
Scratch AI Extensions Tutorial: Build Your First AI Project (2026)
Version 2.4 — Updated April 2026 | Reviewed by Felix Zhao
By KidsAiTools Editorial Team
Reviewed by Felix Zhao (Founder & Editorial Lead)
Step-by-step tutorial to build AI-powered Scratch projects. Image recognition, sound classification, and body pose detection — no coding experience needed.
Scratch AI Extensions Tutorial: Build Your First AI Project (2026)
Scratch isn't just for simple animations anymore. With AI extensions powered by ml5.js and Google's Teachable Machine, kids can build projects that recognize faces, classify objects, respond to voice commands, and detect body movements — all using Scratch's familiar drag-and-drop blocks. Over 100 million kids use Scratch worldwide (MIT Media Lab, 2025), and AI extensions are the fastest-growing category. This tutorial walks you through 3 AI projects, from beginner to intermediate, in about 2 hours total.
What You Need
- A computer with Chrome browser (AI extensions work best in Chrome)
- A webcam (built-in or external)
- A microphone (built-in works fine)
- Scratch account (free at scratch.mit.edu) — optional but recommended for saving projects
- No coding experience needed!
Understanding Scratch AI Extensions
Scratch's AI capabilities come from community-created extensions that connect to machine learning models:
| Extension | What It Does | Best For | Difficulty |
|---|---|---|---|
| ml5.js Image Classifier | Recognizes objects through webcam | "What is this?" games | Beginner |
| Teachable Machine | Train custom AI models | Gesture games, sound apps | Beginner |
| PoseNet | Detects body position/movement | Dance games, exercise apps | Intermediate |
| Speech-to-Text | Converts voice to text | Voice-controlled stories | Intermediate |
| Text-to-Speech | Makes projects talk | Interactive characters | Beginner |
Project 1: AI Animal Identifier (Beginner — 30 minutes)
What you'll build: A project where kids show animals (toys, pictures, or real pets) to the webcam, and Scratch identifies what animal it is.
Step 1: Open Scratch with Machine Learning
- Go to machinelearningforkids.co.uk — this site provides the easiest Scratch + AI integration
- Click "Try it now" → "Log in" (create a free account or use as guest)
- Click "Projects" → "+ Add a project"
- Name it "Animal Identifier"
- Choose "Images" as the recognition type
Step 2: Train Your AI Model
- Click your project → "Train"
- Create labels: "Dog", "Cat", "Bird" (or whatever animals you have)
- For each label, add 10-15 images:
- Click "Add example" → use webcam to take photos of toy animals
- OR drag in images from the internet
- Click "Back to project" → "Learn & Test"
- Click "Train new machine learning model" — wait 1-2 minutes
- Test it! Show an animal to the webcam and see if it's recognized correctly
Step 3: Build in Scratch
- Click "Make" → "Scratch 3"
- Scratch opens with special AI blocks already loaded
- Build this script:
when green flag clicked
forever
recognise image from camera (label)
if <(label) = [Dog]> then
say [That's a dog! 🐕] for (2) seconds
play sound [dog bark]
end
if <(label) = [Cat]> then
say [That's a cat! 🐱] for (2) seconds
play sound [meow]
end
end
- Click the green flag and show animals to the webcam!
What Kids Learn
- AI needs training data (examples) to learn
- More examples = better accuracy
- AI can make mistakes when images are unclear
Project 2: Rock-Paper-Scissors Against AI (Beginner — 45 minutes)
What you'll build: Play rock-paper-scissors against the computer using hand gestures recognized by AI.
Step 1: Train Hand Gestures
- Create a new project → Type: "Images"
- Create 3 labels: "Rock", "Paper", "Scissors"
- For each gesture, take 20 webcam photos from different angles
- Train the model
Step 2: Game Logic in Scratch
when green flag clicked
say [Show your hand! Rock, Paper, or Scissors?] for (2) seconds
wait (3) seconds
recognise image from camera (player_choice)
set [computer_choice] to (pick random (1) to (3))
if <(player_choice) = [Rock] and (computer_choice) = [2]> then
say [Paper beats Rock. Computer wins!]
end
... [continue for all combinations]
What Kids Learn
- AI can recognize visual patterns (hand shapes)
- Computer programs use logic (if/then) to make decisions
- Random number generation creates unpredictable opponents
Project 3: Body-Controlled Dance Game (Intermediate — 45 minutes)
What you'll build: A dance game where a character copies your body movements in real-time using PoseNet body detection.
Step 1: Set Up PoseNet Extension
- Open Scratch at stretch3.github.io (enhanced Scratch with PoseNet)
- Click "Add Extension" → "PoseNet2Scratch"
- Allow webcam access when prompted
Step 2: Make a Character Follow Your Movements
when green flag clicked
forever
set x to (nose x) * (2)
set y to (nose y) * (2)
if <(right wrist y) > (right shoulder y)> then
switch costume to [hands-up]
say [Hands up! 🙌]
end
end
Step 3: Add Dance Challenges
Create a sequence of poses and challenge the player to match them:
- Hands up! ✋✋
- Touch your toes! 🦶
- T-Pose! ✈️
- Dab! 💃
The AI checks if the player's body position matches the target pose.
What Kids Learn
- AI can understand body position in 3D space (pose estimation)
- Real-time AI processing (the model runs continuously)
- Game design: how to turn AI capabilities into fun challenges
Troubleshooting Common Issues
| Problem | Solution |
|---|---|
| "Webcam not working" | Check browser permissions (Chrome → Settings → Privacy → Camera) |
| "Model won't train" | Need at least 10 examples per label; check internet connection |
| "AI keeps getting it wrong" | Add more training examples, ensure good lighting, try different angles |
| "Extension not loading" | Use Chrome (not Safari/Firefox), clear cache, try incognito mode |
| "Scratch is slow" | Close other browser tabs, AI extensions use significant CPU |
Next Steps After These Projects
Ready for more? Try these advanced Scratch AI projects:
- Emotion Detector: Train AI to recognize happy, sad, surprised faces
- Language Translator: Combine Speech-to-Text + Translation API + Text-to-Speech
- Smart Pet Feeder Simulator: AI recognizes when your "pet" sprite is hungry based on time and activity
- AI Music Composer: Use Scratch's sound blocks with AI-generated melodies
Frequently Asked Questions
Do Scratch AI extensions work on tablets?
Unfortunately, most AI extensions require Chrome on a desktop/laptop computer. The webcam and processing requirements are too heavy for tablets. Standard Scratch works on tablets, but AI features do not.
Are Scratch AI extensions free?
Yes, completely. Machine Learning for Kids, PoseNet2Scratch, and all Teachable Machine integrations are free. No paid tier exists. The only "cost" is a computer with a webcam.
Can my child share AI Scratch projects?
Projects using Machine Learning for Kids can be shared within that platform. Standard Scratch community sharing doesn't support AI extensions directly — the shared project won't include the trained model. Workaround: share a video recording of the project in action.
What age is best for Scratch AI?
Ages 8-9 for guided projects (with parent helping). Ages 10-12 for independent projects. Ages 13+ may want to explore more advanced tools (Teachable Machine standalone, Python ML). The AI concepts are the same regardless of tool — Scratch just makes them visually accessible.
Explore more AI coding tools for kids. Try Teachable Machine for training custom AI models. Browse all 55+ AI tools on KidsAiTools.
What Success Looks Like (And What It Doesn't)
Parents often measure AI education success by the wrong metrics. Here's a recalibration:
Success IS:
- Your child asks "how does this work?" instead of just using AI passively
- Your child can explain an AI concept to a friend or sibling in their own words
- Your child spots an AI-generated image or text without being told
- Your child chooses to use AI for creating, not just consuming
- Your child questions AI outputs: "Is this actually true?"
Success IS NOT:
- Your child uses AI tools for X hours per week (time ≠ learning)
- Your child can list 20 AI tools by name (knowledge ≠ wisdom)
- Your child gets A's by using AI for homework (grades ≠ understanding)
- Your child impresses adults by using "AI vocabulary" (jargon ≠ comprehension)
The 3-Month Challenge
Want to put this article into action? Here's a structured 3-month plan:
Month 1: Explore
- Try 2-3 different AI tools from this article
- Spend 15-20 minutes per session, 3-4 times per week
- Focus: What does my child enjoy? What frustrates them?
- Goal: Identify 1-2 tools that genuinely engage your child
Month 2: Build
- Settle on 1-2 primary tools
- Complete at least one structured project or challenge
- Start connecting AI learning to school subjects
- Goal: Your child creates something they're proud of
Month 3: Reflect
- Discuss what they've learned about AI (not just what they've done with it)
- Evaluate: Has their critical thinking about technology improved?
- Decide: Continue with current tools, try new ones, or adjust approach
- Goal: AI literacy becomes a natural part of your child's thinking, not just screen time
Expert Perspective
AI education researchers consistently emphasize three principles:
Process over product — How a child interacts with AI matters more than what they produce. A child who asks thoughtful questions learns more than one who generates impressive outputs.
Transfer over mastery — The goal isn't mastering one AI tool. It's developing thinking patterns that transfer to any tool, any technology, any future challenge.
Agency over compliance — Children who choose to use AI thoughtfully are better prepared than those who follow AI rules without understanding why.
These principles should guide every decision about AI tools, screen time, and learning activities.
Continue learning with our 7-Day AI Camp. Explore AI tools by age group.
Ready to try this with your child?
Knowing which AI tool helps for homework is one thing — getting your child to actually use it productively is another. These five products are how we bridge that gap at home.
| Your child's goal | Try this | Why it works |
|---|---|---|
| Build 3D creations hands-on | 🧱 3D Block Adventure | Browser-based 3D building with 15 AI-guided levels. Ages 4-12, no downloads. |
| Play an AI game right now | 🎨 Wendy Guess My Drawing | A 60-second drawing game where the AI tries to guess. Ages 5-12, zero setup. |
| Learn AI over 7 structured days | 🏕️ 7-Day AI Camp | Day 1 is free. 15 minutes a day covering art, story, music, and safety. |
| Create art, stories, or music | 🎨 AI Creative Studio | Built-in safety filters. Three free creations a day without signing up. |
| Pick the right AI tool for your child | 🛠️ 55+ Kid-Safe AI Tools | Filter by age, subject, safety rating, and price. Every tool parent-tested. |
All five start free, run in the browser, and never ask for a credit card up front.
📋 Editorial Statement
Written by the KidsAiTools Editorial Team and reviewed by Felix Zhao. Our guides are written from a parent-builder perspective and focus on AI literacy, age fit, pricing transparency, and practical family use. We do not currently claim named external expert review or a child-test panel. We may earn commissions through referral links, which does not influence our reviews.
If you find any errors, please contact support@kidsaitools.com. We will verify and correct as soon as we can.
Last verified: April 22, 2026