What AI Video Actually Costs Per Usable Clip: The Credit Math Nobody Shows You

Every AI video price page quotes the same number: what one generation costs. Almost none of them quote the number that decides your bill. Nobody ships the first take. The unit that matters is the usable clip — the one take out of several that survives the delete key — and its real price is the per-run cost multiplied by the number of takes it took to get there. A clip priced at "a dollar" is a ten-dollar clip if you needed ten attempts, and a clip nobody advertises that way is honest only when the math is on the page.

This guide does that math with our own published prices: the exact credit cost of every frontier video model on Moosky, the reroll arithmetic behind cost-per-usable-clip, and a fair way to compare it against per-tool subscriptions without anyone's marketing math.

A methodology note before any number: every credit figure here is computed from the pricing code that drives the live composers (the per-second and per-token rate constants each composer bills against, plus the 10-credit run base), checked at authoring time (October 1, 2026) against the live billing config, which fixes conversion at 100 credits = $1 USD with tax handled at checkout. Failed runs are refunded automatically, so a provider error costs you a retry, not the credits. The composer shows the exact quote for your selections before you press generate — the tables below are what those quotes are made of.

The unit that matters: cost per usable clip

The formula is embarrassingly simple:

cost per usable clip = credits per take × takes until keeper ÷ 100 ($ per USD at 100 credits)

Two of those three variables are yours to control. Credits per take is set by model, resolution, and duration — all visible before you run. Takes-until-keeper is the hidden one: it depends on how well the model fits the shot, how specific your prompt is, and how willing you are to accept "good enough." Comparison pieces in this space have started publishing this unit openly — one recent buyer's guide put a premium flagship at $9.60 per usable shot against a budget model at $3.50, and the cheaper engine won despite needing more attempts (Swfte, 2026 cost-per-usable-shot comparison). That is the arithmetic we will run on Moosky's own four video engines below.

The Reddit refrain this answers is the one about wasted credits — "even if it takes 20 tries, it still eats up credits." That pain is real on platforms where every attempt is metered and nothing is quoted first. The fix is not promising perfection; it is showing the quote, charging less for attempts, and refunding the failures.

The four engines, priced from the code that bills you

All four models below share one credit balance, quote before running, and refund failures. MiniMax H3 figures are for the R2V (reference-to-video) API composer, H3's current video mode on Moosky.

Wan 3.0 — the motion specialist: camera moves, physics, one-takes up to 30 seconds, billed per second of output.

LTX 2.5 — the talking-head workhorse: native joint audio (dialogue, ambience, SFX), billed per second, flat by resolution tier, 1440P and 4K capped at 10 seconds.

Seedance 2.5 — the premium cinematic engine: up to 30-second single shots, the widest reference capacity, billed on output tokens, which is why its 1080P column moves hard.

MiniMax H3 (R2V) — the reference-directed clip: up to 9 images, 3 reference videos, and 3 audio files steer one performance, billed per second at two resolution tiers, drafts at 768P.

Read the grids as dollar figures by dividing by 100: a Wan 3.0 draft is about half a dollar, a Seedance 2.5 1080P hero render is twelve sixty-seven.

Now multiply, honestly: the reroll math

Take one 5-second draft-tier take on each engine and multiply by attempts-to-keeper:

That last row is the whole argument in one cell. The "twenty tries" scenario people fear is only ruinous at the wrong tier: twenty 5-second Wan drafts at 480P is under ten dollars, while a single unlucky 1080P Seedance hero at one-in-eight is a hundred. The premium engine is not expensive because its badge says so — it is expensive only when you iterate on it. The cheap engine is not the cheap option because its badge says so — it is cheap per usable clip when its quality clears your bar in fewer takes.

Which is why the two questions that actually decide spend are never "which model is best," but: (1) what is the lowest tier at which you can judge this shot, and (2) which engine's strengths fit the shot, so the keeper comes sooner. A talking-head hook kept from two LTX 2.5 takes ($1.60) beats the same hook attempted twenty times on a cinematic model. The model-family decision guide covers which engine wins which shot; this article covers what that choice is worth in credits.

Cost per usable clip against a subscription — a fair method

Per-tool subscriptions are not a scam; they are a different risk distribution. The honest comparison is arithmetic, not attitude:

1. Estimate your usable clips per month. Not generations — clips you ship.
2. Multiply by your realistic cost per usable clip using the table above with your keeper rates for your kind of shot.
3. Compare against the all-in subscription cost — the plan price plus what you pay to get past the months you didn't generate: subscription credit pools expire on the billing date, so sporadic months still cost you the plan. (Pay-as-you-go credit balances do not expire, which is the only structural difference worth writing down.)
4. Check the floor price for volume. If a plan's included credits divided by its price beats the grids above at your keeper rate, the plan wins — buy the plan.

The pattern that falls out of step 3 is stable across the market: daily generators tend to break even on flat monthly pools, and episodic creators — the ones who render a campaign in March and nothing until June — pay the expiration tax. If you are the second kind, per-generation billing is simply the shape that matches your usage. This is the same method the neutral comparison sites publish; we are just the only ones supplying the numbers from inside the billing code.

And on Moosky there is one more line in the ledger worth pinning: new accounts start with 10 signup credits plus a 50-credit Project Director planning allowance, and the first real render gate is a 500-credit wallet floor — so you can script, plan, and get an exact quote for a whole multi-scene video before paying anything. The candid version of what is free is on its own page: what is actually free on Moosky.

Three levers that lower your number this week

Draft at the floor tier, ship at the ceiling. Judge motion, framing, and dialogue at 480P/720P where a take is fifty to eighty credits; spend the 1080P render once, on the approved prompt. Same clip, a fraction of the reroll cost. The Wan 3.0 workflow guide walks the full draft-to-hero escalation.

Fix the still before you animate it. Image models bill in single-digit credits. A keyframe you approve for two cents prevents a five-dollar clip you throw away — the cheapest take is the one you never needed.

Batch variants where takes are cheap, spend once on the hero. Five hook angles at $0.80 each and one $12.67 hero is a twenty-dollar campaign test with a known price — versus one premium attempt at a hook nobody validated. The founder ad-creative workflow runs this exact budget end to end.

Honest-credit FAQ

Do failed generations cost credits? No. A run that fails is refunded automatically — the credits return to your balance. You pay for renders, not for provider errors.

What decides the price of one take? Three things, all yours: the engine, the resolution tier, and the duration. Each composer multiplies its billed rate by those choices, adds the flat 10-credit run fee, and shows the total before you confirm. Nothing else changes the quote — aspect ratio and frame rate do not move LTX 2.5's bill at all, and audio is included in the rate rather than billed beside it.

Are credits refundable if I just don't like the clip? No — a take you delete was still rendered, and that compute is billed. That is exactly why this article pushes the draft-tier loop: the refund covers failures, and cheap drafts cover taste.

Can I compare models on the same prompt without three subscriptions? Yes — one credit balance runs all of them, so a three-engine bake-off on the same brief costs the sum of three draft-tier takes (about two dollars at the rates above) instead of three monthly plans.

Do unspent credits expire? No. Credits are bought, not subscribed: the balance carries until you spend it. The 500-credit floor on starting a full Project Director render is the only threshold to plan around.

What is the cheapest honest way to find out what my clips cost? Plan the whole project in Project Director on the free planning allowance: it prices your scene list and shows the quote before rendering. You leave with the exact credit number for your finished video, spend nothing, and decide from the quote on the credits page.

The number to plan against

The right budget line is never "price per generation." It is credits per take, times your takes to keeper, on the engine whose strengths fit the shot — and drafts at the floor tier until the shot is earned. Run that math on your own brief: start free at moosky.ai/start, compare the engines on one balance — the Seedance 2.5 page and the Wan 3.0 page have the capability details — and judge the finished number by the clip you ship, not the clip you paid for.