AI Video Editing Flows

Reel teardowns: short-form reels measured and taken apart

Reel teardowns take one short-form reel apart and measure it: length, hard cuts, shot length, loudness and speech pace, with the frames read one by one. So far there are 26 teardowns, drawn from 138 reels on 69 pages, all measured on 5 Nov 2026. Every teardown links to the original reel and never copies it.

The teardowns so far

#TeardownFormatLength (s)Hard cuts/minBig visual changes/minLoudness (LUFS)
#01Stick-Figure Survival Reel, Fast CutsInteractive survival series (audience votes next episode)61.132.442.2-14.1
#02Phrase Captions Over B-Roll, Cut per LineWisdom, stoic, shayari and quote pages47.624.031.5-14.1
#03Counter-Led Slicing Sim, No Hard CutsSatisfying escalation ('10 vs 1,000 cuts')72.90.02.5-14.5
#04Stick-Figure Horror Reel, Vote EndingInteractive survival series (audience votes next episode)91.55.24.6-14.2
#05Code Physics Simulation Reel, One TakeCode physics simulation (race, battle, merge)33.70.00.0-14.6
#06Stick-Figure Comedy, Reaction-Clip CutStickman / simple 2D comedy and explainers17.710.26.8-14.5
#07Doodle Explainer, New Picture per LineStickman / simple 2D comedy and explainers225.918.313.0-14.1
#08Stick-Figure Race, Choose Your ColourCode physics simulation (race, battle, merge)65.30.00.9-14.1
#09Paper Quiz Reel With Countdown CirclesQuiz / would-you-rather / pick-one63.80.02.8-14.2
#10AI Macro Fact Reel With Soft CaptionsAnimal biology explainer61.54.923.4-14.2
#11AI Giant-Fruit Rooms, Repeated PatternAI giant-object room reveal (ASMR)69.06.119.1-14.2
#12Rules-Horror Reel With a Title PosterNarrated horror story over AI visuals (Hindi)64.28.418.7-14.1
#13AI Bodycam Clip: One Shot and a RevealAI recurring character IP (pets, oddballs, character universes)10.10.011.8-14.1
#14AI Film Episode in a Letterbox StripAI cinematic short-film series261.50.00.0-14.1
#15Horror Reel With a Recap and Vote ButtonsInteractive survival series (audience votes next episode)79.83.80.0-14.1
#16Retro 3D Survival Reel, Pixel CaptionsInteractive survival series (audience votes next episode)119.311.612.1-14.1
#17Split-Screen Ball Simulation ReelCode physics simulation (race, battle, merge)34.20.00.0-14.5
#18First-Person History Tour, Place LabelsHistory POV / reconstruction32.111.213.1-14.0
#19Hands-Only Dessert ASMR, Tight ShotsHands-only real-process ASMR (food, craft, care)27.339.622.0-15.9
#20AI Pet-Per-Month Reel, Equal SegmentsQuiz / would-you-rather / pick-one96.16.96.2-15.9
#21Still-Panel Horror Reel, Speech BubblesNarrated horror story over AI visuals (Hindi)36.10.05.0-14.1
#22AI Vegetable Serial With Long HoldsAI anthropomorphic object / vegetable serial (Hindi comedy)97.83.14.9-14.1
#233D Animal Gag Reel, Fast Cuts After SetupAI anthropomorphic object / vegetable serial (Hindi comedy)25.219.023.8-14.0
#24AI Dessert-Room Reel, One Timing TemplateAI giant-object room reveal (ASMR)51.95.88.1-14.4
#25AI Myth Reel in a Letterbox StripAI cinematic short-film series78.60.00.0-14.1
#26Ten-Second AI Kitten Clip, One ShotAI recurring character IP (pets, oddballs, character universes)10.40.00.0-14.2

What the 138 reels have in common

  • 126 of 138 reels had an audio track; 12 files had none, so loudness is blank for those.
  • Median integrated loudness was -14.2 LUFS across the 126 reels with audio, and the 10th to 90th percentile ran from -14.8 to -14.1 LUFS.
  • 117 of those 126 reels (92.9%) sat within 1 LU of -14 LUFS.
  • The first hard cut came after 3 s in 67 of 138 reels (48.6%), and a further 38 (27.5%) had no hard cut at all.
  • The first hard cut came within 3 s in 33 reels (23.9%).
  • Reels with a series number in the caption (such as Part 3 or Night 2) reached 3x their page median in 23 of 95 cases (24.2%), against 178 of 827 (21.5%) for reels without one. This uses the 922 of the 935 reels we collected that have a page median, and our own matching rule, so the gap is small and approximate.

Editing benchmarks by format

We also put the same measures side by side for all 23 formats, with plays per follower and confidence. See Short-form editing benchmarks by format.

Formats we wrote up

How we measure a reel

  • Pick the best reel and a typical reel from each page, and download them for measurement only.
  • Run scripts for length, hard cuts, big visual changes, loudness and speech pace.
  • Read a contact sheet of each reel frame by frame and write down what is on screen. A model does this step, so we label it as observation.
  • Write the page from the measured numbers and the notes. We never copy or embed the reel.

Method and limits

  • Plays, likes, comments, followers and posting dates were collected from Instagram by Apify on 5 Nov 2026.
  • For each of 69 pages we took the best reel and a typical reel of its last ten, 138 reels in all, and measured the files by script.
  • Hard cuts, shot length and the first cut come from a script that finds hard cuts between frames.
  • Big visual changes count hard cuts plus large camera travel or scene changes, so they are always at least as many as hard cuts and also catch some fast moves.
  • Loudness is the integrated loudness in LUFS of the audio track. Speech pace is words per second from a speech-to-text pass, with a language guess and its probability.
  • The frame-by-frame notes were written by a model reading contact sheets of each reel. They are observations, and each teardown page labels them that way.
  • Ratio to the page median is plays divided by the median plays of the page's last ten reels, as measured on 5 Nov 2026.
  • Format statistics are recomputed here from the raw files and compared with the workbook's Format scoreboard; any difference is recorded under recheck.
  • Pattern counts per format are keyword searches over the frame-by-frame notes of that format, so read them as approximate.
  • Hard-cut counts miss dissolves, so a reel built on dissolves can show close to zero hard cuts.
  • The big visual change count also counts camera travel and scene changes, so it is not the same as the cut count. We show both.
  • Some full-HD files had no audio track, so loudness and speech are blank for those reels.
  • Plays include replays. Shares and follower conversion cannot be seen from outside.
  • Ten reels from one page are not ten independent tests.
  • Hashtag discovery favours pages that already had a hit, so breakout pages are over-represented.
  • Plays per follower favours small, newer pages; it measures reach health, not size.
  • Formats measured on only one to three pages are hypotheses.
  • No number here is a promise about what a new reel will do.

Plays and followers as shown on Instagram on 5 Nov 2026.

Page checked on Nov 4, 2026.