Kling motion control or Recast: a photo moves, or people get swapped

Both tools take a video and a photo, and both can make a person in your photo appear to do what someone in your video does. The difference is the starting point. Kling 3.0 motion control starts from your still photo and drives it with the motion of a reference video, so the result is a clip of the photo's subject performing that movement. h3-max-recast starts from your video and replaces the people in it with the people in your photos, keeping the original camera, cuts and sound. Pick Kling when the photo is the scene, and Recast when the footage is the scene.

On Sume the two also bill differently: motion control at about $0.1575 per second, Recast at about $0.375 per second at 768p (both list prices times 1.25, read 2026-10-03).

What each one takes in

Kling motion control is a Sume model-run surface of its own, not a Video Router id. It takes one visual source, either a public image_url or a ready avatar, plus a required motion_video_url that drives all output motion. You declare duration_seconds (1 to 30) so Sume can price the job; the provider has no duration knob and the output length follows the driving video. keep_original_sound defaults to true and character_orientation is video by default, or image if you want the photo's facing to win.

Recast is a Video Router id. It takes a source video_url and 1 to 4 reference_image_urls, one per person, and the output keeps the source's motion, camera, cuts and audio. The source must be 5 to 30 seconds with no shot over 15, and Recast's resolution is 768p or 1080p.

Worked cost for a 10-second clip

At 10 seconds the arithmetic is simple. Kling motion control reserves 10 times $0.126 times 1.25, which is $1.575 and rounds up to $1.58. Recast at 768p is 10 times $0.30 times 1.25, $3.75, and at 1080p $5.63. So motion control is less than half the price per second of Recast at its cheaper tier. The comparison is only fair when both give you what you need, and they usually do not: a motion-control clip has no original camera or cuts from your footage, and a Recast clip has no way to make a photo do a new movement.

Use that to decide on need first and price second. If the deliverable is "the existing ad, but with a different person", only Recast fits. If the deliverable is "my mascot photo doing this dance", only motion control fits.

A motion-control request, for contrast

The body below is the documented shape. There is no reference_image_urls or model field; the URL itself identifies the model, and the strict body rejects Video Router vocabulary.

curl -X POST https://api.sume.com/v1/kling/3.0/motion-control \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: kling-mc-001" \
  -d '{
    "image_url": "https://example.com/mascot.png",
    "motion_video_url": "https://example.com/dance.mp4",
    "duration_seconds": 10,
    "keep_original_sound": true,
    "character_orientation": "video",
    "mode": "async"
  }'

Picking in practice

Run through three questions. Does the output need your original camera work and cuts? Then Recast. Does the output need a single character from a photo to move in a specific way? Then motion control. Is the person in the source a real individual whose likeness you are replacing? Then make sure you have the right to use both the source performance and the replacement face; Sume documents the tools, not your clearance to use them.

For the many-people case, Recast takes up to four photos in one call, one per person, while motion control is a one-image surface. For a deeper look at Kling's prices at 30 seconds see Kling motion control price per second and the cost of a 30-second clip, and for reference-video requirements see Kling motion control reference video requirements.

What neither does

Neither is a free-form prompt edit and neither writes new dialogue. Both keep or take audio from the video you supply rather than creating a voice for the new person. If you need speech from a still image with your own audio, that is the lip-sync surface, which has its own 5 to 14.8 second window. The practical outcome is a pipeline: choose the motion tool, then add captions or a voice step afterward.