Why are AI videos so expensive? What drives the cost
On Sume, AI video is billed by output length at a rate that rises with resolution, and discarded takes are billed too. Why it adds up, how to cut it.

AI videos get expensive because you pay for every second of output, the per-second rate climbs with resolution on most models, and every finished take you throw away is billed like the one you keep. On Sume, before the agent fee, 10 seconds of wan-3.0 is priced at $1.25 at 720p and $2.50 at 1080p, and 10 seconds of seedance-2.5 at 1080p 16:9 at $14.22.
This post explains how Sume prices a clip, from its pricing code and the Video generation docs, read on 2026-09-27. It does not cover what providers spend to run their models. Per-model rates are in How much does an AI video cost to generate?
What makes an AI video cost more?
Every model is billed at the provider's list price × 1.25, reserved when you submit, plus a 5.5% agent fee by default. These move the amount:
| Driver | What it does to the price | Example |
|---|---|---|
| Length | Every model bills per second of output, or per video token, which also grows with seconds. | wan-3.0 at 720p: $0.63 for 5 seconds, $1.25 for 10 |
| Resolution | Most models have a higher per-second rate at a higher resolution; Seedance counts more tokens per frame at higher resolutions. | gemini-omni-flash-1.1, 10 seconds: $1.25 at 720p, $3.75 at 4K |
| Sound | kling-3 has a higher rate with sound, and sound is on unless you send generate_audio: false. | kling-3, 10 seconds: $1.40 silent, $2.10 with sound |
| Model | Each model has its own rate. | 10 seconds at 720p: $0.13 on grok-imagine-video-1.5, $5.78 on seedance-2.5 in 16:9 |
| Reference video | On Seedance, Sume's current estimate adds 15 assumed input seconds to the token count, then multiplies by 0.6. | seedance-2.5, 10 seconds at 720p 16:9: $5.78 without, $8.67 with a reference video |
| Discarded takes | A completed take is billed whether you keep it or not. | Three takes of a clip cost three times one take. |
Why does resolution raise the price so much?
Because the rate is set per resolution. On wan-3.0, a second costs $0.125 at 720p and $0.25 at 1080p. On gemini-omni-flash-1.1, it costs $0.125 at 720p and $0.375 at 4K.
Seedance prices pixels directly. Sume's pricing code currently counts 2.25 times as many video tokens in a second of 1080p 16:9 as in 720p, and seedance-2.5 also charges a higher per-token rate at 1080p, so its 10-second 16:9 clip goes from $5.78 at 720p to $14.22 at 1080p. How much does an AI video cost to generate? shows the token math.
Do failed or rejected takes cost money?
A failed job doesn't: Sume refunds the reserved amount, as Do failed AI video jobs cost money? explains.
A take you reject is different. It completed, so it is billed like the one you keep, and once generation work has started a job can no longer be canceled: the API returns 409 job_generation_already_started and the job runs to completion. So the number of takes you need, not only the rate, sets what a usable clip costs.
How do I spend less on AI video?
Pay for fewer seconds, a lower rate, or fewer takes:
- Test a prompt short and small.
wan-3.0accepts clips from 2 seconds and lists 480p, and a 2-second 480p test reserves $0.13. - Generate at the resolution you will publish, not above it.
- Turn sound off where you don't need it: on
kling-3, sendgenerate_audio: false. - Settle the opening frame as a still first, then send it as the clip's first frame, so each take starts from a picture you have already approved.
- Reuse finished clips instead of regenerating them. To make a clip you already have bigger, Video Upscale 1.0 is listed at $0.009 per video second on API pricing; see the AI video upscaler API.
- See the price before you pay, and cap it: estimate AI video cost before you run and set spend caps on unattended runs.
Sources
Related posts
More in Pricing
- How Sume pricing works: plans, one wallet, published model rates
Sume plans set access and concurrency. Usage draws from one prepaid wallet at each model's published USD rate, for generation, the Agent, Formats, and the API.
- AI avatar video API pricing: cost per second and per minute
Sume bills AI avatar video per second by quality tier, with separate rates when you send a product image. Per-minute costs for standard, plus, and max.
- AI video generation cost per video: what one Sume API run cost
To see what one Sume run or video job cost, call GET /v1/usage with run_id or job_id and read debited_usd: the wallet deduction, agent turns included.
- Insufficient credits error 402 on the Sume API: add funds and retry
A Sume 402 insufficient_credits means the balance cannot cover the request's estimated cost, so nothing ran. Check GET /v1/balance, add funds, then retry.
Written by Sume