Test Kling motion control on 3 s: $0.47 before a $4.73 run

Run a 3-second motion video through Kling motion control before you commit to a 30-second one. On Sume the route reserves ceil(duration_seconds) x $0.1575, so a 3-second test holds $0.4725 and a 30-second run holds $4.725. The rate is the fal list of $0.126 per second times a 1.25 margin. A test this small tells you whether your still and your motion video work together, for under 50 cents.

What the route takes

The request goes to POST /v1/kling/3.0/motion-control (the same job is also reachable on /v1/avatar-1.0/motion-control and two model-run paths). It needs an image_url or a saved avatar, and a motion_video_url. duration_seconds runs from 1 to 30 and sets only the amount reserved; it is not sent to the provider, so the length of the output follows the motion video. prompt steers appearance, not motion. keep_original_sound defaults to true, and character_orientation is video by default or image.

Steps for a cheap test

• Cut a 3-second piece of the motion video that has the move you care about.
• Pick a still with the character fully visible, matching the pose of the first frame of the motion.
• Submit with duration_seconds of 3, and keep_original_sound set to what you want in the output.
• Poll GET /v1/jobs/:id/status; read /result when it completes.
• Check that the character matches the still and that the move transferred. Only then submit the long version.

The cost helper

A few lines of Python reproduce the table and help you budget batches.

import math

RATE = 0.126 * 1.25

def reserve(seconds):
    if not 1 <= seconds <= 30:
        raise ValueError("duration_seconds must be 1 to 30")
    return round(math.ceil(seconds) * RATE, 4)

for s in (1, 3, 10, 30):
    print(s, reserve(s))

Why a short motion clip is the right test

Motion control is the one place where the input you pick matters more than the settings. The model copies the move from the motion video onto the character in the still, so the three questions to answer are whether the character keeps its look, whether the body proportions survive the move, and whether the background stays plausible. All three show up in 3 seconds. A 30-second run answers nothing more about those, and it costs ten times as much if it fails. If a test passes, run the full length with the same still and the same orientation setting, and keep the test as your reference for what good looks like. Two test runs of 3 seconds cost $0.945 in all, which is still under a fifth of one 30-second run.

What Sume does not do

Sume does not forward duration_seconds to the model, so you cannot shorten a long motion video by sending a small number: you would reserve less than the job might use. Keep the number equal to the motion video's length, rounded up. The prompt cannot change the motion either; for that you change the motion video.