AI Motion Transfer
Upload a reference video and a photo to explore how a person, pet, or character can follow the video's movement. Add optional instructions and use @ to refer to your materials.
What is AI motion transfer?
AI motion transfer uses movement from a reference video to guide a new clip featuring the subject in your photos. In TokAtlas, upload one reference video and 1–5 subject photos, then optionally describe the intended action and appearance. The model attempts to follow the movement while retaining the subject’s visual identity; poses, faces, hands, clothing, and backgrounds may differ from your inputs.
Creation examples
How to use AI motion transfer
Upload a reference video
Choose one MP4 video with clear motion and minimal obstruction. Follow the upload instructions for file and model limits.
Upload subject photos
Add 1 to 5 photos of a person, pet, or character. Choose clear images with a visible outline; supported formats are shown on the page.
Add details and generate
Optionally describe the result you want, review the displayed settings and quote, and submit. Check the finished motion and appearance; adjust your materials or description and try again if needed.
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?
TokAtlas combines a motion reference, subject photos, and optional instructions in one workflow. Each input has a different role; the generated clip may vary from the source footage.
Use video to guide the action
Provide one reference video to show the movement, timing, and pose transitions you want. The video supplies motion guidance; it is not a frame-by-frame editing instruction.
Define the subject with photos
Upload 1–5 clear photos to show the intended subject. When using several photos, keep the subject consistent and explain each photo’s role in your instructions.
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
For full-body movement, try a photo that shows the whole subject; for expressions, use a clear face photo. People, pets, and illustrated characters have different proportions, so review cross-subject motion carefully.
State the main request clearly
Use @ to select the video and photo, then describe your goal. For example: “Follow the waving rhythm of the selected video, use the selected photo for the subject’s appearance, and keep a fixed camera.” Keep motion, identity, and camera instructions consistent.
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 photo(s); the video guides the intended action.
- When to choose it
- You already have a movement example and want to explore it with a different 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 + face reference; 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?
A motion transfer clip can serve as a character-performance shot in a longer video. Define the subject and action first, then plan the surrounding scenes. Consider text-to-video for a new scene described in words, or image-to-video when extending a still image. Assemble the clips in a video editor to adjust the sequence, captions, and soundtrack. Each workflow has different input requirements; check the features available in your chosen tool.
Ready to try your own motion reference?
Prepare one reference video and 1–5 subject photos. Add optional instructions, use @ to select the materials you mean, and review the available settings and displayed cost before submitting. Start with one clear action so you can judge the result.
Frequently asked questions
What do I need for AI motion transfer?
You need one reference video and 1 to 5 subject photos. The video provides the motion; the photos provide the person, pet, or character to appear in the result. You can also add an optional description.
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?
Choose a well-lit reference video with clear movement and minimal obstruction, and a photo with a clear subject and outline. Complex movement, fast camera motion, or heavy obstruction can make generation harder. Try different materials if needed.
Will the result preserve the subject and motion exactly?
Not necessarily. The model attempts to follow the reference motion and retain the subject’s appearance, but identity, clothing, pose, details, or background can change. Review the result before using or sharing it.
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?
Available output settings depend on the model and options shown in the generation panel. Check them before submitting. An upload file-size limit does not guarantee a particular clip duration or output resolution, and the output may differ from the reference video’s length or framing.