Motion transfer copies someone else's moves onto your photo

Motion transfer copies someone else's moves onto your photo. Your photo. Someone else's moves. generate_video_from_image in Kling Motion Control mode takes a character still as image_url and a reference clip as video_url. It does not invent a dance from a sentence. Asking it to "make me dance" with only a portrait wastes credits on ordinary image-to-video motion you will not keep.

Transfer a dance or motion onto my photo is the named capability. The AI video generator can animate a still without a driver. That is a different contract. AI motion transfer is the tool-shaped door for this one.

Two inputs, or it is not this job

The capability notes are not optional: you need the photo and a separate motion-reference video. The agent maps the still to the character and the clip to the motion source. Do not bury the driving URL in the prompt text. If you only have a still, you wanted turn a photo into a video: one image, invented camera and body motion.

If you only have a dance clip and no character still, you wanted a different generate, not a transfer. There is nothing to transfer onto.

Output is a video of your photo's subject performing the reference clip's motion. Premium-tier motion-control models, priced per model and shown before you confirm. That price is for a copy, not for a vibe.

What wastes credits

Prompted choreography. "A woman does the exact dance from that trending sound" is text-to-video guessing joint positions. Generate a video from text will give you a dance. It will not give you that dance. Motion control exists because language cannot specify 240 frames of timing.

Image-to-video with extra adjectives. "Animate this photo, make it look like she's dancing" is a pan and a smile, not a copied routine. Same primary tool name, different mode, different required fields. Missing video_url is how you pay motion-control prices in your head and receive an I2V take in the file.

First-and-last-frame interpolation. Two stills with no driving clip is a first and last frame transition. That interpolates plates. It does not retarget a performance.

The test

Count the attachments.

Two files — a character still plus a motion clip you would actually copy — and this is the job. One still, and you are in image-to-video. One clip and a text description of a person, and you are inventing a performer, not transferring motion.

If the movement is the content — a dance trend, a product-demo gesture, a specific hand action — copy it. If the movement is "slow push-in, soft smile," do not buy a motion-control row. This is one named agent job. Using it for a different job wastes credits.

FAQ

Can I describe the dance instead of attaching a reference clip?

Not for this job. The driver is the motion. A paragraph about choreography is a different generate, and a worse one when timing is the point.

Is this the same as animating a product still?

No. A product orbit or a blink from one photo is turn a photo into a video. Motion transfer is for when an existing performance must land on your subject.

Why is this a premium-tier model?

Because it is retargeting motion, not sampling a clip from a prompt. The cost is shown before you confirm. Treat that quote as the price of a copy. If you did not bring a copy-worthy driver, cancel.

Can I use any video as the driver?

Use a clip whose motion you actually want copied, with a readable body. A messy party video with five people and a whip pan will transfer mess. The still is the character contract; the clip is the motion contract. Both have to be good.