Scratch AI extensions for kids: what's really AI, and a first project

Scratch has one built-in extension that is genuinely AI: Face Sensing, added in October 2025, which uses a pre-trained machine learning model to find a face and its features through the webcam, on your own computer. Text to Speech and Translate send words to outside services, and Video Sensing isn't AI at all. To train your own model and use it in a Scratch project, children use free add-ons: Machine Learning for Kids, MIT's RAISE Playground, or the volunteer-run Stretch3. This guide was written by Peabody (peabody.kids), a free reading and learning site that is free through the end of 2026. None of these tools is ours, and nobody paid to be listed. We opened every one on 18 October 2026.

Which Scratch AI extensions exist? The quick table

Scratch itself is free. Its parents' page says it is "primarily designed for 8 to 16 year olds," and younger children use it with their parents.

Is Scratch's text to speech AI? Being precise about each extension

Open the Scratch editor, click the blue button at the bottom left, and you'll see the extension library. Here's what each camera or language extension really does.

Face Sensing. The Scratch Team announced it on 8 October 2025 as an "AI-powered feature." Its blocks were trained in advance on photos of faces and don't learn from children's data. Scratch's Face Sensing guide says "None of your data is stored or sent to Scratch or any other site." The guide also warns it may not work on some Android and iOS devices, and may run slowly on some Chromebooks and older computers. The blocks we saw on 18 October 2026: go to (nose), point in direction of face tilt, set size to face size, when face tilts (left), when this sprite touches a (nose), when a face is detected, a face is detected?, face tilt and face size. The "go to" list offers nose, mouth, left eye, right eye, between eyes, left ear, right ear and top of head.

Text to Speech. Makes a sprite talk, with voices like alto, tenor, squeak, giant and kitten. The extension library marks it as a "Collaboration with Amazon Web Services": the speech comes from an Amazon service, so the words your child types go to Amazon to be turned into sound. The child doesn't train anything or see how the voice is made, so it's a useful tool rather than a lesson about AI.

Translate. One block, translate ( ) to ( ), which the library marks as a collaboration with Google. It needs the internet. Same as above: handy, but the AI part is hidden.

Video Sensing. Reacts to movement in front of the webcam. It isn't machine learning: Scratch's own code works out motion by comparing each video frame with the one before. It's still a great first camera project, and Scratch's Video Sensing page says the blocks don't collect or store recordings or images of your face.

So if your child asks "where's the AI in Scratch?", the honest answer is Face Sensing, plus the outside tools below.

Does Scratch need an account?

Not to make things. The editor opens without signing in, and the File menu has "Save to your computer" and "Load from your computer." An account is needed to save projects online and to share them in the community, and sharing needs a confirmed email address. Scratch's parents' page says it limits what it collects at sign-up and doesn't sell or rent account information.

On AI and children's data: the Scratch Foundation's Scratch & AI page says "right now, we are not training any AI models on Scratchers' data," and sets out its rules for any AI features it adds.

A first Scratch AI project: a hat that follows your head

This takes about 15 minutes, needs a computer with a webcam, and no account.

1. Open the Scratch editor. Click the cat sprite in the sprite list and delete it (the little bin icon).
2. Choose a hat. Click "Choose a Sprite" (the cat button, bottom right), search "hat", and pick one, such as Wizard Hat or Party Hats. Or click the paintbrush and draw a crown.
3. Add Face Sensing. Click the extension button at the bottom left and choose Face Sensing. The browser asks to use the camera; say yes. You'll see yourself on the stage.
4. Build the script. From Events, drag when green flag clicked. From Control, add forever. Inside it, put the Face Sensing blocks go to (top of head) and set size to face size.
5. Click the green flag. Move your head. The hat follows, and gets bigger as you lean in.
6. Change one thing. Switch "top of head" to "nose" and draw a clown nose. Or add point in direction of face tilt so the hat tips with you.
7. Save it to your computer from the File menu, if you want to keep it.

What to talk about. "How does it know where your nose is?" It was trained on lots of photos of faces before it reached you, so it guesses. Try it with a hand over half your face, or in a dim room, and watch it lose track. That's the lesson: AI guesses from patterns, and it can be wrong.

No webcam? Try a talking translator instead. Add Text to Speech and Translate, then build when green flag clicked, ask (What should I say?) and wait, then speak (translate (answer) to (Spanish)). Remind your child not to type names or anything personal, because the words go to Amazon and Google.

Machine Learning for Kids: train your own model, then use it in Scratch

Machine Learning for Kids is a free site built by Dale Lane using IBM Watson services. Its three steps: collect examples of the things you want recognised, train the computer, then make a game in Scratch that uses what it learned. It can learn to recognise text, numbers, images and sounds.

You don't need to register. "Try it now" offers every feature without logging in, but its own note says projects made that way are deleted after at least four hours. When we opened it, a "Verify you are human" check sat beside the button. Class accounts are set up by a teacher. Our parent's guide to Machine Learning for Kids walks through it.

RAISE Playground: MIT's Scratch with AI blocks

The RAISE Playground comes from the MIT RAISE Initiative and the Personal Robots Group at the MIT Media Lab. It looks like Scratch, and opening it needed no sign-in. Its extension library, as of October 2026, includes Text Classification, Teachable Machine ("use your Teachable Machine models in your Scratch project"), Hand Sensing, Face Sensing, Body Sensing, Object Detection and Music Creation, alongside the usual Scratch extensions. Its curricula, such as Dancing with AI and How to Train Your Robot, are written for middle school. It's a good next step after Face Sensing for a child of about 11 and up.

Stretch3 and Teachable Machine in Scratch

Stretch3 is a modified Scratch made by volunteers, not by the Scratch Team. Its extension list includes TM2Scratch ("recognize your own images and sounds," made with Google's Teachable Machine), TMPose2Scratch for poses, ML2Scratch, and hand, face and speech recognition blocks. It's how many teachers connect Teachable Machine to Scratch. See our Teachable Machine walkthrough for the training half.

One caution: Stretch3's list also has ChatGPT and other generative AI extensions. Those put a chatbot inside a project, which is a different thing from a child training a small model. That's a choice for a grown-up to make, not one to stumble into.

What about "AI coding" tools that write the code for you?

Some tools let a child describe a game and have AI write it. Peabody has one, the Game Studio: on a computer or tablet, a child describes a game in words, AI writes it, and it plays inside a locked box with no internet. It's fun, and it teaches how to give clear instructions. But it isn't learning to code. The child never sees or changes the code. If you want your child to understand how programs work, Scratch is the better place, and the projects above add AI to it without hiding the logic.

Where Peabody fits

Peabody is mainly a reading site: free AI-written stories at a child's level in the library, and learning games with questions made and marked by code, not AI. It doesn't teach coding. For a child who's just starting to ask what AI is, our AI explained for 7-year-olds page is written to read together, and the Google AI games guide has gentler first steps than Scratch.

In a homeschool

Scratch suits a homeschool as a weekly coding block: one 30- to 45-minute session on the same day each week, starting with the Face Sensing hat and moving on to Machine Learning for Kids or the RAISE Playground after a few weeks for children of about 10 and up. Save each project to the computer and keep a screenshot for your records. Peabody fits the other days, not the coding: 10 to 15 minutes of reading from the library at your child's level, and typing practice helps a child who types slowly in Scratch (free homeschool typing). The honest limit: Peabody teaches no coding, and its Game Studio writes the code for the child, so it is no substitute for Scratch. See AI in homeschooling and the homeschool hub.

Learn more elsewhere

• Scratch: Face Sensing guide: the official how-to and privacy answers.
• Scratch & AI: the Scratch Foundation's rules for AI features and children's data.
• Scratch for parents: ages, accounts and privacy in plain words.
• RAISE Playground: MIT's Scratch-style editor with AI blocks and middle-school curricula.

For about seventy more free AI-learning sites, sorted by age, see where to learn about AI.

Questions parents ask

Does Scratch have AI?

Yes, one extension. Face Sensing, added in October 2025, uses a pre-trained machine learning model to find a face and its features through the webcam. Text to Speech and Translate send words to Amazon and Google services, and Video Sensing measures movement without AI. Training your own model needs an add-on such as Machine Learning for Kids or the RAISE Playground.

Is Scratch free, and does my child need an account?

Scratch is free. A child can open the editor, build projects and save them to the computer without an account. An account is needed to save online and share projects, and sharing needs a confirmed email address.

Is Scratch's Face Sensing private?

Scratch says Face Sensing runs on your own device and that no data is stored or sent to Scratch or any other site. The model was trained before it reached you and doesn't learn from your child. The browser asks for camera permission first, and you can say no.

Is Video Sensing in Scratch AI?

No. Video Sensing works out how much and which way things move by comparing each webcam frame with the one before. It's a good first camera project, but it doesn't learn or recognise anything. Face Sensing is the extension that uses machine learning.

How do I use Teachable Machine with Scratch?

Train a model on the Teachable Machine site, then load it into a Scratch-style editor that has a Teachable Machine extension. The MIT RAISE Playground has one, and the volunteer-made Stretch3 has TM2Scratch. Scratch's own editor doesn't include it.

What age is right for Scratch AI projects?

Scratch is designed mainly for ages 8 to 16. The Face Sensing hat project suits about 8 and up with a grown-up nearby. Machine Learning for Kids and the RAISE Playground suit about 10 or 11 and up, and RAISE's curricula are written for middle school.