AI Video Editing Flows

@threeavatars.ai: how an AI ASMR theme page edits its reels

@threeavatars.ai is an AI ASMR page in AI sensory and ASMR, with 39,963 followers on 5 Nov 2026. Posts short, single-shot AI macro clips of a beauty ritual with an odd twist, left uncut and almost silent, with no on-screen text.

Watch the reels

Most-played reel in our sample. Posted Nov 3, 2026 Open on Instagram (opens in a new tab)
Open on Instagram (opens in a new tab)
A typical reel. Posted Nov 1, 2026 Open on Instagram (opens in a new tab)

Press Play here to watch a reel in Instagram's own player; nothing loads from Instagram before that. Plays as counted on Nov 5, 2026; the account on Instagram has the rest.

Independent analysis of public data. We are not affiliated with this creator and they have not endorsed this page. The reels above load from Instagram's own player only when you press Play here; we never copy or re-upload them. If you run this account and want this profile corrected or removed, write to us through our contact page and we will do it.

The page in numbers

Account: @threeavatars.ai on Instagram. Niche: AI sensory and ASMR. Format: AI ASMR cutting / creepy-cute. See the format playbook.

MeasureValue
Followers39,963 on Nov 5, 2026
Reels measured10
Median plays per reel56,734
Plays per follower1.42 (middle third of the 71 pages in our study)
Most-played reel in the sample11,115,996 plays, posted Nov 3, 2026
Reels at 3x or more the median of the page's other reels1 of 10
Posts per week, over the dates of the measured reels21
Median reel length8.0 s
Last post before we measuredNov 4, 2026
Spoken languageno spoken language
Pictures made withAI video

Plays per follower is the median plays of the page's recent reels divided by its followers. The median across the 71 pages in our study is 0.55; it favours smaller pages, plays include replays, and a page's figure is not a forecast for a new account. In the reel table, each multiple compares a reel with the median of the page's other sampled reels, so the reference differs from row to row.

How its reels are edited

MeasureTypical reelBest reel
Reelopen on Instagramopen on Instagram
Plays66,02911,115,996
Plays as a multiple of the median of the page's other sampled reels1.4x234.3x
PostedNov 1, 2026Nov 3, 2026
Length (s)8.18.1
Hard cuts per minute0.00.0
Big visual changes per minute0.00.0
Loudness (LUFS)-28.7-28.0

Not shown in the table: no big visual change was detected, no speech was detected in the measured reels.

The best reel (11,115,996 plays, 8.1 s) is a single uncut shot. It begins on an extreme macro of a closed eye, still for 0.5 s, and then a make-up brush loaded with tiny wet sea creatures sweeps in. By 3 s they cling to the brush, and the shot ends mid-application so it can loop. There is no on-screen text and no speech. Our frame notes say the caption was only hashtags.

The look is photoreal AI macro beauty, with hard flash-like lighting, deep shadows, pink eyeshadow and a cyan nail as the accent. Sound is quiet ambient, measured at -28.0 LUFS.

The typical reel (66,029 plays, 8.1 s) is built the same way. The eye, the single shot and the quiet audio (-28.7 LUFS) stay, but the brush carries tiny mice. Only the odd item changes between reels, which is what our frame notes mean by swapping the product on the brush.

Why it likely works

One reading is that a very clear first half-second, a calm and familiar beauty image, followed by an odd intrusion, is easy to understand without sound. The short loop gives viewers little reason to leave. The format recipe frames this as one shot, no captions and a seamless loop.

The best reel reached 11,115,996 plays while the page's median is 56,734, on a page with 39,963 followers. That suggests a single standout idea, and it may not repeat. The page also posts at 21.0 reels a week, so volume may matter as much as any one reel.

How this format works across the pages we measured: the format playbook.

Make this without footage

Our proposed build, not the creator's method: their tools are not known to us. Use your own subjects, objects and rules rather than copying a creator's finished reel.

  1. Pick a calm, recognisable everyday ritual and one odd addition, such as a tiny object arriving on the tool.
  2. Generate a locked macro plate as an AI image, such as a tight crop of a closed eye, and keep it the same for every variant.
  3. Generate one image-to-video clip of about the same length as the measured reels. Swap only the odd item per variant.
  4. Keep the edit minimal in a HyperFrames composition (HyperFrames hub): a light colour grade, trimmed to a clean loop, no text. See how we work: colour and sound.
  5. Keep the sound very quiet or use the generator's own audio. The measured reels sit near -28 LUFS.
  6. Export vertical. We have not built this exact reel ourselves, and our frame notes suggest there is little for an editing engine to add beyond the grade and the loop.

What to test before copying it

  • Whether reach depends on one outlier reel. Only one reel in the sample ran at three times the median.
  • Whether the very quiet audio helps or hurts, since most viewers have sound off or low.
  • Whether the odd creatures or morphing bodies at close range could trigger a platform AI label or content rule.
  • Whether volume, rather than any single idea, is what drives the page's reach.

Method, sources and removal

  • Numbers were collected from public Instagram data on Nov 5, 2026 and are as of that date. Each page is judged on its last ten or so reels.
  • We measured the reels in the table by script, and read at least one of them frame by frame. Reels are shown through Instagram's own embed and never copied or re-uploaded.
  • Pages were found through hashtags and seed lists, so pages that already had a hit are over-represented. What this page does is not proof of why it grew.
  • Plays include replays. Shares, saves, watch time and follows cannot be seen from outside Instagram.

Corrections or removal: write to us through our contact page.

Back to all theme page profiles or read what the measured pages have in common.

Page checked on Nov 4, 2026.