Teaching machine learning at school: an honest comparison of the tools

Machine learning has moved from a curiosity to something students use every day. The best way to understand it is to train a model yourself: collect examples, watch the model learn, find where it fails and fix it. Several tools make that possible in a classroom. They differ in the devices they need, where student data goes and what students take away at the end.

What to look for in a classroom tool

• The devices you already have. Buying new hardware for one unit is rarely an option.
• Where student data goes. Photos of faces, voices and handwriting are personal data. The less leaves the device, the easier the conversation with families and your data protection officer. And when a tool is free, it is fair to ask how it is paid for.
• Accounts. Every account is something to create, remember, reset and delete.
• Works without internet. School networks block, slow down or go down at the worst moment.
• Variety. Images are a great start, but data, text and sound show students that machine learning is everywhere.
• A real outcome. A model students can use inside their own app turns an exercise into a project.
• Ready-made lessons. Teachers need materials, not just software.

The tools, side by side

	MagicaLCore EDU	Teachable Machine	Machine Learning for Kids
Device	iPad with M1 or later	Computer with a browser	Computer with a browser
Model types	Image, sound, tabular data, text, hand pose, hand action	Image, sound, body pose	Text, images, numbers, sounds
Where training happens	On the iPad	In the browser	Depends on the project: text projects are trained in the cloud
Student accounts	None	Not needed to train	Class accounts, or a quick try without one
Works offline	Yes	Opened from its website	Needs a connection
Students build with it	An app in Swift Playgrounds	Web projects	Scratch projects
Teaching materials	Teacher’s guide with six solved exercises and rubrics, student notebook, trial guide	Lessons made by other educators	Worksheets and project guides

Teachable Machine

A wonderful first contact. It runs in the browser, needs no installation and gives instant results with the webcam. It is limited to images, sounds and body poses, and its models are made for the web. It is a free Google product: training can stay in the browser, but saving projects in the cloud needs a Google account, something to check against your school’s policy. Great for a first lesson, less so for a unit that ends with students’ own apps.

Machine Learning for Kids

Built for younger students and Scratch. It covers text, images, numbers and sounds, and teachers can set up class accounts. According to its own help pages, some project types, such as text, are trained and stored in the cloud, while others are trained on the student’s computer. Ideal if your school teaches with Scratch.

Lobe

Many teachers liked Microsoft’s Lobe for image classification. Its desktop application is no longer under development, so it is not a safe choice for a new unit.

MagicaLCore EDU

Made for schools that teach with iPads:

• Runs on the school iPads you already have, with an M1 chip or later and iPadOS 26.
• Private by design. Student photos, sounds and data stay on the iPad and are never uploaded to train a model. It collects no personal data, shows no ads, tracks nothing and has no accounts. Its terms are accepted by the school, not by students.
• Works offline. Training happens on the iPad, so a weak network never stops a lesson.
• Six model types, so students see machine learning in images, sounds, data, text and hand gestures.
• Fits in a lesson. Pause a training when the bell rings and continue next class from History.
• From a model to an app. Students export a .mlmodel and build an app around it in Swift Playgrounds, on the same iPad. It can even be the heart of a Swift Student Challenge project.
• A complete unit. A teacher’s guide with six solved exercises, one per model type, each with a rubric, plus a student notebook and a trial guide, in 13 languages.
• Simple to buy. Pay once, with no subscription, through Apple School Manager. Schools that buy 20 or more licences get 50% off.

Our recommendation

Use Teachable Machine for a five minute first taste on any computer. Choose Machine Learning for Kids if your students live in Scratch. If your school teaches with iPads and you want students to train real models, keep their data on the device and finish with an app of their own, try MagicaLCore EDU. Write to us from your school’s email and we will send a free trial through TestFlight, together with the trial guide and its first exercise.