Kling motion control vs H3 Max Recast: $4.73 vs $11.25 per 30 s

A 30-second Kling 3.0 motion control job costs $4.725 on Sume ($0.126 list x 1.25 = $0.1575 per second), and a 30-second H3 Max Recast at 768p costs $11.25 ($0.30 x 1.25 = $0.375 per second). Motion control costs about 42 percent of Recast per second.

But the price gap is a result of what each route does. Motion control gives your character a movement from a reference video; Recast keeps the source video and puts different people into it. Choose by the input you have, then check the price.

The two inputs

Motion control takes a still (or a ready avatar) and a motion_video_url. The reference video drives all of the output motion and the output length. Sume's docs say the prompt is optional guidance for appearance only, and character_orientation is video by default or image. Duration is 1 to 30 seconds.

Recast takes a source video_url and 1 to 4 reference photos. The Sume docs say the output keeps the source motion, camera, cuts and sound. The source is 5 to 30 seconds, with no shot longer than 15 seconds, and the resolution is 768p by default or 1080p.

Price by length

The table lists Sume billable amounts at the documented rates. Kling uses a single rate; Recast has two.

When each is the wrong tool

If your goal is to put a different spokesperson into an existing ad, motion control does not help. It generates new footage from your still and the motion reference, and the original scene does not remain. If your goal is to make a mascot do a dance, Recast does not help either, because it needs a source video that already has people in it.

There is one overlap. A source video with a person in it can serve as the motion reference for Kling, and the same video can serve as the source for Recast. In the first case you get the still character performing the movement, in a new scene built by the model. In the second case you get the original scene with the person replaced. Sume's own face swap beta worker uses the Kling motion queue for its motion stage and muxes the original audio back at the end.

• Want the original background and camera: Recast.
• Want a character that is not in any video yet: motion control.
• Want the soundtrack of the source: both can keep it; check keep_original_sound on motion control.

Run a cheap test first

For either route, test on the shortest clip that shows the problem. A 5-second motion control test is $0.79 and a 5-second Recast at 768p is $1.88. Each route refunds the reservation if the job fails, per the docs, and each submit accepts an Idempotency-Key, so a retry after a timeout does not bill twice.

A worked scenario: one dance, three characters

Suppose you have one 15-second dance reference and three mascot stills. With motion control that is three jobs at 15 seconds each, or 3 x $2.37 = $7.11 after rounding each job up to the cent. The three outputs share the same choreography because they share the motion reference, and each character is generated from its own still.

With Recast you would need a source video that already contains a person doing the dance, and you would replace the person with one photo per job. At 768p, three 15-second swaps cost 3 x $5.63 = $16.89. The result keeps the original scene, camera and cuts, which is what you want for a swap and not what you want for a mascot, which has no source scene at all.

What both routes have in common

Both are job-based. You submit, receive a job id, and poll GET /v1/jobs/:id/status or use a webhook. Both reserve the price at admit, capture it on completion and refund it on failure. Both accept an Idempotency-Key, which makes a retry after a timeout safe. Neither takes a seed-style knob on Sume, so a re-run gives you another take, and you pay for it.