12-second clip: face swap $2.21 vs Kling motion control $1.89

For a 12-second source video, Sume's Avatar Face Swap beta costs $2.208 at standard and Kling 3.0 Motion Control costs $1.89 (as of 2026-10-09). The two are not the same product, but they are related: the face-swap worker uses a Kling 3.0 motion-control stage after it swaps the first frame.

What each one takes

Face swap takes an avatar handle, a public source video and a quality tier. It returns the source video with the avatar's face on it, and the source audio comes back in the final file. Motion control takes your own still image (or avatar) and a motion reference video, and returns the still performing the reference's movement.

Why the cheaper one is not always the answer

The $0.318 gap (2.208 - 1.89) buys the first step: swapping the face in the first frame so that the still you animate already has the avatar's face, and putting the source audio back afterwards. If you already have a clean still of the character, call motion control directly and save it. If you start from a video and want the face replaced throughout, face swap does the extra work.

There is a second difference in range. Face swap is a beta with a window of about 4 to 15 seconds. Motion control accepts 1 to 30 seconds, so a 25-second dance reference only works on the motion route: 25 x 0.1575 = $3.9375.

Worked example: ten clips

Suppose you have ten 12-second source clips. Face swap at standard is 10 x 2.208 = $22.08, plus the one-time $0.95 for the avatar, for $23.03. Calling motion control ten times with a ready still is 10 x 1.89 = $18.90. The difference of $4.13 is small next to the time it takes to prepare ten clean stills, so count your own preparation before you decide the cheaper route is cheaper.

Both routes are asynchronous. Submit with an Idempotency-Key, poll the status URL, and read the result when it is ready. A failed job is refunded, so a clip that fails validation, for example a source with no usable audio on face swap, costs you nothing but the attempt. Test one clip of each kind before you queue the batch.

Choosing in practice

Use face swap when the input is a recorded clip and the goal is a different face on the same performance. Use motion control when the input is a still and a movement reference, and the goal is a new performance from that still. Both are paid jobs with the same lifecycle, covered in Jobs and results.

If you only need the avatar's face on a short loop, motion control with a still is the simpler path.