Today you will train an AI pose model, using pictures of your own body, to tell apart three positions. These are tilting left, tilting right and not tilting.

Each position is a category, which means a group the model learns to recognise. You teach it by recording pictures of yourself holding each pose. The diagram shows that the tool looks for patterns in your pictures, and it becomes most sure about the pose with the most pictures.

Before you record anything, think about what might make the model confident and what might confuse it.

Google's Teachable Machine can learn poses as well as hand gestures. Today it will learn from pictures of you tilting left, tilting right or not tilting, and work out which one you are doing. You do not need an account or any code.

The picture shows Teachable Machine's own example, a model that tells whether someone is standing up or crouching down low. Your model will learn the three tilts instead.

Open Google's Teachable Machine, the website where you will make your model. When it opens, click the Get Started button.

Click the Pose Project button to start a new pose project. This takes you to the screen where you set up the classes for your model.

A class is a category that the model learns to recognise. Your model will have three classes, which are Tilt Left, Tilt Right and No Tilt. The AI learns what pictures of each class look like, so that it can tell them apart.

Your model already has two classes, called 'Class 1' and 'Class 2'. Click the 'Add a class' button to add a third class. Then rename your three classes to these names:

• Tilt Left
• Tilt Right
• No Tilt