YT Life — Eight Pieces of Brocade, in three dimensions

Why a video isn't enough

A recording shows you one viewpoint. A form has all of them.

Every instructional video was shot from wherever the camera happened to stand. When the movement turns away from it, the detail you needed turns away too. This is the same performance, but the viewpoint is yours.

Walk around the form

Orbit, move in close on the hands, drop to floor level to read the stance. The motion keeps playing while you move.

Slow it to a quarter

Quarter, half, three-quarter speed — the transitions that blur past at full tempo are where the technique actually lives.

iii

Loop three seconds

Mark A, mark B, and the passage repeats until your body has it. Set it on the wrist rotation and leave it running.

Choose the body

Three avatars carry the same performance. Watch the form on a build closer to your own — proportions change how a movement reads.

Practice somewhere

A wooden practice hall by default, or step the form into a scanned basilica, gallery or garden.

Nothing to install

It runs in the browser on ordinary hardware. No plugin, no download, no dedicated graphics card.

The set

Eight pieces of brocade, each one captured whole.

A thousand-year-old health form, performed by one practitioner and recorded end to end — no cuts, no cleanup between repetitions. Hover any panel to watch it move.

Stand on the toes and bounce out the heels — Ⅷ
Stand on the toes · bounce the heels

How it's made

Five cameras in, one moving body out.

No suits, no markers, no gloves. The practitioner performs the form in an ordinary room; everything after that is solving.

Record the form
Five synchronized cameras, calibrated to each other, running the full set end to end at sixty frames a second.
Find the body in every view
A whole-body detector marks 133 points per camera — face, torso, feet and, critically, every finger joint.
Solve a real skeleton
Those views are fitted to a parametric human rig (Meta's openly licensed MHR): one identity, calibrated once from a calibration hold and frozen, so the skeleton never quietly changes size mid-form.
Repair what the cameras couldn't see
Arms disappear behind the back; hands cross the body. Occluded spans get a second opinion from a single-image 3D model, and the result is checked joint by joint before it's accepted.
Refuse to ship it until it passes
Automatic gates measure limb penetration, per-frame rotation jumps and fit residuals. A take that fails doesn't go in the studio.

One instant of the form, four viewpoints at once. The rig watches from five angles so the solve never has to guess at a limb the camera lost — and once it's solved, standing anywhere is free.

The details

Hands first. Everything else is easier.

A form is judged by whether it looks right, and what breaks that illusion is rarely the big movement — it is a finger passing through a thigh, or a wrist snapping between two frames. Those are the things the pipeline is built to catch.

The hands set the bar

Fingers are small, fast and constantly hidden behind each other — and they are the first place your eye goes when something is wrong. The solve weights them accordingly, and holds them steady across neighbouring frames rather than letting them flicker.

Occlusion is expected, not ignored

An arm sweeping behind the back is invisible to most of the rig. Those spans are identified and given a second opinion, then checked joint by joint before any of it is accepted.

iii

Nothing ships unchecked

Every rebuild re-measures limbs passing through the body, frame-to-frame snapping and how far the body sits from what the cameras saw. A take that fails is not quietly published — it goes back.

The channel atlas · 经络图谱

The traditional map, on the same body.

Acupuncture's channel-and-point map has lived on flat charts for centuries. Here, the WHO's standard point locations — all 361 of them, across the fourteen channels — are solved onto the same captured body that performs the form, each point carrying its source text and its classical needling depth. A flow animation traces the circulation sequence the classical literature describes, breath by breath.

Located, not illustrated

Every point is constructed from the written rules of the WHO Standard Acupuncture Point Locations (2008) — solved onto the body's surface, with the page reference a click away. Where a location rests on soft tissue the atlas says so, plainly.

Depth you can see

Switch to x-ray and the points sink to their classical depths — the Lingshu's own per-channel figures, drawn to scale inside the body rather than printed as a footnote.

The circulation, traced

The flow animation follows the classical sequence — lung to liver and around again — as one continuous cycle. An illustration of the traditional model, presented as exactly that.

Explore the channel atlas → Watch it move →

The road ahead

Next: your body in the studio, and a second opinion on your form.

What exists today is one practitioner's set, performed by borrowed bodies. The work underway points it at you.

A captured person, rigged and ready to move.

This part already works — on a capture stage

That is the studio's own creator, reconstructed from a capture session into geometry that carries his build, his clothing and his face, then rigged so any of the practitioner's takes can drive it.

Getting here currently takes a rig and a room. The work below is about getting here from a phone.

In development

Your own body, from ordinary video

The bodies in the studio began as scans. The same reconstruction that turns a capture into a mesh can be pointed at footage you already have — a few passes around yourself — and give the studio a body with your proportions instead of a stand-in's.

In development

The form, performed on you

Motion and body are deliberately kept apart, so a take can be replayed on any build. Carrying it onto your mesh is the same problem one step harder: a reach has to clear your frame, at your limb lengths, without borrowing someone else's proportions to do it.

Research

Guidance from your own attempt

Record yourself doing a movement and have it solved by the same pipeline that solved the reference. The comparison is where instruction begins: which joint arrived late, where the weight shifted early, and which of those differences actually matter.

In the studio now

The traditional map, drawn accurately

The WHO-standard channel-and-point atlas is now explorable on the captured body — see it above. Next: the same points riding the moving form itself, so the map travels with the practice.

Calibrated confidenceconformal prediction: Telling someone their elbow is wrong is easy; being right about it is not. Conformal methods put a stated error rate on that judgement, so a flag means the difference sits outside what measurement noise can explain — not that it crossed a threshold somebody guessed.
What leads whatGranger causality: Good movement is sequencing: the turn starts in the hips, or it doesn't. Testing whether one segment's motion helps predict another's tells you which one leads — the difference between "your arm is late" and the correction a teacher would actually give.

Your practice

The motion is the practitioner's work. It stays that way.

A teacher's form is the thing they spent decades building. It is treated here as theirs, not as a file to be passed around.

Streamed, never downloaded. Movement data is fed to your session as it plays and lives only in memory. There is no motion file to find in a cache.

Signed in, always. Every avatar and every take is delivered against your session and marked never-to-be-stored.

Credited where it's due. Scanned practice environments carry their creators' names and licences in the studio's credits panel.

Watch the form
from any angle,
as slowly as you need.