Skip to content
Train Your Own ModelLesson 5 of 12

5 · Hands-on: train a model in your browser

Time to actually do it. You're going to train a real image model — no coding, no app to install, and no account needed — using a free tool from Google called Teachable Machine. It runs right in your web browser at teachablemachine.withgoogle.com.

Good to know (privacy + cost): Teachable Machine is free, and Google says your images, sounds, and poses stay on your own computer while you train — they don't get uploaded unless you choose to save your project to Google Drive. Nothing to buy. (Do this with a parent or teacher nearby, and use your own safe, non-private examples — see Lesson 10.)

Teachable Machine can train three kinds of models: Image (recognize pictures or webcam frames), Audio (recognize short sounds), and Pose (recognize body positions from a camera). We'll do an Image project because it's the easiest to see working. A fun first goal: teach it to tell "thumbs up" vs. "thumbs down."

The steps (this is the whole game — features, labels, training data, all in action):

  1. Open the tool. Go to teachablemachine.withgoogle.com and choose Get Started → Image Project → Standard image model.
  2. Name your labels. You'll see two empty classes ("Class 1," "Class 2"). Rename them to your two labels — for example, Thumbs Up and Thumbs Down. (Labels, from Lesson 2!)
  3. Collect training data. Under Thumbs Up, click Webcam and hold the pose while it captures a bunch of frames. Aim for lots and varied — different distances, angles, and backgrounds (remember Lesson 3: variety matters). Do the same for Thumbs Down. Try to keep the two roughly balanced in count.
  4. Train. Click Train Model and wait a few seconds while it learns the pattern. Don't switch browser tabs during training — it pauses.
  5. Try it live. A Preview panel turns on your webcam and shows its guess plus a confidence bar for each label. Give it a thumbs up — does the right bar shoot up?

That's it — you trained a model! The bars you see are the model's confidence: how sure it is, shown as a percentage. In the next lesson we'll test it properly and learn what those numbers really mean.

Maker tip: If you don't have a webcam, you can upload image files instead. And you can add a third label (like "no hands") to make it smarter about when neither pose is showing.

Think about it. Before you train, predict: which two labels did you pick, and what kind of variety will you add to your examples so the model doesn't accidentally learn your background instead of your hands?

Sources

Sign in to track your progress