How a video take is priced

The claim

A price on a button can only be honest if its formula is the provider’s. Close is not enough: a button that rounds a short count says less than the take costs.

The short answer

Each provider prices a take its own way — by the second, by the token, per image, per character — and the desk carries each formula. It rounds the estimate to a ten-thousandth of a dollar; the button rounds it up to the cent. Change a setting in the calculator below: it prints the formula it used.

The instrument

Instrument

Price a take

The desk’s built-in catalogue and its own functions, imported into this page. Change anything: the formula follows.

Model

Seedance 2.5By the token · OpenRouterSeedance 2.5Per second · BytePlus, directSeedream 5.0 ProPer image · BytePlus, directEleven v4Per 1,000 characters · ElevenLabs

Length

Include the take’s first frameClear it to replay the count the desk used before the fix.

The desk’s formula

1,280 × 720 px × (24 × 12 + 1) frames ÷ 1,024
=260,100 tokens
×$0.0000107 a token
=$2.7831

Button preview · nothing is sent

The button rounds the estimate up to the cent. An estimate, not an invoice: the provider’s own cost replaces it when it reports one.

Recorded These are the settings of a real bill: $2.78307 on October 6, 2026, through OpenRouter.

Rates from the desk’s catalogue (models.ts, checked against the providers’ documentation on 2026-10-03). The desk reads OpenRouter’s rates live; here the token rate is the one the bill of 2026-10-06 shows.

It opens on a real bill. Clear “Include the take’s first frame” to see what the desk said before the fix, and change the length: on most lengths the old count puts the button under the bill.

How each provider prices a take

The calculator applies four formulas from the desk’s built-in catalogue:

BytePlus, direct, bills Seedance 2.5 by the second, at a rate per resolution: $0.103 at 480p, $0.231 at 720p, $0.569 at 1080p.
OpenRouter bills the same model by the token: width × height × frames ÷ 1,024, where a clip of n seconds has 24 n + 1 frames — its first frame, then 24 a second.
Seedream 5.0 Pro costs $0.045 an image at 1K, $0.09 at 2K, plus $0.003 for each reference image.
ElevenLabs counts characters: $0.30 per 1,000 in the desk’s catalogue.

The 289th frame

During a test run of the desk on October 6, 2026, 12 seconds of Seedance 2.5 at 720p, 16:9, through OpenRouter were billed $2.78307. At $0.0000107 a token, that is 260,100 tokens — 1,280 × 720 pixels × 289 frames ÷ 1,024.

The desk had counted 288 frames, 24 a second. Its estimate came to $2.7734, and the button said $2.77 for a take billed $2.78 to the cent. The difference is one frame: 900 tokens, $0.00967.

Twelve seconds of Seedance 2.5 at 720p, priced by the token
Count	Frames	Tokens	Amount	What the button said
The desk, before the fix	288	259,200	$2.7734	$2.77 under the bill
The desk, today	289	260,100	$2.7831	$2.79 covers the bill
The bill recorded	289	260,100	$2.78307	—

Rounded up to the cent, as the buttons are today, the old count would still say $2.78 — under the $2.78307 bill. Counting the first frame is what makes the button cover the take.

Since then videoTokens counts fps × seconds + 1. The same take now estimates at $2.7831 and its button says $2.79. Rounding up alone would not have saved it: the old count, rounded up, still says $2.78.

Why a vertical take costs the same

At a named resolution, video models keep the pixel area of the 16:9 frame across ratios — at 720p, 1280 × 720, 960 × 960 or 720 × 1280. The desk’s dimensions keeps that area, so a vertical take is priced like a horizontal one. Only a ratio that does not divide evenly moves the count, slightly: try 21:9 in the instrument.

Limits

The rates are the desk’s built-in catalogue, checked against the providers’ documentation on 2026-10-03. The desk reads OpenRouter’s rates live; this page cannot, so its token rate is the one the bill shows.
An estimate is not an invoice. The provider’s own cost replaces it in the project when the provider reports one.
ElevenLabs bills characters on the creator’s own plan; the desk’s rate is for estimating.
One bill checked the token formula to the token. The per-second and per-image rates are the providers’ published prices, not bills we measured here.

What it means for Kiju

The formulas live in one file of the engine, core/src/models.ts, and this page imports that same file: the figures above are the desk’s figures, computed in your browser.
Where a provider bills by the token, a bill that disagrees with the formula is a bug in the desk. This one became a test and a fix.

Sources

Kiju’s engine: core/src/models.ts — estimateVideo, estimateImage, estimateVoice, videoTokens, dimensions, upToCents, the catalogue BYTEPLUS_MODELS and ELEVEN_MODELS.
The bill: a test run of the desk on 2026-10-06, recorded in videoTokens’s own comment.