AI Video Editing Flows

We measured 93 faceless Instagram theme pages: what the 71 camera-free ones have in common

We measured 115 public Instagram accounts and 935 reels on 5 Nov 2026. 71 are faceless theme pages that make their own pictures without a camera. This page describes patterns among them, including the pages with unusually high recent plays per follower, and where common editing advice did not hold up. It shows patterns, not proof of cause.

The short answer

Sample: 115 public accounts and 935 reels, then 93 faceless theme pages, then the 71 that make their own pictures without a camera. 138 reels measured file by file, 86 read frame by frame. Measured on Nov 5, 2026.

  • Pages that make their own pictures without a camera had more plays per follower than pages built on other people's clips: a median of 0.56 plays per follower against 0.07 (68 pages against 17).
  • A page's best reel was shorter than its typical reel on 33 of 53 pages, longer on 15 and the same on 5 (median 39.5 s against 57.9 s).
  • Cut rate, first visual change and speech pace did not separate a page's best reel from its typical one. None of those differences passed a simple sign test.
  • Few posts did not hold pages back: the 19 pages with 40 posts or fewer had a median of 2.7 plays per follower against 0.34 for the 50 with more, and 11 of them had a reel past 1,000,000 plays. Survivorship is strongest here.
  • Running word-by-word captions were rare: 7 of 67 frame-read reels. The most common choice was no text at all (32 reels).
  • Most measured reels with sound share the same loudness: 93 of 99 sit within 1 LU of -14 LUFS.

These are patterns among the pages we sampled, measured on Nov 5, 2026. They are not proof that any one choice causes reach. The pages were found through hashtags and seed lists, which favour pages that already had a hit.

Who we measured

On Nov 5, 2026 we collected the last ten or so reels of 115 public Instagram accounts, 935 reels in all. 93 of the accounts are faceless theme pages; the other 22 are creators who appear on camera, and we left them out.

GroupPagesIn this study
Made without a camera (AI video or images, 2D or 3D animation, code, motion graphics, type)68Yes
Made mostly from stock footage plus AI images3Yes
Own camera footage (hands-only process videos, filmed paper or posters)5No, shown for comparison only
Built on other people's clips (film, news, archive or reposted clips)17No, shown for comparison only

So the study covers 71 pages that make their own pictures. For 69 pages we also downloaded two reels, the best of the ten and a typical one near the page's median, and measured the files: length, hard cuts, big visual changes, the first cut, loudness and speech pace (138 reels). 86 of those reels were read frame by frame for the hook, the text on screen, the look and the ending.

Camera-free pages had more plays per follower in our sample

Median plays per follower, by where the pictures come from (measured 5 Nov 2026)
Made without a camera0.56 (68 pages)
Mostly stock plus AI images0.16 (3 pages)
Own camera footage1.1 (5 pages, too few to judge)
Other people's clips0.07 (17 pages)

Plays per follower is the median plays of a page's recent reels divided by its follower count: recent plays relative to follower count, not unique reach (plays include replays). It favours small pages.

27 of the 68 no-camera pages get more plays per reel than they have followers. Among pages built on other people's film, news or reposted clips, 1 of 17 do. The no-camera pages are not all small: the median of their median plays per reel is 42,683, against 31,961 for the clip pages.

One untested reading is that originality is rewarded: Instagram has said that reposts lose recommendations to the original, as reported by Social Media Today. Another is that the clip pages we found are older pages past their peak. Our data cannot separate the two, and the five camera-footage pages are too few to compare.

Reach by niche

Median plays per follower by niche, pages that make their own pictures
Satisfying and simulations12.47 (4 pages)
Interactive stories5.45 (5 pages)
AI sensory and ASMR2.76 (6 pages)
2D animation and stick figures2.04 (6 pages)
AI short films1.29 (6 pages)
History0.39 (6 pages)
AI stories and horror0.3 (7 pages)
AI characters0.29 (17 pages)
Mythology and devotional0.23 (4 pages)
Facts and explainers0.16 (7 pages)

Only niches with 4 or more measured pages are shown. Edits (1), Finance (1), Wisdom and quotes (1) had fewer pages and are left out of the chart.

The two niches at the top are small: 9 pages in all, 4 of them with under 20,000 followers. The AI character niche has the most pages (17), with a median of 184,039 followers, and sits low on reach per follower. A niche median says what the measured pages have in common, not what a new page will get.

Ideas from each niche, with the pages that show them, are on the theme page ideas list.

Viral with 40 posts or fewer

For 69 of the 71 pages we also have the post count shown on the profile on Nov 5, 2026 (every post type, not only reels). We split them at 40 posts.

Median plays per follower, by number of posts on the page
40 posts or fewer2.7 (19 pages, median 20 posts)
More than 40 posts0.34 (50 pages, median 162 posts)

Plays per follower favours small pages, and a page with few posts usually has few followers, so part of this gap is arithmetic. Read it with the counts below.

Pages with at least one reel past 1,000,000 plays in the sample
40 posts or fewer11 of 19
More than 40 posts23 of 50

Survivorship is strongest here. We find pages through hashtags and seed lists, so a page with few posts is in our sample only if one of its reels already travelled. New pages that posted a few reels and stopped are not here. What the numbers do show is that a large audience does not need a long back catalogue: 5 of these 19 pages are interactive survival story series, which run as numbered episodes from the first post.

PageFormatPostsFollowersMost-played reelPlays per follower
@bubblesimulator.aicode physics simulation41,02243,0433.14
@project.red.hatinteractive survival story814,3001,202,52023.42
Not namedAI character9242,979167,0320.16
@stickocalypse_shortsinteractive survival story9178,11618,944,2422.7
Not namedinteractive survival story108,2061,422,21713.43
@duzianoanimal biology explainer1227,975416,2582.63
Not namedmythology animation164,9414,2290.12
@miniworldsiminteractive survival story1771,7021,796,7195.45
@racing2thetop.igcode physics simulation1865,4195,684,99517.38
@infiniteslice01satisfying escalation2058,59839,036,9827.56
Not namedstick-figure2012,0845,817,1004.25
Not namedinteractive survival story2257,43575,3600.69
@waghya_universeAI short-film259,478613,5391.84
Not namedAI short-film2710,202633,9500.73
@ms._stretchyAI character29279,7102,302,5496.2
Not namedhistory POV3222,36546,1650.53
@ranjanbabita781AI talking-baby comedy3462,0123,205,3565.14
@chhotay_babuAI talking-baby comedy3550,3551,301,2121.04
@satify.ai.studioAI ASMR3614,4281,220,7281.62

We name the pages we profiled. The others are counted but not named; some of them we also leave out on purpose (health topics, or reels we judged unsuitable to send readers to).

Newer low-post pages we measured on 6 and 7 Nov 2026

After the study we measured more low-post pages: eight named to us as examples, and pages suggested by ChatGPT Deep Research that we then checked with Apify (we checked 45 of its suggestions; 39 exist and are public). These pages are not in any statistic above. Most began posting in October 2026.

PageFormatPostsFirst reel in our dataFollowersMost-played reelReels past 1,000,000
@mindunblindstill-artwork dialogue4Nov 4, 202628,49013,398,233 (reel 2)1
@geetillustrationsillustrated song-lyric19Oct 5, 202662,69825,560,229 (reel 5)8
@notruecolorstory3D rules-horror13Oct 16, 202679,3027,513,373 (reel 8)6
@plotshift.studioschoice-story9Oct 15, 202658,8212,018,055 (reel 1)2
@thebig.whatifwhat-if scenario31Oct 9, 202612,6551,900,341 (reel 8)2
@map_sodyanimated map22Oct 5, 20267,342995,980 (reel 10)0
@desmontandoomundo3D engineering explainer19Oct 21, 2026676,37116,779,523 (reel 16)19
@akhyaan.officialfolk-art mythology story34Aug 20, 202689,54313,890,289 (reel 10)6
@vedastovikings3D mythology film11Oct 24, 202613,3051,394,115 (reel 5)2
Still-artwork dialogue page. @mindunblind profile, posted Nov 4, 2026 Open on Instagram (opens in a new tab)
Illustrated song-lyric page. @geetillustrations profile, posted Oct 11, 2026 Open on Instagram (opens in a new tab)
3D engineering explainer page. @desmontandoomundo profile, posted Oct 31, 2026 Open on Instagram (opens in a new tab)

The most-played reel of three of these pages, ready to play from Instagram. Plays as counted on 6 and 7 Nov 2026.

All 9 make their pictures without a camera: still art, 2D illustration, low-poly 3D, 3D mannequin figures or animated maps. The two still-picture pages (@mindunblind and @geetillustrations) never change the picture at all: every reel we measured is one image with words animating over it. 2 more pages we measured are left unnamed because their reels use photoreal people who could read as real.

A page's best reel was shorter than its typical reel on 33 of 53 pages

Best reel against the same page's typical reel, length (53 pages)
Pages33 shorter, 5 same, 15 longer

We compared each page's best reel with a typical reel from the same page, so the page, its niche and its audience stay the same. Of the editing measures we compared, length showed the clearest difference: the best reel was shorter on 33 of 48 pages where the two differed (exact sign test, unadjusted, p = 0.013). The median best reel ran 39.5 s against 57.9 s. We tested eight measures, so treat this as an exploratory finding: it does not show that shortening a reel improves it.

MeasurePages comparedBest reel, medianTypical reel, medianBest higherBest lowerSign test p
Length (s)5339.557.915330.013
Hard cuts per minute536.15.821180.749
Big visual changes per minute537.57.426190.371
Median shot length (s)537.36.322240.883
First hard cut (s)374.65.815190.608
First big visual change (s)374.24.515190.608
Loudness (LUFS)46-14.1-14.220100.099
Words per second (reels with speech)322.672.5816110.442
Direction of the difference on each page, best reel against typical reel
Length (s)33 lower, 5 same, 15 higher (p = 0.013)
Hard cuts per minute18 lower, 14 same, 21 higher (p = 0.749)
Big visual changes per minute19 lower, 8 same, 26 higher (p = 0.371)
First big visual change (s)19 lower, 3 same, 15 higher (p = 0.608)
Words per second (reels with speech)11 lower, 5 same, 16 higher (p = 0.442)

Social Media Today reported in January 2025 that Instagram uses watch time as a ranking signal, but our data measures only length, not completion rates. But a best reel is also the one that travelled to people who did not follow the page, and the topic may matter more than the length. Treat it as a reason to test a shorter cut of your next reel, not as a rule.

Faster cuts were not a feature of the higher-play reels

We split the 53 pages with two torn-down reels into thirds by plays per follower and compared each page's typical reel.

Typical reel length (s), by reach third
Bottom third56.1 s
Middle third47.8 s
Top third71.0 s
First big visual change in the typical reel (s), by reach third
Bottom third3.2 s
Middle third4.4 s
Top third7.8 s
Reach thirdPagesPlays per followerHard cuts per minuteBig visual changes per minuteReels with speech
Bottom third170.03 to 0.464.17.412 of 17
Middle third190.46 to 3.148.19.014 of 19
Top third173.24 to 34.695.86.712 of 17

These are two different comparisons. Inside one page, the best reel tended to be shorter than the typical one. Across pages, the pages with the most reach per follower post longer typical reels with a later first change. Both can be true at once. Format and topic are plausible explanations, but these comparisons cannot separate them.

The top third does not cut faster, and its first big visual change comes later (7.8 s against 3.2 s). Part of this is format: 6 of the 17 top-third pages are code simulations, escalation reels or survival story series, which play as one continuous take or long story beats, where the hook is the premise on screen in the first frame, not a fast cut. Across all 106 torn-down reels, only 27 change the picture in a big way within the first 3 seconds.

When the picture first changes in a big way (106 torn-down reels)
Reels27 within 3 s, 57 after 3 s, 22 with no big change

A big visual change is a cut or a large shift in the picture. A reel can still hook in the first second with text, a question or motion inside one shot, which this measure does not count.

One page shows how much else is at play. @pastlooksindia posted two reels a day apart with matching measured length, matching cut counts and near-identical narration transcripts, but very different plays: 891,369 and 89,343.

Posted Jan 22, 2026 Open on Instagram (opens in a new tab)
Posted Jan 23, 2026 Open on Instagram (opens in a new tab)

What text the reels put on screen

Text on screen in 67 frame-read reels
Running captions of 1 to 4 words, synced to the voice7 reels
Running phrase or sentence captions, subtitles or speech bubbles8 reels
A fixed title, question, labels or counters only20 reels
No text, or only a watermark or handle32 reels

Labelled by us from the frame-by-frame notes. Many no-text reels have no speech either: simulations, ASMR and short AI clips.

Running captions show up where someone is talking: the survival story pages and a Hindi stick-figure explainer page caption every phrase. Pages without a voice tend to use one fixed line instead, a question or a title that sets up the reel. We counted styles; we did not test which style performs better. Our captions and text notes and the one clip, every caption style test show how to set both up.

Most measured reels had similar loudness

93 of the 99 torn-down reels with sound measured within 1 LU of -14 LUFS (median -14.2 LUFS). 75 reels have speech, at a median of 2.5 words per second. This describes the files we measured, not how they were mastered. The difference between best and typical reels was small and not significant (p = 0.099). Our colour and sound notes give the FFmpeg loudness command we tested.

What this means if you are starting a page

  • Make the pictures yourself. The measured pages with the most reach per follower make theirs with AI video, 2D animation, code simulations or motion graphics, and none of those needs a camera. The ideas list groups them by niche.
  • Pick a format with a premise that reads on the first frame: a question, a choice or a scene already in motion. The 30 page profiles show how each page does it.
  • Cut a shorter version of your next reel and compare the two. Length was the one difference we measured between a page's best and typical reels.
  • Test running captions when there is speech, and a fixed label or question when there is not.
  • Master to about -14 LUFS and check the safe zone with our safe zone overlay.

To build one: choose a format from the ideas list and read its playbook, then use our effects library for the captions, counters, titles and transitions these pages use, each rendered by us with the prompt and code. Our tests show what worked and what broke when we ran the tools.

Method and limits

  • Page and reel data (followers, plays, likes, comments, length, posting dates) were collected from public Instagram pages with the Apify scraper on Nov 5, 2026. Follower and play counts are as of that date.
  • Each page is judged on its last ten or so reels. Plays per follower uses the median plays of those reels.
  • Files were measured by script: hard cuts by a scene-change detector, big visual changes by a second, looser detector that also counts camera travel, loudness as integrated LUFS, speech pace from a transcript.
  • Best against typical: for each page, the reel with the most plays in its sample against a reel near its median. The sign test counts pages where the best reel was higher or lower and ignores ties.
  • Page groups and niches were labelled by us from each page's bio, captions and contact sheets. Captions were labelled by us from the frame-by-frame notes.
  • Survivorship: hashtag discovery finds pages that already had a hit, so successful pages are over-represented and failed pages are missing.
  • Ten reels from one page are not ten independent tests. Plays include replays. Shares, saves, watch time and follows cannot be seen from outside Instagram.
  • Niches with 1 to 3 pages are hypotheses. The best-against-typical comparison covers pages with two torn-down reels only.
  • Hard-cut counts miss dissolves, and the big visual change count also counts camera travel.
  • We corrected one of our own earlier figures: numbered series reach 3x their page median 24.2% of the time against 21.5% without, not 28% against 21%.
  • Nothing here is a promise of views or followers.

Page checked on Nov 4, 2026.