Kling 3 two 15 s jobs make a 30 s clip: $4.30 silent, $6.40 with sound

Kling 3 on Sume stops at 15 seconds, so a 30 second Kling-style clip is two 15 second jobs joined on Timeline. That costs $4.20 for the two clips with audio off or $6.30 with audio on, plus $0.10 for a Timeline render of up to one minute: $4.30 or $6.40 in all. A single 30 second job on Seedance 2.5 at 720p is $17.33, and on Wan 3.0 at 720p it is $3.75.

Kling 4.0 is listed with a 3 to 30 second range, according to the fal explainer (read 2026-10-05), and its full launch is due in October. Until a Kling 4.0 model is in the Sume catalog, the nearest Kling model is kling-3 (Kling Video v3 Pro).

What Kling 3 gives you

The Sume catalog describes kling-3 as text-to-video and image-to-video with an end frame, 4 to 15 seconds, 720p or 1080p, in 16:9, 9:16 or 1:1, with optional audio and no reference_*_urls fields. Pricing is per second: $0.14 with audio off and $0.21 with audio on, after the 1.25 margin. A 15 second job is therefore $2.10 or $3.15.

Two jobs do not make one continuous shot by themselves. The second job starts from whatever its prompt says. To make the join feel continuous, take a still from near the end of the first clip and send it as the first_frame of the second, which is the first and last frame mode that Kling 3 lists.

A 30 second clip, five ways

When two jobs beat one

The table shows why a single long take is not always the best price. Seedance 2.5 costs the same whether you run one 30 second job or two 15 second jobs, apart from the join, so the long take wins there. For Kling 3 the cap forces a join, but the clip price is still about a quarter of the Seedance figure.

Retries are the other reason to prefer two jobs. If the second half fails or looks wrong, you rerun a 15 second Kling 3 job at $2.10 or $3.15, not the whole clip. A 30 second Seedance 2.5 job that you have to rerun costs $17.33 again. For a plan with several iterations per shot, the cheaper unit of retry is worth more than the saving on the join.

Keep the two prompts close. Repeat the subject, the lighting and the lens in both, change only the action, and name the camera move you want to continue. If you use audio, expect a seam in the sound at the join, because each job makes its own track, and consider generating the clips silent and adding one music or voice track on the Timeline audio spine.

Take the join point seriously. Pull the frame with the Sume video frames endpoint, which accepts up to 24 instants in at[], each between 0 and the clip duration, and fps no higher than 2. Choose an instant a little before the end, because the last instant must be less than the duration.

Totals and the second job

This prints the cost table and builds the frame_images entry for the second job. It sends nothing.

import json

RATE = {"off": 0.14, "on": 0.21}
JOIN = 0.10

for audio, rate in RATE.items():
    clips = round(2 * 15 * rate, 2)
    print(f"kling-3 audio {audio}: clips ${clips:.2f}, total ${clips + JOIN:.2f}")

second = {
    "model": "kling-3",
    "prompt": "Same scene, the camera keeps moving to the left",
    "duration": 15,
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "frame_images": [{
        "type": "image_url",
        "image_url": {"url": "https://media.sume.com/artifacts/artf_demo/tail.jpg"},
        "frame_type": "first_frame",
    }],
}
print(json.dumps(second, indent=2))

Join and check

Render the join with Timeline: two slots of 15 seconds each with a short fade, audio.duration_seconds of 30 and an output size that matches the clips. Timeline bills $0.10 per output minute, rounded up, so a 30 second cut is one billable minute. If you want a continuous voice track over both halves, use the audio spine rather than the clips' own audio.

Check the final price on each job's usage.cost, and send an Idempotency-Key on each of the three requests so a retry cannot create a second paid job. The video generation docs list the field.