Kling motion control silent clip with a Lyria music bed for a Short

To avoid carrying the motion video's soundtrack into a Kling motion control clip, send keep_original_sound: false, which returns a silent clip, then add a new track from the Music Router and join the two with a Timeline render. For a 12-second clip the three steps cost about $2.115 on Sume.

Why bother

keep_original_sound defaults to true. If the motion reference is a dance clip filmed with a popular song playing, the output can inherit that audio. YouTube's help page says a Short over one minute with an active Content ID claim of any type is blocked globally, and that nothing changes for Shorts under one minute (read 2026-10-07). A 12-second clip sits under that line, but a clip you later extend or join into a longer Short does not. Starting from silence avoids the question.

Rights are your responsibility either way. A silent clip only removes the original track; it does not clear the motion video itself.

Step 1: the Kling job

POST /v1/kling/3.0/motion-control takes image_url or an avatar (not both), a required motion_video_url, duration_seconds from 1 to 30, an optional prompt that steers appearance only, and character_orientation of video (default) or image. The price is ceil(duration_seconds) times $0.126 times 1.25, which is $0.1575 per second. It is reserved at admission, captured on completion and refunded on failure. The duration_seconds value is the reservation basis; the motion video's length sets the output length. See the silent clip note for the field in isolation.

Step 2: the bed

POST /v1/music-router/generate with model: "sume/music-auto" and a prompt. It rejects duration and duration_seconds, so put the length in the prompt text: "a 12-second instrumental, no vocals". The price is a fixed $0.125 per generation. Check the length you actually received before the join, and set the Timeline's duration_seconds to match it.

Step 3: the join

A Timeline render needs one audio spine and at least one video slot, with all URLs on media.sume.com. Billing is $0.10 per output minute rounded up, so a 12-second render is $0.10.

curl -X POST https://api.sume.com/v1/timeline-1.0/render \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: kling-bed-001" \
  -d '{
    "audio": {
      "url": "https://media.sume.com/artifacts/artf_demo/bed.wav",
      "duration_seconds": 12
    },
    "video": [
      {
        "source_url": "https://media.sume.com/artifacts/artf_demo/kling-silent.mp4",
        "start": 0,
        "duration": 12
      }
    ]
  }'

Picking the motion reference

The output motion comes entirely from motion_video_url, so pick a clip with one clear performer, a steady camera and the full body visible if the still shows a full body. character_orientation defaults to video; image is the other value. The prompt is for appearance only, so do not try to direct movement through it. Strict validation also means a stray model or reference_image_urls field is rejected rather than ignored.

Keep duration_seconds honest. It is the basis for the reservation: declare the length of your motion video, rounded up, and the price follows that value.

What to check

Probe the silent clip with video inspect and frames: false; its probe.has_audio should be false. The bed should be quieter than speech if you add any, and soundtrack.duck_db exists for that, but it needs a real audio spine. For a clip over 30 seconds see the 60-second two-job pattern.