Motion matching
AI generates new video matching the motion of a reference video.
Motion matching is a technique that lets AI create new video content that looks and moves like a different video. Imagine you have a video of a dancer. Motion matching allows an AI to generate a new video of that dancer performing new actions, but with the same style of movement and camera work as the original video.
For example, if you have a video of a car driving around a track, motion matching could help an AI generate a new video of that car driving on a different track, but it would copy the car's speed, turns, and the camera's panning from the first video.
Motion matching is a video generation technique where the temporal dynamics, including camera motion and subject movement, of a source video are used to guide the generation of new video content. This is achieved by extracting motion features from a reference clip and conditioning the generative model, such as a diffusion model or a transformer, to synthesize novel sequences that adhere to these extracted motion patterns. The goal is to ensure temporal coherence and realistic movement in generated videos, often by aligning latent representations or using explicit motion embeddings.
This approach allows for greater control over the animation and cinematography of AI-generated video, enabling the creation of content that matches specific stylistic or kinetic requirements. Tradeoffs can include increased computational cost for motion analysis and conditioning, and potential artifacts if the reference motion is too complex or dissimilar to the target content.
This term is appearing now because frontier AI labs are pushing the boundaries of video generation, aiming for more control and realism. Techniques like motion matching are crucial for blending AI-generated elements with live-action footage seamlessly, as demonstrated by Google's recent work.
- Behind “ANCESTRA”: combining Veo with live-action filmmakingGoogle · Google DeepMind Blog ·
- Google Gemini
- Google DeepMind's work on ANCESTRA uses motion matching to combine its Veo model with live-action filmmaking, blending AI video with real footage.
- Motion transfer
- Motion transfer typically applies the motion of one object to a
- Style transfer
- Style transfer focuses on applying the visual aesthetic (colors,
- Video prediction
- Video prediction aims to forecast future frames based on past
The concept of motion matching in AI video generation has evolved from research in computer vision and graphics, drawing inspiration from motion transfer and style transfer techniques. Specific implementations and terminology are often defined within research papers and technical blogs from leading AI labs.
Try using a text-to-video tool that allows for reference images or video clips. Upload a short video of a person walking and then use a text prompt like "a person running" to see if the generated video captures some of the original walking motion style. Experiment with different reference clips to observe how the AI adapts.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- PromptRole-play a student struggling with math for AI tutor simulationSimulate a student's math struggles and uncertainty for AI tutor evaluation. · 5 min
- TemplateReusable workflow templates for healthcare teamsTeams can create and share templates for common workflows like documentation or prior authorizations. · 15 min
- TemplateTemplate for building an AI tool tailored to your workA structured approach to defining and building a custom AI tool for specific business needs. · 15 min
- TemplateEvals CSV prompt list for agent skillsA CSV file template for listing prompts used in evaluating agent skills. · 20 min
- HeuristicDouble coding agent task completion with docs linksProviding coding agents with a documentation link significantly increases their task completion rate. · 2 min
Questions people ask
- How does motion matching make AI video look more realistic?
- It helps by ensuring that the movements in the AI-generated video, like how a character walks or how the camera pans, match the natural flow and dynamics of real-world motion captured in a reference video.
- Can motion matching be used for any type of video?
- It's most effective for videos where specific, discernible motion patterns are important, such as character animation, sports, or dynamic camera work. Highly abstract or chaotic motion might be harder to match precisely.
First seen on Subvolts Oct 25, 2025. Updated Sep 15, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.