Kling motion control duration_seconds 1: shortest clip is $0.1575
Kling motion control on Sume takes a 1-second reference, so a probe costs $0.1575. kling-3 starts at 4 seconds ($0.56 silent). Test before you pay for 30 s.

Kling 3.0 motion control on Sume accepts duration_seconds from 1 to 30, so the cheapest real test is a 1-second reference at 1 x $0.1575 = $0.1575. The kling-3 video id cannot go below 4 seconds, which is $0.56 silent. If you want to prove a still and a reference work together before you pay for 30 seconds, the motion route is the cheaper probe.
Why a probe is worth it
Motion control is the one Kling surface on Sume where the reference decides everything about the output. A bad crop, a character that is cut off at the frame edge, or an orientation setting that does not match the still, will spoil a long clip as surely as a short one. At 30 seconds that mistake costs $4.725. At 1 second it costs the price of a coffee stirrer.
The probe does not need to look good. It needs to show whether the face is kept, whether the body follows the reference, and whether the edges of the frame hold. Those three checks are visible in one second of video.
The price ladder
The reserve is ceil(duration_seconds) x $0.1575 and the API rounds a fraction up to the next whole second. Here is the ladder for a probe-then-ship plan.
| Step | Declared seconds | Motion control at $0.1575/s | kling-3 silent at $0.14/s |
|---|---|---|---|
| Smallest probe | 1 | 0.1575 | not offered (4 s floor) |
| Floor of kling-3 | 4 | 0.63 | 0.56 |
| Short test | 5 | 0.7875 | 0.70 |
| Full reference | 30 | 4.725 | not offered (15 s cap) |
How to cut the probe
Take the first second of the reference with POST /v1/video-trim. It needs a source on media.sume.com and an Idempotency-Key, and the output may be as short as 0.2 seconds. A trim is $0.02 per job, so the probe plus the trim is about $0.18. Then submit the trimmed URL with duration_seconds: 1.
Pick the second that has the hardest motion in it, not the first one. A turn, a raised arm or a step shows the failure that a still pose hides.
One caveat
The output length is set by the reference you send, not by the number you declare. Declare what you sent. If you send a 5-second clip and declare 1, you have asked for a reserve that is lower than the real job, and the mismatch is your bug to fix, not the API's. Measure the trimmed file and declare its rounded-up length.
What a 1-second probe cannot tell you
A one-second clip will not show drift over time, a lip that slowly loses sync, or a hand that bends oddly at second twenty. It answers the first-order questions only. After it passes, run a 5-second clip at $0.7875 on the hardest part of the reference, and only then pay for the full length.
The ladder costs $0.1575 + $0.7875 + $4.725 = $5.67 in all if every step passes, compared with $4.725 if you go straight to 30 seconds. The extra $0.945 is the price of catching a bad setup early, and it pays for itself the first time that a step fails.
Related posts
More in Pricing
- Largest single AI video job each Sume model allows, by price
The most one Sume video job can reserve ranges from $1.13 on H3 to $42.65 on Seedance 2.5 at 30 seconds and 1080p. Eleven rows.
- Listing narration in 3 languages: 1,200 characters, $0.17 on Sume
A 1,200-character listing description spoken in three languages is 3,600 characters of Sume TTS at $0.0475 per 1,000, which is $0.17 per listing.
- Lyria 3.5 costs $0.08 a song at Google; Sume charges $0.125
Google lists Lyria 3.5 at $0.08 a song; Sume's Music Router bills a fixed $0.125 per accepted generation. The gap is $4.50 per 100 songs. What it covers.
- MAI-Transcribe-2-Streaming $0.54/hour vs Scribe v2 Realtime $0.39
MAI-Transcribe-2-Streaming is $0.54 per hour (intro, through 2026), Scribe v2 Realtime $0.39, AssemblyAI Realtime $0.45. Sume has no live STT.
Written by Sume