Do Anything

Shape is an AI lab building a world model for human movement. Our first product is an app that watches you practise through your phone’s camera and coaches your technique.

Multiple exposures of a dancer tracing the sweep of her arms and the rotation of her body
Concept visualization

A world model
for movement

The model learns how people move from video.

We’re building it for our own products and will offer it to other companies through an API.

A sculptural human figure with arms outstretched, expanding into the four-arm, four-leg Vitruvian pose

01Measure

Measure a movement

Move through a recorded movement and watch the angles change.

02The app

Coaching from your phone’s camera

Prop your phone where it can see you and practise. Shape recognises the activity and coaches you in two modes.

Live

Visual and spoken feedback while you practise.

Review

Pause any session to see what to fix, drawn on the replay and compared with professional demonstrations.

Both plates show the same recorded piano passage with the player’s fingers tracked. Live measures the middle finger’s angle as the passage plays; it bends from about 155 to 137 degrees on the held chord. Review pauses at 6.88 seconds, at 137 degrees, and shows the angle over the whole passage.

03The lab

How the model works

It is designed to estimate the state of a body, predict what comes next and compare each attempt with its goal. Coaching states only what the evidence supports.

The app launches on established models. From July to September 2027 it runs entirely on Shape’s own.

Inputs
Video and audio from the phone
Architecture
Perception encoders, a world-state representation and a temporal joint-embedding predictive model
Output
A corrected motion state, shown in 3D, in AR or as a ghost, and explained in words
Inputs
Video and audio from the phone
Architecture
Perception encoders, a world-state representation and a temporal joint-embedding predictive model
Output
A corrected motion state, shown in 3D, in AR or as a ghost, and explained in words

04Plan

October 2026 to September 2027

  1. 01Launch

    October 2026 to January 2027

    1 December 2026iOS and Android release

    App
    Runs on established models
    Models
    Research on Shape’s own models begins
  2. 02Expansion

    January to June 2027

    App
    Product research, including VR headsets and smart glasses
    Models
    Proprietary model development
  3. 03Own models

    July to September 2027

    App
    Fully native Shape
    Models
    All inference on Shape’s own models

Join the waitlist