Why This
Matters

Children today grow up surrounded by AI in a way no generation has before, and visual generative tools can now create finished paintings from a few words. That makes it easy for anyone to skip years of practice — but it can also quietly change how children learn to draw in the first place. Instead of spending time on brush strokes, many now start from an AI image and build from there.

Painting matters for children well beyond the picture itself: it trains their hands, builds creativity, and gives them something to feel proud of. The risk sits in the quality gap — when a five-year-old compares their own painting to a flawless AI one, the result is either disappointment or a quiet habit of just copying what the machine made instead of creating from their own ideas.

• 01Find design ideas that let AI help children without making them feel disappointed or dependent on copying.
• 02Build a simple, safe AI painting tool — working name Peekaboo, shipped as Picabon — that encourages a child's own style.
• 03Give suggestions for future apps and toys, so AI can be used in a genuinely positive way for children's art and learning.

What Painting
Is Actually For

Pablo Picasso spoke often about the value of children's painting and its role in real art.

"It took me four years to paint like Raphael, but a lifetime to paint like a child."Pablo Picasso

He learned skilled, realistic painting in a few years as a young man — but the free, honest way children paint took him a lifetime, and he spent his later Cubist years chasing that same childlike freedom. Research backs the instinct: Viktor Lowenfeld's stages of artistic development show that early free painting builds imagination and problem-solving, skills that support creativity for life. Left unprotected, that instinct fades fast — George Land and Beth Jarman's longitudinal creativity tests found a decline from 98% "genius level" divergent thinking in young children to far lower by the teenage years.

Pablo Picasso

Three existing tools were researched directly, and each — in its own way — nudges kids toward copying rather than creating.

Picabon fills the gap between the two failure modes above: it never locks a child into copying fixed samples, and it never hands over a perfect result. It works in small, encouraging steps — light suggestions only when needed — while the child always leads.

From Troll
to Yarn

Step 1 · Abandoned Sketch

Step 2 · Reference

Step 3 · Finalized Character Design

Step 4 · Mia's AI Avatar (Imaginary Story)

The Prototype

Core Experience

A functional prototype was built end to end. It takes a child's painting, extracts its visual elements, retrieves relevant chunks from the children's-literature database, and generates a short, personalised, spoken story plus one gentle suggestion — delivered by the knitted 3D character.

Seven Components

• Painting input & feature extraction
• Textual description generator
• Story retrieval engine (semantic search)
• Story & feedback generator (LLM + system prompt)
• Character database + voice
• Text-to-speech + animation trigger
• UI / character display layer, in Figma

Design Constraints

Young children's limited reading ability demanded a very simple interface. Child safety and content appropriateness had to be designed in, not bolted on. And the system had to hold up against wildly different drawing styles — a scribble and a careful sketch both had to earn a story.

Mia's session, in the Art room

01 — Reads the painting back to Mia

02 — Closes with an original song, not a critique

What
Surprised Me

Picabon is the only one of the reviewed tools that avoids visual perfection or copy-models altogether, using storytelling as the entire feedback mechanism.

What Actually
Happened

Mixing drawing and storytelling genuinely engaged children in testing. They didn't just listen to a story — they became active participants by feeding their own ideas in through painting first. Children aged 6–10 showed more interest in trying new colours, shapes and ideas after story-based suggestions than after copying-based tools, evidence that Picabon builds confidence and personal style early, letting technology support real growth instead of replacing it.

Against the research question: by using verbal stories with no visual models and no copying rules, Picabon proves AI can support a child's personal style when it's built around motivation and play instead of perfection.

What This Proves,
What's Still Open

The project delivered an AI-supported storytelling application that responds directly to a child's own drawing — built to preserve their unique visual expression rather than push them toward a predefined style. Children are encouraged to retell the generated story to parents or friends, exercising narrative thinking and confidence in self-expression along the way. AI acts as a creative partner here, not a replacement for a child's creativity.

The storybook database isn't complete, which caps narrative diversity; the system still struggles to interpret very abstract drawings accurately; and time constraints ruled out the originally planned full 3D animation. The app also hasn't been tested extensively across different environments yet — age adaptability, broader language support, and parental controls all need more work before this is real-world ready.

• →Multiplayer or collaborative storytelling, so painting and story-building happen between kids, not just one child and one character.
• →Parental controls and progress tracking, closing the safety gap flagged above.
• →Improved drawing interpretation — moving to a real multimodal model such as GPT-4o or LLaVA.