Samsung Patents an AI System That Puts Your Body Double in Workout Videos
Instead of watching a stranger do squats and hoping you can follow along, Samsung's latest patent describes a system that generates a workout video starring a digital model of your own body, built from your personal measurements and fitness data.
What Samsung's personalized AI workout video actually does
A generic workout video shows a fit stranger demonstrating moves that may not match your body, your pace, or your current fitness level. You have to mentally translate what you see into what you should do, which is harder than it sounds.
Samsung's patent describes a device that collects your body measurements and your exercise history, then builds a 3D digital model of your body. It picks a workout program that fits your level, finds an appropriate exercise video, and uses generative AI to re-render that video so the person performing the exercises looks like your digital double, moving through the program at the right pace for you.
The result is a guide video personalized to your shape and your routine, not a one-size-fits-all clip filmed for an imaginary average person. Think of it as a workout video where the trainer in the frame has been replaced by a version of you.
acquire user information, the user information comprising body information and exercise performance information, generate a user body model based on the body information, identify an exercise program based on the exercise performance information …
Translation: The system collects your physical stats and fitness history to build a digital avatar and pick a workout.
How Samsung builds your body model and rewrites the video
The system starts by collecting two kinds of data: body information (likely measurements, proportions, or scan data used to construct a 3D avatar) and exercise performance information (your fitness history, ability level, or workout preferences).
From the body data, the device generates a user body model, a digital representation of your physical form. From the performance data, it selects an appropriate exercise program and locates one or more real exercise videos that correspond to that program.
The core step is then handled by a generative AI model. That model takes the real exercise video and your body model as inputs and produces a new guide video in which the movements from the original footage are performed by your digital avatar. The human demonstrator in the source video is, in effect, replaced by a model built to your specifications.
- Acquires body and fitness data from the user
- Generates a 3D body model from that data
- Selects a matching exercise program and source video
- Uses generative AI to re-render the video with the user's avatar performing the exercises
- Delivers the finished guide video as part of an exercise guide
… acquire, by using a generative artificial intelligence model, a guide video from the identified at least one exercise video and the user body model, the guide video including exercise images of the user body model according to the exercise program …
Translation: AI combines a workout video with your avatar to show your digital self doing the exercises.
What this means for fitness apps and AI-generated content
For fitness apps and smart devices like Samsung's Galaxy Watch or future health-focused hardware, this kind of personalization could make workout guidance feel less like watching someone else exercise and more like previewing your own movement. Seeing a body model that mirrors your proportions could also help with understanding form, since the visual reference actually matches your build.
Samsung's steady investment in health and fitness patents points to a broader push to differentiate its devices through on-device AI. If the generative rendering runs locally on the device rather than in the cloud, it also raises interesting questions about how much personal body data is stored and where. For now, the patent says little about that, which is worth keeping in mind.
That makes this Samsung's 30th filing we've tracked in AI image and video since May, a topic that already covers compressing AI without real data and rating photo quality in words.
Claim 1 as written is broad. It covers any electronic device that takes body data, builds a body model, finds an exercise video, and uses any generative AI model to produce a guide video with that body model in it. There is no requirement that the AI be a specific architecture, that the body model meet a minimum quality threshold, or that the output video be photorealistic. That scope could cover everything from a basic avatar animation system to a high-fidelity video diffusion approach.
That breadth is a double-edged thing. A wide claim is harder to work around, but it is also more likely to face prior-art challenges during examination, since avatar-based fitness guidance and AI video generation have both been active research areas for years. The claim does not appear to require any novel combination beyond the basic pipeline.
What the claim does NOT cover is telling. It says nothing about real-time feedback, motion correction, or comparing your actual movements to the generated guide. The patent is purely about generating a personalized reference video, not coaching you as you move. That narrower scope makes the filing more defensible, but it also means competitors could build interactive coaching on top of a similar pipeline without touching this claim.
There are more where this came from
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
The drawings
15 drawing sheets from US 2026/0263897 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →