone
• A moment of genuine reaction, laughter or disagreement

Copy each candidate into a list with its start timestamp, end timestamp and a one line note about why it works. Aim for more candidates than you need; you will cut roughly a third of them once you watch the footage back.

Step 3: Set the hook, then trim the in and out points

The first two seconds decide whether the clip gets watched. In the transcript, find the single strongest sentence in your candidate and make that the first line of the clip, even if it originally appeared thirty seconds into the answer. Setup can follow the hook; it rarely survives in front of it.

Then tighten the edges. Move the in point to the exact word the sentence starts on, not the breath before it. Move the out point to the end of the final word, not the pause after. Target thirty to sixty seconds for Reels and TikTok, and under sixty seconds if you want the clip treated as a Short on YouTube. Anything over ninety seconds needs to be genuinely gripping to hold.

Automatic hook detection does the same job by scoring transcript segments on question-and-answer structure, sentiment shifts and topic completeness. It is good at surfacing candidates and less good at knowing which ones fit your audience, so treat its ranking as a shortlist and make the final call yourself.

Step 4: Cut the clips out of the source video

With timestamps in hand, this part is mechanical.

In a normal editor (Premiere, Final Cut, DaVinci Resolve, CapCut), drop the full video on the timeline, use the timecode field to jump straight to each in point, mark in and out, and export that range. Do not re-edit inside the clip yet.

If you prefer the command line, FFmpeg cuts without re-encoding, which is close to instant:

ffmpeg -ss 00:14:22 -to 00:15:07 -i source.mp4 -c copy clip-01.mp4

Stream copying cuts on the nearest keyframe, so the start can drift by a second or so. If precision matters, drop -c copy and let it re-encode.

Step 5: Reframe the clip to vertical without cutting off heads

A 16:9 frame cropped to 9:16 loses most of its width, and whoever is speaking is rarely dead centre.

Manually, set your sequence to 1080x1920, drop the clip in, scale to fill the width, and keyframe the horizontal position so the active speaker stays in frame. For two person interviews, the cleaner solution is a stacked split: top half on one speaker, bottom half on the other, both cropped to squares. It reads well on a phone and avoids constant repositioning.

Automatic reframing tracks faces and pans the crop window for you. Check every clip afterwards regardless. The usual failures are a hard pan when someone gestures off screen, and a slow track on quick cuts. Where the tracking gets jumpy, lock the crop to a static centre position for that clip; a slightly off-centre static frame beats a crop that swims.

Step 6: Add burned-in captions to every clip

Most short form gets watched muted, so captions are not optional. Because you already have a transcript, you are not typing anything.

Export an SRT for each clip, then import it into your editor as a caption track and style it: a heavy sans-serif, large size, high contrast, positioned in the middle third of the frame rather than the very bottom where platform UI sits. If you cut clips from the full video's SRT, the timings will be offset by the clip start time, so either shift them or generate a fresh SRT per clip. If you have not worked with subtitle files before, our guides on creating an SRT file from video automatically and adding subtitles to video automatically cover the formatting and styling in more detail.

For word by word highlighting, split the captions into short phrases of three or four words rather than full sentences. It holds attention better and keeps the text block small.

Step 7: Batch export for TikTok, Reels and Shorts

You want one master export per clip, not three. Export 1080x1920, H.264, at the source frame rate, with captions burned in. That single file uploads cleanly to all three platforms.

Name files so you can find them later: episode-14_clip-03_hook-pricing.mp4 beats clip3_final_v2.mp4. If your editor supports render queues or batch export presets, set the preset once and queue every clip in one go rather than exporting one at a time.

Then vary only what needs varying per platform: the caption text in the post description, the cover frame, and the hashtags. Keep a simple spreadsheet with one row per clip holding the filename, the hook line, the source timestamp and the platforms it has been posted to.

Common problems and how to fix them

The clip starts mid-sentence. Your transcript timestamp marks the start of a caption block, not the start of the sentence. Nudge the in point back a beat and listen to the first word before exporting.

Automatic hook detection keeps picking the same kind of moment. Scoring models favour clear question and answer structure, so long narrative sections get overlooked. Read the transcript yourself for stories and pull those manually.

Captions drift out of sync partway through. Almost always a frame rate mismatch between the source video and the sequence, or an SRT generated from a differently trimmed version of the file. Regenerate the SRT from the exact clip you exported.

Speaker's head gets cropped off in vertical. Turn off auto tracking for that clip and set a static crop, or switch to a stacked split for two person sections.

The clip makes no sense without context. Add a one line text overlay at the top of the frame naming the topic, or record a five second intro line. If neither fixes it, the moment is not a clip.

Names and technical terms are wrong in the captions. Fix them in the transcript before generating clip SRTs, not in each clip afterwards. Tapescribe lets you correct the transcript once and re-export the subtitle files, which saves repeating the same fix eight times.

Frequently Asked Questions

How long should clips from a long video be?

Thirty to sixty seconds works for most platforms and gives you room for a hook plus a complete thought. Keep clips under sixty seconds if you want them treated as Shorts on YouTube. Longer clips can work when the moment genuinely holds attention, but the retention bar is much higher.

How many clips should I pull from one long video?

From a one hour episode, five to ten strong clips is realistic. Pulling twenty means most of them are filler, and posting filler trains the algorithm against you. Quality of the hook matters more than volume.

Can I make clips automatically without an AI tool?

Yes. Generate a transcript, mark the timestamps of the best moments, then cut those ranges with FFmpeg or your editor and add captions from an SRT. The manual route in this guide is the same workflow automated tools follow; it just takes longer per clip.

Do I need to re-transcribe each clip for captions?

No, and you should not. Generate one accurate transcript from the full video, fix any name or jargon errors once, then export per-clip subtitle files from it. The same source transcript also gives you automatic chapters for the long video, so one transcription pass covers both jobs.

The workflow that actually holds up over time is transcript first: transcribe once, search the text instead of the timeline, and let timestamps drive the cuts, the captions and the chapters. Automation speeds up finding candidates and reframing, but the two judgement calls that decide whether a clip performs, which moment to pull and which sentence opens it, stay yours. Build the transcript step into your routine and the rest becomes an afternoon of mechanical work rather than a week of scrubbing.