Kling 3.0 Motion Control price per second and a 30-second clip

Kling 3.0 Motion Control costs $0.1575 per output second on Sume. fal's page for Kling Video v3 Standard Motion Control says a request costs $0.126 per second, and Sume prices the route at that list times 1.25. Output length follows the driving video, so a 10-second motion reference costs $1.575 and the 30-second maximum costs $4.725.

The billed length is not the length of your still image or your prompt; it is the motion video you send, rounded up to a whole second. This post lays out the per-clip numbers, the one field that decides your reservation, and the two cases where the submit fails before you are charged.

What does the vendor list, and what does Sume add?

fal's model page, read on 2026-10-03, states one price line: $0.126 per second. It does not list a duration limit in the text read. Sume's own constants record the same $0.126 per output second as the provider list for this route, and its catalog entry for kling/3.0/motion-control multiplies it by the standard 1.25 margin, giving $0.1575 per second.

The difference is $0.0315 per second. On a 30-second clip that is $0.945 of margin on $3.78 of provider cost.

Which field sets the reservation?

The OpenAPI schema for POST /v1/kling/3.0/motion-control marks two fields as required: motion_video_url and duration_seconds. The second is described as the motion video duration used to reserve credits at admit, computed as the ceiling of the seconds times the per-second rate times the margin, with a maximum of 30. Output length follows the driving video.

That makes duration_seconds the cost lever. If you declare 12 but the motion video is 12.4 seconds long, you have under-declared the length you meant to pay for; round up yourself. The still image can be a public URL (image_url) or a ready avatar (avatar_id or avatar_handle), but not both. Neither changes the rate.

What does one clip cost?

Each row is whole seconds times $0.1575 on Sume and $0.126 on the vendor page. They are list arithmetic, not a measured run.

What happens to a 12.4 second reference?

The duration_seconds field is a number, and the docs say the reservation uses the ceiling of it, so 12.4 bills as 13 seconds: 13 x $0.1575 = $2.0475, against $1.953 if billing were exact to the tenth. If your driving videos come from a phone, trimming them to whole seconds before you submit removes the rounding cost and also keeps the declared length equal to the real one.

• Maximum motion video length is 30 seconds; a longer file must be cut first.
• keep_original_sound defaults to true, so the driving clip's audio comes through unless you set it false; sound does not change the rate.
• Reservation is exact micro-dollars, shown rounded up to cents on the balance.

How does that compare with Sume's other per-second routes?

Motion Control sits in the middle of Sume's per-second video prices. The rates below come from Sume's catalog; they bill different things, so read the unit column before comparing. Motion Control and recast bill the length of the video you give; lip-sync and Fabric bill the length of the audio you give.

If the job is to make a still image move like a reference performance, Motion Control is the route that takes a driving video. If the job is to make a still image speak a recording, the audio-driven routes are cheaper per second but have their own audio-length windows.

When do you get charged, and when not?

Sume reserves the amount at submit, captures it on a completed job, and releases or refunds it on a failed job or a cancel before generation starts. A balance that cannot cover the reservation returns 402 insufficient_credits and starts no job. A full workspace queue returns 429 queue_full, and the failed admission releases its reservation; the generation admission docs say so directly. Once generation has started, cancel returns 409 job_generation_already_started and the job finishes or fails on its own.

To see what a run actually cost, read GET /v1/usage?job_id=...: its summary shows debited, held and refunded amounts for that one job, and the final flag turns true when no hold is open.

If you only need part of a long source, trim it before you submit. Motion Control bills the seconds you send, rounded up, and the 30 second cap means a longer source has to be cut either way, so cutting to the exact moment you need keeps the reservation as small as it can be.