Motion Transfer
What is Motion Transfer?
Motion Transfer extracts movement from one video ( how a person walks, dances, or moves ) and applies that exact movement pattern to a different AI-generated character or subject.
At a glance
- Also known as
- Pose transferPose-guided video generationMotion-conditioned generationReference motion animation
- Used for
- Animating AI-generated characters using real-world movement referencesApplying dance, athletic, or performance motion to synthetic subjectsCreating character animation without manual keyframing or live motion captureTransferring actor performance to different visual characters
- Common tools
- ControlNet (pose conditioning)AnimateDiffRunway gen-3Kling motion reference featuresMorphic motion transfer tools
- Related terms
- ControlNetPose estimationAnimateDiffImage-to-videoKeyframeCharacter consistency
- How it works in simple terms
- The AI analyses a reference video to extract the skeleton and pose sequence of the moving subject frame by frame. It then uses those poses as a guide to generate a new video in which a different character ( defined by appearance reference or description ) performs the same sequence of movements.
- Where you encounter this
- Motion transfer features appear in AI video platforms including Runway, Kling, and Morphic's video generation tools, as well as in open-source workflows built on ControlNet and AnimateDiff. It is increasingly a standard feature in professional AI video production pipelines.
Ready to create?
Direct scenes, design characters, and ship full films
All-in-one AI creative platform with simple, transparent pricing, no speed throttles, and an infinite Canvas for max creativity.
How it compares
Compared with related concepts
Motion transfer is related to but distinct from traditional motion capture. Motion capture uses physical sensor suits, marker arrays, or optical tracking to record actual performer movement data for subsequent application to digital characters in a controlled production environment. Motion transfer uses any existing video footage as input, applying computer vision to extract movement without physical sensors, and generates a new video output rather than producing riggable animation data. Motion transfer is accessible with consumer hardware and standard video inputs; traditional motion capture requires specialised equipment, studios, and production infrastructure.
Think of it like…
Motion transfer is like placing a transparent sheet of tracing paper over a dancer's movement, capturing the pattern of every pose and transition, then lifting that pattern off and laying it over a completely different figure: so the new figure moves with the same timing, rhythm, and physicality as the original, but looks entirely different.
Pro tip
When using motion transfer, the quality of the source reference video significantly affects the quality of the transferred motion. Choose reference footage with clear, unobstructed body visibility, consistent lighting, a single subject occupying the majority of the frame, and relatively stable camera movement. Complex multi-person scenes, heavy occlusion (limbs crossing over each other), fast or chaotic motion, and unstable handheld camera movement all reduce the accuracy of pose extraction and degrade transfer quality. Clear, well-lit, single-subject reference footage produces the most reliable and controllable motion transfer results.
Types and variations
- Skeleton-based motion transfer uses extracted pose skeletons as the conditioning signal, transferring body structure and movement while allowing full visual re-styling of the subject.
- Dense flow-based transfer uses pixel-level motion vectors for finer motion detail but is more computationally demanding.
- Video-to-video motion transfer applies the motion of a source video to a new video generation, allowing both the motion and the visual context to change simultaneously.
- Face motion transfer specifically transfers facial motion and expression from a source to a target face, enabling talking head generation and facial animation.
- Object motion transfer, still an emerging capability, extends motion transfer beyond humanoid subjects to objects and non-character elements.
Ready to make your first scene in Morphic?
Try MorphicCommon use cases
Motion transfer is used in AI video production to animate characters with reference performance footage, in content creation for applying viral dance trends to branded or stylised characters, in education and training for visualising athletic technique or movement instruction with custom characters, in visual development for previewing how choreography or physical performance will look on specific characters before production, in film and advertising for creating character animation without the cost and logistics of traditional motion capture, and in any AI generation context where specific movement quality must be replicated in a generated character.
Ready to create?
Direct scenes, design characters, and ship full films
All-in-one AI creative platform with simple, transparent pricing, no speed throttles, and an infinite Canvas for max creativity.
FAQs
Motion transfer is an AI technique that extracts movement patterns from a source video: the timing, direction, and body mechanics of how a subject moves: and applies those patterns to a different generated subject or character. It enables creators to animate AI-generated characters using any existing video as a movement reference, without manual keyframing or traditional motion capture equipment.
Most motion transfer approaches use pose estimation to identify and track anatomical key points (joints, limbs, body contour) across the frames of a source video, producing a pose skeleton sequence. A generation model: often a diffusion model conditioned by ControlNet or a similar framework: then uses that pose sequence as a guide to synthesise a new video in which a specified target character performs the same movement sequence.
Large-scale, clearly visible whole-body movements: walking, running, dancing with broad gestures, athletic movements with clear body trajectory: transfer most reliably with current AI systems. Complex fine motor movements, rapid multi-directional motion, detailed hand and finger articulation, facial micro-expressions, and interactions involving physical contact between multiple people remain more challenging and produce less consistent results.
Yes, any video can in principle serve as a motion reference, but the quality of the transfer depends significantly on the reference video's characteristics. Clear, well-lit footage with an unobstructed, single subject, minimal camera movement, and clearly visible body positions produces the most accurate and usable pose extraction. Footage with heavy occlusion, multiple subjects, low resolution, or chaotic camera movement will produce lower quality motion extraction and less reliable transfer.
Motion transfer and deepfake technology share some technical foundations: both involve applying characteristics extracted from one video to another: but they have different objectives and concerns. Motion transfer focuses on transferring movement patterns to generated or synthetic characters for creative production purposes. Deepfake technology typically involves replacing the face or identity of a real person in a video with another real person's face, raising significant concerns about consent, authenticity, and misinformation. Ethical motion transfer applications use generated or consenting-subject targets.
Traditional motion capture requires physical sensor suits, specialised studios, and controlled production environments to record performer movement data. Motion transfer uses any existing video footage as input and applies computer vision to extract movement without physical sensors. Motion capture produces riggable animation data usable across 3D production pipelines; motion transfer produces new generated video output directly. Motion transfer is significantly more accessible and requires far less production infrastructure.
Motion transfer and character consistency are related challenges in AI video production. Motion transfer addresses whether the generated character moves the way the source reference moves. Character consistency addresses whether the generated character's appearance ( face, costume, body proportions ) remains stable across the frames of the generated video. Achieving both simultaneously: consistent character appearance performing transferred motion accurately: is one of the primary technical challenges in current AI video generation.
Practical applications include applying dance choreography to branded AI characters for marketing content, visualising athletic coaching material with custom instructional characters, animating game or film characters for pre-visualisation using actor reference performance, creating stylised social media content by applying trending movement to original characters, and producing training or educational material that requires specific movement sequences performed by consistent, reproducible synthetic subjects.
