Landscape2026 · 08 · 08 Reelly183-app census

Auto-assembly works only where the format kills the discretion.

Not “automation doesn’t work.” Four constraint regimes are live in the market right now. Three of them make money. Reelly was built in the one that doesn’t.

The finding

Four constraint regimes, all live

Multi-clip auto-assembly works exactly when the format or the footage is constrained enough that creative discretion disappears. Where discretion remains, it fails. Each cell below is a real app shipping today.

No constraint

Any footage, any narrative

Elva: AI Video Director Agent

3.89★ · 6,000 dl/mo · $60k

Reelly as designed is here

Format constraint

One second per day, chronological

1 Second Everyday

4.79★ · $200k · $20.00/dl

Footage constraint

Action-cam footage only

GoPro Quik

4.78★ · $2.03M/mo

Template constraint

Fixed slots, fixed pacing

Muse · Vids AI

4.66–4.78★ · $1.52M combined

The fix is not to abandon assembly. It is to pick a constraint.


Coverage

What was actually read

AppKittie third-party estimates, pulled live 2026-08-08 · directional, not audited. Per-download revenue is monthly revenue over monthly downloads — a monetisation-intensity ratio, not lifetime value, and inflated for mature apps with an installed subscriber base. Classification is judgment: boundaries between assist, template and generation are genuinely fuzzy in about a dozen cases, and category totals move ~15% depending on those calls. Roughly 100 of the 183 are photo editors, beauty cameras, storage or streaming apps and are excluded from category totals.


Category totals

Where the money is

CapCut alone · general editor, 11.2M downloads/mo1 app$49.1M
Single-asset transform & generation40+~$12M
Assist-the-edit · AI does part of the job on your footage18$11.7M
Template / reel makers · the naive slot-fillers8$4.12M
Constrained auto-assembly4$2.63M

The single-asset transformation cluster is the biggest by count, but it is mostly thin clients over the same third-party generation APIs. The differentiator there is paywall design and distribution, not technology — and Reelly has no edge in either.


$11.7M/mo · 18 apps

Assist-the-edit, the largest addressable category

The AI does one part of the editing job on footage the user already shot. Note the ratings.

Apple App Store · monthly estimates
AppRel.DownloadsRevenue$/dl
Splice — Video Editor2010203,292$3,049,627$15.004.60
Captions: AI Edits Your Video2020326,813$2,184,112$6.684.74
Video Up! Video Editor2022100,164$1,001,661$10.004.57
开拍 — zero-skill talking head2023200,000$1,000,000$5.004.91
Filmora2016201,081$603,271$3.004.68
Alight Motion2019305,546$509,324$1.674.41
BIGVU Teleprompter Captions AI201661,306$408,831$6.674.73
快影 — Kwai official AI editor2017402,343$402,351$1.004.92
VivaVideo2014100,000$400,000$4.004.77
OpusClip — AI video clipping202574,917$321,778$4.304.85
VLLO — Vlog Edits2015100,000$300,000$3.004.83
Teleprompter for Video201660,000$300,000$5.004.92
PowerDirector201933,092$220,642$6.674.64
YouCam Video202150,532$202,138$4.004.71
VEED Shorts: AI videos202210,025$200,452$20.004.84
VN: AI Video Editor20181,000,000$200,000$0.204.89
TrimCut2024100,000$200,000$2.003.77
Teleprompter-Record Script2020197,778$162,800$0.824.90
Healthiest profile in the census

OpusClip: $321,778/mo, 4.85★, and +35.4% revenue growth in 30 days at one year old. It is AI selection over video, and it works.


$4.12M/mo · 8 apps

Template reel makers — the naive slot-fillers

Trending template, drop clips into slots, apply effects. The slot-filling takes the head of each clip. That is the opening.

AppRel.DownloadsRevenue$/dl
Muse: Reels Video Editor2022405,928$913,429$2.254.66
Instories2019300,000$1,000,000$3.334.82
Vids AI — the comp2022201,093$603,290$3.004.78
Unfold201730,000$400,000$13.334.86
Reel Maker for Instagram・BEAT2022100,000$400,000$4.004.73
Hula AI: Trend Video Generator202560,155$300,784$5.004.59
Mojo201830,000$300,000$10.004.90
Templify202140,282$201,398$5.004.77
$2.63M/mo · 4 apps

Constrained auto-assembly — and what it earns

AppDownloadsRevenue$/dlConstraint
GoPro Quik202,523$2,025,552$10.004.78action-cam only
1 Second Everyday10,000$200,000$20.004.791s/day, chrono
TimeHut — Baby Album40,000$200,000$5.004.63one child, chrono
iMemories6,073$200,000$32.934.89digitised home video

Note the per-download revenue: $5 to $33. Far above the $0.86–$3.00 consumer transformation band. Constraining the format does not shrink the value, it raises it, because the output becomes reliable.

The honest counter-example

GoPro Quik does unconstrained-looking multi-clip auto-editing and makes $2M/mo. But it has two things Reelly does not: footage constrained to one camera type with consistent framing and content, and a hardware distribution moat. Read it as proof that the constraint is what makes it work — not as proof that unconstrained assembly is viable.


The cleanest signal in the dataset

Assist rates 4.8–4.9. AI-magic rates 2.4–4.1.

Assist & constrained
快影 (Kwai editor)4.92
Teleprompter for Video4.92
开拍4.91
Mojo4.90
VN4.89
iMemories4.89
OpusClip4.85
VEED Shorts4.84
VLLO4.83
AI-magic & unconstrained
Pixi AI2.42
Genie AI3.55
Reelina3.66
TrimCut3.77
Elva — director agent3.89
Vixel3.96
AURA4.10
Movia AI4.12

An assist tool promises a predictable improvement to something you already have, and delivers it. An AI-magic tool promises a result and frequently misses. Since organic TikTok is the only distribution channel available without a UA budget, and sentiment feeds it, over-promising is not just a quality problem, it is a distribution problem.


Why one works and one doesn't

OpusClip’s selection versus Reelly’s

Both are “AI picks the good parts.”

OpusClipReelly as designed
Inputone long asset300 unrelated clips
Structuretimeline already existsnone; must be authored
Signalspeech — strong, legiblenone; visual taste only
Taskfind best 60s of 40 minauthor a narrative
Ground truthexists, checkablenone; differs per person

OpusClip solves a well-posed problem. Reelly was solving an ill-posed one. The lesson is not “don’t do selection” — selection is running at $322k/mo and 4.85★. It is do selection where there is a right answer that does not depend on the individual’s taste.


Already in the repo

The constraint was found a month ago

muse-demo-recipe.md · 2026-07-14
“Coherence comes from the treatment, not the footage. One color grade + uniform ~1s pacing + beat sync + one text style makes unrelated clips read as ‘my month / my trip.’”

That is a format constraint, written down before this session. It dissolves the creative-discretion problem for one specific shape:

ElementSpec
Clip count15–25
Clip length0.5–1.5s, on beat
Duration15–25s
Transitionshard cuts only, on beat
Text0 or 1 static serif overlay
GradeONE look across all clips
Content mix~40% people, ~60% scenery / food / objects

A 20-shot beat-synced montage under one grade requires no narrative discretion. It requires picking decent moments and applying a treatment uniformly. Picking moments is what Indexing/ already does at 3/3 and 4/4 against ground truth. Applying treatment uniformly is what ReelRenderer already does. The discretion wall was hit while trying to author a long-form travel vlog — the montage format does not have that wall, and Muse is doing $913k/mo inside it.


Recommendation

The smart-slot-filling montage maker

Enter the template reel-maker category with the one thing every app in it does badly. Every reel maker fills slots with the head of each clip. Reelly’s index knows what is actually in each clip and can put the right moment in the right slot.

The three-second pitch

Every other app takes the first two seconds of your clip. We take the right two seconds. That is a before/after demo — and the before-after creative already hooked at 60.1%.

Versus the alternatives

PathVerdict
Smart slot-filling montage makerRecommended. $4.12M/mo category, reuses every asset, differentiator is the capability already built
Single-asset transformationBiggest by count, no edge. Thin API wrappers; abandons the render pipeline
Full-auto multi-clip directorElva runs it live at 3.89★ and 6,000 downloads
OpusClip-style long→shortHealthiest profile in the census, but long-form single assets: different user, desktop-first
1SE-style constrained diaryMonetises beautifully at $20–33/dl, but 1SE owns the format and it’s a daily habit, not a trip

Where the human stays in the loop

Human decides: which template, which clips are in or out, final reorder, swap a slot’s pick. App decides: which moment inside each clip, which clip fits which slot, effects, timing, beat sync, grade. That maps onto the existing Clip Timeline Editor — which stops being a fixup tool for bad automation and becomes the product surface.


Unresolved

Risks and open questions

What changes about the plan

Week one does not change. Wire Indexing/ into the shipping path so the renderer stops using CMTimeRange(start: .zero, ...). Required under every path.

Weeks 2–8 change. Replace content-driven narrative assembly with slot requirements plus clip-to-slot matching. Smaller, better-posed, more defensible.

The gate gets sharper. Render one real trip two ways — head-of-clip versus matched-and-trimmed, both under the Muse format constraints — and see whether the difference is visible. Better than the original gate, because it isolates the one variable that matters.