Kling Motion Control 2.6

It is the motion-transfer model from Kuaishou's Kling, also called Kling AI Motion Control: it makes the person in a photo perform the moves from a reference video. blubiu runs the 2.6 motion-control model and bills it in blubiu credits; this is not the Kling app, and you don't need a Kling account.

This page runs 2.6. According to Kling's user guide (checked 2026-11-04), 3.0 builds on 2.6 with a focus on faces: facial features stay stable and expressions smooth through complex, multi-angle and long motions, and you can bind face references. In Kling's own app, 3.0 also costs more: 12 credits per second in Professional mode against 8 for 2.6. For full-body dances and walks, where the face is small in the frame, 2.6 does the job, as the examples on this page show; close-up facial acting is where 3.0's changes matter.

Right at the top of this page, on the AI motion transfer page, or on the motion control AI page; all three use the same form: add the full-body photo and the motion clip, choose the orientation and 720P or 1080P, and generate. The motion transfer page is about the job; this page explains the model, its limits and its price.

No. On blubiu there is no prompt box: the photo decides who moves and what they look like, and the motion video decides how they move. Kling's own app takes an optional prompt for background details; the background in the result comes from your photo (see the rooftop and lobby examples), so to change it, put the person in that setting in the photo before you start.

You can try it free with the Free plan's monthly credits, which cover a short 720P test. It is not unlimited on any plan: every run is billed by the second, the Generate button shows the exact cost before you confirm, and failed runs are refunded automatically. Plans and credit packs are on the pricing page.

By the length of your motion video in whole seconds, at the rate for the resolution you pick, 720P or 1080P; the result normally runs the same length. Current rates are in the table above, the Generate button shows the total, and failed runs are refunded.

The motion video can be 3 to 30 seconds, and the result matches its length. Very fast or complex moves can come out shorter. For a longer routine, split it into parts and run each with the same photo.

Several minutes, and longer at 1080P. In our runs, a 5-second 720P clip took about 3 minutes (2026-10-08), 6- and 8-second 720P clips about 7 and 8 minutes, and an 8-second 1080P clip about 21 minutes (2026-11-04). You can leave the page: the result lands in your Creations, and a failed run refunds its credits.

No. In our tests on 2026-09-20, non-human characters (robots, animals, mascots) were rejected upstream with "no valid characters detected", and the credits were refunded. Use a full-body photo of a person; a human-shaped 3D cartoon character also worked in our 2026-11-04 test (see the examples). For a non-human character, Wan 3.0 can replace the person in a video instead; recast video puts your character into a dance, an ad or a trending clip that way.

Use motion control when the character is a person in a photo and the clip runs up to 30 seconds; only the motion video's seconds are billed. Use Wan 3.0 person replacement when the character is non-human, or when you want to keep the original video's background and camera; it takes source clips up to 15 seconds and bills both the source and the output.

Use a photo of yourself or of someone who has agreed, and a clip you have the rights to use. The result shows that person performing moves they never made, so get their permission before you post it.

Start at 720P to check the moves, then rerun the keeper at 1080P. For 2K or 4K, upscale the 1080P result with FlashVSR on the AI video upscaler; it keeps the motion, length and frame rate.