AI Motion Transfer
Use subject photos to replace a video’s main subject and face, remove face-blocking props, and retain source action, scene and timing. Use a 4–30 second clip and choose 1080p or 720p.
What is AI motion transfer?
This play edits the reference video to put your photo subject into the original action, scene and camera movement. The video supplies timing and motion; the subject photos supply identity and appearance. Output follows the source duration and aspect ratio, with a possible duration difference of up to about 0.4 seconds. Choose 1080p or 720p. Facial details still require review.
Replacing the subject includes replacing the face. The subject photos define facial identity, body and clothing; the source performer is not an identity reference.
The instructions also ask to remove props that block the target face, such as a microphone, while preserving hands, body movement and the surrounding scene. Hidden facial details cannot be guaranteed from an obscured photo.
Creation examples
How to use AI motion transfer
Upload a reference video
Upload one 4–30 second MP4 with a clearly visible main subject. Start with a clip within 20 seconds and avoid rapid cuts or heavy occlusion.
Upload subject photos
Add 1–5 subject photos to define identity, body and clothing. For a human subject, choose photos with a clear, visible face. When several photos show the same person, keep their identity consistent; explain separate subjects using @. Choose an example before uploading your own photos.
Add details and generate
Choose 1080p for higher output detail or 720p for a lower-cost trial. Duration and framing follow the source video. Add optional subject instructions, review the displayed quote, then generate and inspect the face, action and complete timeline.
Ideas for motion transfer
Creative character clips
Explore how a character photo might move with a reference performance for a short creative concept.
Pet motion experiments
Try a clear pet photo with a simple reference movement to explore playful video ideas. Results will vary.
Social video concepts
Prototype a movement-led short video using footage and subject images you are authorized to use. Review the result before sharing.
What controls does AI motion transfer give you?
Video editing preserves the source timeline and framing while the subject photos guide replacement identity. Select output resolution separately; a higher resolution does not guarantee facial similarity.
Use video to guide the action
The reference video supplies the action sequence, speed, camera and scene. Editing follows the original duration, allowing a small difference of up to about 0.4 seconds.
Define the subject with photos
The subject photos define the replacement’s facial identity, hairstyle, body and clothing. Use consistent photos for each intended subject and assign their roles clearly when several people are involved. The original performer’s face is not the identity reference.
Refer to materials with @
In Additional instructions, type @ and select an uploaded material to say which video or photo you mean. Describe the intended action or appearance instead of manually guessing material numbers.
How can you prepare better motion transfer inputs?
Align your materials with one clear creative goal, then refine the result. These tips help you select inputs and review output; results still depend on the model, materials, and instructions.
Choose readable reference movement
Start with a well-lit clip, one clear subject, and few obstructions. Walking, waving, or a simple dance is easier to assess than a sequence with fast cuts, overlapping people, or heavy camera shake.
Match the photo to the framing
Choose subject photos with visible eyes, nose, mouth and jaw. The instructions ask to remove face-blocking props, including microphones, without removing hands or changing the action. An obscured photo does not reveal the hidden facial details; use clearer subject photos when available and inspect the result through turns and occlusion.
State the main request clearly
Use @ to select the photos and state who should replace the main performer. Keep the original motion and camera unless you want them changed. If several images show one person, say so; avoid conflicting identity instructions.
Review the result and change one variable
Check the face, hands, feet, occlusions, background, and motion transitions. If needed, replace the reference clip or simplify the instructions, changing one factor at a time so you can compare results. Fine details may not be preserved in every generation.
Motion transfer vs image-to-video and text-to-video
Choose motion transfer when a reference video already shows the action you want. Image-to-video starts from a still image, while text-to-video starts from a written scene. The comparison below explains these workflow concepts; available models and controls may differ.
Workflow
Select an option to explore the possibilities
Motion transfer
- Inputs and motion guidance
- Reference video + subject photos; this play edits the original clip, preserving its timing, framing and action.
- When to choose it
- Keep an existing performance and scene while replacing the main subject with your photo subject.
Image-to-video
- Inputs and motion guidance
- Still image + written instructions; the model creates movement from the image and prompt.
- When to choose it
- You have a visual starting point but do not need to follow a specific reference clip.
Text-to-video
- Inputs and motion guidance
- Written scene and motion description; a source image or reference clip may not be required.
- When to choose it
- You want to invent a scene or action from a written idea.
Face replacement
- Inputs and motion guidance
- Existing image or video + subject photos; the focus is changing facial identity.
- When to choose it
- Your main goal is to change a face rather than transfer whole-body movement.
How does motion transfer fit into a video workflow?
Use this play when you want to retain a recorded performance, timeline and scene while changing the main subject. If you want a completely new scene based on a photo, use image-to-video instead. Assemble completed clips in an editor for captions, soundtrack and final delivery; verify identity and motion before sharing.
Ready to try your own motion reference?
Prepare a 4–30 second source video and 1–5 subject photos. For a human subject, choose photos with clear facial features. Choose 1080p or 720p and review the displayed cost before generating. Check that the result shows the intended face rather than the original performer.
Frequently asked questions
What do I need for AI motion transfer?
Use one 4–30 second MP4 and 1–5 subject photos. The subject photos define facial identity, body and clothing; the video provides action, scene and timing. For a human subject, choose photos with clear, visible facial features.
Which file formats and sizes are supported?
The current setup accepts an MP4 reference video up to 100 MiB and JPG, PNG, or WebP subject photos up to 10 MiB each. The model may impose lower limits; follow the upload instructions.
Can I use a person, pet, or fictional character?
You can try a clear photo of a person, pet, or character. Motion and appearance will vary with the subject and reference video. Use materials you have permission to use and obtain any necessary consent for a real person’s likeness.
How can I improve the result?
Use clear subject photos of the target person and identify the performer to replace. The instructions ask to remove face-blocking props, such as microphones, while preserving hands, action and the scene. Hidden facial details cannot be guaranteed from an obscured photo. 1080p increases output dimensions but does not itself solve identity errors.
Will the result preserve the subject and motion exactly?
The prompt requires using the subject photos to replace facial identity, head, body and clothing and forbids retaining or blending the source performer’s face. This guides a generative model; it does not guarantee a face lock. Inspect turns, occlusion and reappearance before using the result.
How much does generation cost?
The cost can vary with the model, materials, and settings. Check the quote displayed before submission, and review it again if you change the settings.
Can I use someone else’s video or photo?
Upload only materials you have the right or permission to use, and obtain any required consent for likeness, performance, and other rights. Check applicable rights and rules before public or commercial use.
How do I use @ to reference a video or photo?
Upload your materials first. In Additional instructions, type @, select the intended material from the list, and describe how to use it. Select the material rather than manually typing Image 1 or Video 1; references are assembled from the submitted materials.
What video duration, aspect ratio, and resolution can I use?
Input video: 4–30 seconds; start within 20 seconds for easier review. Output follows source duration and aspect ratio, with a possible difference of up to about 0.4 seconds. Default resolution is 1080p; 720p is available for a lower-cost trial. Resolution and duration affect the quote. 1080p output may use HEVC, so playback requires a compatible device.