ReelPulse Community Cut · Field note 001 · Updated 03 Sep 2026

Video can
beat the clock.Can it hold a story?

H3 Max moves AI video into the playback window. ReelPulse is starting smaller: can ordinary site visitors finish one vertical scene, choose its next beat and return for the reviewed result?

What is real

Generation throughput

Five-second 768p output is publicly reported below playback duration, with native synchronized audio.

What is unproven

Narrative continuity

A fast clip does not establish stable characters, props, voice or plot state across dozens of requests.

The ReelPulse wedge

Community Cut

Let site visitors vote asynchronously, generate only the winning continuation and publish the evidence—not an open-prompt free-for-all.

The technical opening

The wait moved inside the shot.

fal says H3 Max is post-trained from MiniMax H3 and co-optimized with its inference engine. At 768p, a five-second clip renders in under three seconds; longer durations scale until a 15-second clip takes around 15 seconds.

5–15 sec

Generation duration

A Community Cut needs only one controlled continuation after the vote closes.

768p / 24 fps

Default fast route

Enough for a mobile pilot episode; it is not the base H3 model's 2K delivery route.

Native audio

One-pass output

Picture and sound arrive together, but voice identity and audio cuts still need continuity scoring.

$0.08 / sec

Published list rate

One generated hour is $288 before rejects, extra buffer or the rest of the production stack.

Infinite Slop

Observed snapshot

2,278 displayed watching

Free-text prompt queue and an ad-buy route were visible in ReelPulse's September 3 snapshot.

fal.live

Observed snapshot

23–80 displayed per channel

Voting cards and several themed channels were visible; the selected broadcast was reconnecting during the snapshot.

RENOISE LIVE

Observed snapshot

Feed lost during check

The polished director-and-cast interface did not provide a stable programme in ReelPulse's September 3 check.

Displayed viewer counts are point-in-time interface observations, not verified active users, retention or independent audience measurements.

Episode economics

One winner. One generated cut.

At 0.08 dollars per generated second, there is no reason to fund an endless stream before demand exists. One reviewed setup and one winning continuation can test the behavior for under five dollars.

Community Cut economics

Model one reviewed setup and a few attempts at the single winning continuation.

Uses the conservative 768p list rate of $0.08/generated second. Streaming infrastructure is unnecessary for this asynchronous format.

Modeled cost per episode

$4.50

45 submitted seconds · 3 attempts at the winning cut · 25% allowance

Raw model
$3.60
Per visitor
$0.045
Four episodes
$18.00
Episode cap
$5.00

Inside the stop rule

$0.50 remains beneath the $5 episode cap.

RP-CUT-001

One audience. One winning cut.

Test whether ReelPulse visitors will finish a short scene, choose its next beat and return for the generated result—without running a continuous AI livestream.

Enter episode 001 →
  1. DAY 0

    Watch

    Publish one reviewed 30-second setup and unlock voting only after the viewer reaches the decision point.

  2. DAYS 1–7

    Vote

    Offer three pre-moderated story beats. Record one choice per browser and hide the running tally to limit social proof.

  3. LOCK

    Generate

    An editor locks the winner, submits one controlled H3 Max prompt and stops if the episode reaches its $5 generation cap.

  4. NEXT CUT

    Return

    Review the output before publication, notify opted-in viewers and measure how many return for the result.

Page visitor to completed vote

Video starter to decision-point reach

Result-notification opt-in

Return visit for the winning cut

Rejected and failed generation attempts

Spend per published usable second

Why not open chat first?

Votes constrain chaos.

Public prompt queues immediately create moderation, rights and story-coherence problems. Episode 001 offers three pre-moderated beats, hides the running tally and generates only the winner after an editor locks the vote.

Why short drama fits

Every beat already asks for the next.

Research with 28 micro-drama creators describes compressed, audience-responsive production loops. Short drama supplies the hooks, reversals and recurring state that random infinite video lacks.

Pilot gate · not a launch promise

Watch the setup.
Own the next cut.

Episode 001, Floor 27, is ready in pilot preview. ReelPulse will record the vote, generate only the winning continuation and disclose attempts, failures and cost.

Watch and choose →

Evidence and limits

Source the claim. Test the system.

Product speed, ranking and price statements remain vendor claims until ReelPulse completes a recorded run. Live-site observations were checked on September 3, 2026 and can change without notice.

H3 Max FAQ

Before opening the vote.

Is MiniMax H3 Max faster than real time?

fal reports that a five-second 768p H3 Max clip renders in under three seconds, with about 2.5 seconds of backend inference time. It also says a 15-second clip takes around 15 seconds. Queueing, prompt expansion, transfer and playback assembly still need to be measured end to end.

How much does a ReelPulse Community Cut episode cost?

At fal's published 768p list rate of $0.08 per generated second, a 30-second reviewed setup plus three attempts at one five-second winning cut has $3.60 of raw output cost. ReelPulse adds a 25% allowance and enforces a $5 per-episode stop rule.

Can H3 Max make an infinite AI short drama?

It can generate clips quickly enough to support a buffered shared broadcast, but an infinite coherent drama is not established by the public speed claim. Cross-request character identity, wardrobe, props, geography, voice and story state all require a controlled continuity test.

What is the difference between MiniMax H3 and H3 Max?

MiniMax H3 is the base multimodal model announced by MiniMax with up to 15-second 2K generation and broader reference and editing modes. H3 Max is fal's post-trained variant optimized around its inference stack for faster 480p and 768p generation.

Why does ReelPulse use asynchronous voting instead of a livestream?

An asynchronous vote lets ordinary site visitors participate without arriving at the same time. Pre-moderated choices reduce unsafe requests and narrative drift, while generating only the winning continuation keeps the experiment below a fixed episode budget.