G↗GENJUTSUVIDEO STUDIOOpen studio ↗

Motion transfer

AI motion transfer.
Your character, their moves.

Bring a character image into motion with a reference performance. Dance, gesture, and act using a video as your guide.

INDEPENDENT AI VIDEO STUDIOVIDEO + IMAGE → NEW VIDEOKLING · DREAMACTOR · ONE-TO-ALL
THE TRANSFORMATION ROOM● REC
TAKE
02
ACTUAL STUDIO OUTPUTMake it
your own.
GENERATED TAKE / KLING 3.0 MOTION CONTROL
↳ SAME IDEA. NEW DIRECTION.16:9
A DIFFERENT REALITY ↗

See what a new take looks like.

Real inputs. Real generated takes.

These are actual tests from our studio. Each result uses the source clip and reference shown below.

Rooftop run → neon city

MOTION TRANSFER / 3s
01 / PERFORMANCE VIDEO
Character reference used in this test
02 / CHARACTER REFERENCE
03 / GENERATED RESULT
See the prompt and model

The man in the tan overcoat runs with the same performance as the reference video. Keep his outfit, facial identity and neon city atmosphere.

Model: Kling 3.0 Motion Control · Actual output from a completed generation.

Rooftop run → night street, DreamActor

MOTION TRANSFER / 3s
01 / PERFORMANCE VIDEO
Character reference used in this test
02 / CHARACTER REFERENCE
03 / GENERATED RESULT
See the prompt and model

No description: this model follows the image and the video only.

Model: DreamActor v2 · Actual output from a completed generation.

DreamActor follows the image and the video only; no description was used. The character image was sent as a JPG copy of the same file.

Rooftop run → neon crossing, One-to-All

MOTION TRANSFER / 3s
01 / PERFORMANCE VIDEO
Character reference used in this test
02 / CHARACTER REFERENCE
03 / GENERATED RESULT
See the prompt and model

The man in the tan overcoat runs with the same performance as the reference video across a city rooftop at night.

Model: One-to-All 1.3B (480p) · Actual output from a completed generation.

How AI motion transfer works

AI motion transfer reads the movement in a short video you upload, frame by frame, and gives it to the character in an image you upload. The pose, the timing and the camera path come from your clip; the character, the outfit and usually the setting come from your image. A dancer’s routine, a sprint across a rooftop or a wave to the camera can be performed by a cartoon mascot, a game character or any original figure you have the rights to.

The people in your source clip never appear in the result. AI motion transfer is not a face swap: it animates your image rather than editing your footage.

Four AI motion transfer models

Each model has a different balance of cost and fidelity. The studio lists them with a credit price for five seconds.

Kling 2.6 Standard Motion Control is the default: reliable full-body motion for 105 credits per five seconds. Kling 3.0 Standard Motion Control is the tested quality option for 189 credits, when hands, fast turns or camera moves matter. DreamActor v2 from ByteDance also follows facial expression and head movement for 75 credits; it needs a JPG or PNG under 4.7 MB and ignores the text description. One-to-All 1.3B is the lowest-cost draft at 480p for 45 credits.

Scene presets for motion transfer

Eight presets set the scene around your character: keep my setting, neon city, soft studio, concert stage, beach sunset, snowy street, rooftop at night, fantasy forest. Choose one in the studio and edit the wording if you want a different place or light.

Getting a clean AI motion transfer

Match the framing of the clip and the image. A full-body performance needs a full-body character image; a waist-up gesture works best with a waist-up portrait. Keep one person in the clip, avoid hands hiding the face, and choose a continuous shot without cuts. Bright, even light and a plain background in the image help the model keep the character’s details.

Fast spins, crowds and tiny subjects can be rejected by the model or come out inconsistent. When a job fails, its credits return automatically.

Where AI motion transfer is useful

Creators use it to put a mascot into a trending dance, to preview choreography on a character before an animation is drawn, to animate game and comic characters, and to turn one performance into several variations. Add a restyle step afterwards to redraw the take as anime, claymation or a neon night.

Motion transfer compared with filming or animating

Filming a new performer means casting, a location and a shoot day. Keyframe animation means hours of work for each second. Motion capture needs suits or multi-camera rigs. AI motion transfer reuses a performance you already have: one clip and one image give you a new take in minutes, at a cost you see before you start. It is best for short, single-character moments; long scenes with several people or exact lip sync still need traditional tools.

It also works well as a planning tool. Directors and choreographers can see a routine on the intended character before a shoot, and small teams can test an idea on several characters in an afternoon instead of commissioning each version separately.

Common mistakes and how to fix them

Most failed or weak takes come from a mismatch between the clip and the image: a seated portrait for a full-body dance, a character facing sideways for a clip shot head-on, or an image with several people in it. Very dark footage, motion blur and a moving crowd behind the performer also confuse the model. Fix the input first, then try a higher-quality model if you still need more detail.

Their moves.
Your character.

Three steps. One performance.

01 /

Upload the performance

A continuous 3–10 second MP4 with one clearly visible person, recorded or licensed by you.

02 /

Add the character image

JPG, PNG or WebP up to 10 MB, framed like the performance: full body for full body, portrait for portrait.

03 /

Choose a model and generate

Pick Kling, DreamActor or One-to-All, check the quote and generate. The completed result appears automatically in My results.

AI motion transfer
questions.

About motion transfer

Which AI motion transfer model should I start with?

Kling 2.6 Standard at 105 credits per five seconds. Try One-to-All for a cheaper draft or Kling 3.0 for the final take.

Does the person in my video appear in the result?

No. Only the movement is used. The character, outfit and usually the setting come from your image.

Can I animate a photo of a real person?

Only with that person’s permission, and never to impersonate them or deceive anyone. The Acceptable Use Policy lists what is not allowed.

Why did DreamActor reject my image?

DreamActor needs a JPG or PNG under 4.7 MB. Choose another model, or export the image again in one of those formats.

How long can the clip be?

Between 3 and 10 seconds. Credits are charged per second of the clip, rounded up.

Your character.
Their moves.

Your next take starts here

Open the studio ↗