StoryForge

STORYFORGE Your Story. Your Voice. Your Book

Inspiration

The best storytellers have always known something that technology forgot: a story told to you is entertainment, but a story told with you is something else entirely. It is a collaboration. It is ownership. It is the reason children ask to hear the same tale again and again, not because they forgot the ending, but because they want to live inside it one more time.

StoryForge was built from that understanding.

We did not want to build a story generator. We wanted to build a storyteller. One that listens as much as it speaks. One who notices when a child goes quiet and asks what they are thinking. One that takes a seven-year-old's suggestion that the dragon should actually be friendly and runs with it, then asks what the friendly dragon's name should be.

The insight at the centre of StoryForge is simple: the best story a child will ever hear is the one they helped create. We built a live AI storyteller that makes that true for every child, every time, using nothing but their voice.

The Problem

Children's technology today does not know how to listen.

Passive screen entertainment captures attention without building anything. Edtech products structure learning without leaving room for imagination. AI story generators produce polished output from a prompt and then go silent, waiting for the next instruction. None of them are in conversation with the child. None of them adapt to what the child says, feels, or wants in real time.

The result is a fundamental mismatch. Children are wired for dialogue. They do not experience stories as audiences. They experience them as participants, talking back to the television, narrating their own play, casting themselves as the hero without being asked. Every child who has ever been read a bedtime story has interrupted it. That interruption is not a problem. It is the most important part.

No product today treats a child's voice as a live, continuous input that shapes the story as it unfolds. StoryForge does.

What StoryForge Does

StoryForge is a live AI storytelling companion that builds a fully illustrated, narrated storybook through real-time voice conversation with a child. Gemini Live is always on throughout the entire session, listening, responding, and co-creating. The book is not generated from a prompt. It is the artifact of a live conversation.

The core experience works as follows:

The child opens the app and the AI storyteller greets them by name, warm and present, and asks what kind of story they are in the mood for
The child responds however they like. One sentence, one word, or a rambling description of everything they want. The AI listens, reflects back what it heard, and begins
The AI narrates the first page aloud. Then it pauses and asks the child a question that shapes what comes next: "There are two paths in the forest. Which one should she take?"
The child answers by voice. The AI builds the next page from their response, incorporating their exact words and ideas into the story
An illustration for each page generates inline as the AI speaks, rendered in the visual style the child chose at the start
The child can speak at any point. They can interrupt, redirect, add a character, change the tone, or ask the AI to slow down and describe a scene in more detail. The AI responds naturally, mid-story, without breaking the flow
The AI notices emotional cues. If the child laughs, the story gets funnier. If they go quiet, the AI checks in. If they ask a question about the world of the story, the AI answers in character and weaves the answer into the next page
The session closes when the story reaches its ending, collaboratively decided, and the full illustrated book is exported as a PDF and shareable web link, the child keeps forever

Google Technologies We Leveraged

Gemini Live API The soul of StoryForge. Gemini Live is active for the entire session, from the opening greeting to the final page. It listens continuously, handles natural interruptions, responds to emotional cues in the child's voice, and maintains the full narrative context across the conversation. This is not a voice input layer on top of a generation pipeline. The story is built turn by turn through live dialogue. Without Gemini Live, StoryForge is a generator. With it, it is a storyteller.

Gemini Multimodal Output with Imagen Each page generates an inline illustration anchored to the scene as the AI narrates it, rendered in the visual style chosen at the start of the session. Prompts are scoped narrowly per style to maintain visual consistency across pages, so the finished book feels like a coherent illustrated work rather than a collection of individual images.

Gemini AI Studio Used during development to prototype the storyteller persona, conversational turn architecture, and child-safe content layer. Core prompts were developed and refined to shape how the AI opens a session, asks generative questions, incorporates child responses into the narrative, handles unexpected input, and closes a story warmly.

Model Configuration via Environment Variable The Gemini model name is managed through a MODEL_NAME environment variable, keeping the codebase flexible and easy to update across model versions without hardcoding.

Challenges We Faced

Keeping Gemini Live active and contextually coherent across an entire storytelling session required significant prompt architecture work. The AI needed to hold the full story state, character names, world rules, established tone, and prior child responses, across every turn without losing the thread or drifting in personality.

Designing generative questions was harder than it sounds. The AI's mid-page questions needed to genuinely shape the story, not feel like a quiz. They had to be open enough that any child's answer could be incorporated naturally, but specific enough to create meaningful narrative stakes.

Handling child interruptions gracefully in a live voice session required careful turn-taking design. A child who shouts "wait, I want a dragon" mid-narration needs the AI to pause, acknowledge, adapt, and continue without the experience feeling jarring or robotic.

Balancing AI authorship with child co-creation was an ongoing design tension. Too much AI control and the child is just an audience. Too little and the story loses coherence. The right balance required the AI to hold strong narrative intent while genuinely yielding creative decisions to the child at the right moments.

Accomplishments and Learnings

StoryForge is a genuine Live Agent. Gemini Live is not the input method. It is the relationship. The AI storyteller listens, adapts, and responds across the full arc of a session the way a skilled human storyteller would.

The conversational story structure works. Children who are given genuine creative agency over a narrative produce more engaged, more emotionally invested sessions than passive listeners. The questions the AI asks unlock that agency.

The finished storybook as a session artifact is a meaningful differentiator. The child does not just have a memory of a conversation. They have an illustrated, narrated book that is entirely, provably theirs. Their words are in it. Their decisions shaped it.

The child-safe content layer holds across a wide range of unpredictable child-led inputs without the AI becoming evasive or robotic in its refusals.

What we learned

Live Agent design for children is fundamentally about the quality of the questions the AI asks, not the quality of the content it produces. A great question from the AI creates more value than a great paragraph of narration.

Narrative state management in a live conversation is a different problem from narrative state management in a generation pipeline. The AI must track not just what happened in the story but what the child said, how they said it, and what it chose to do with that input.

Co-creation requires the AI to have genuine opinions. An AI that agrees with everything the child says produces a flat story. An AI that occasionally pushes back, "are you sure you want the hero to give up? What if there is one more thing they could try?", creates drama, investment, and better stories.

Future Work

StoryForge has a clear path forward beyond the hackathon.

Persistent story memory so the AI remembers a child's favourite characters, recurring worlds, and past adventures, building a creative universe across sessions
Collaborative mode where two children co-create with the AI mediating, incorporating both voices and attributing each contribution in the finished book
A personal library to save, revisit, and collect every storybook a child has made, building a creative archive that grows with them
Classroom mode designed for teachers, with group storytelling sessions, shared narrative decisions, and exportable class storybook collections
Narration voice personalisation, letting the child choose their storyteller character and voice
Multilingual support so a child can converse with the AI storyteller in their own language and receive a finished book written and narrated in kind

The core insight, that the best story a child will ever hear is the one they helped create through live conversation with an AI that genuinely listens, is one we believe has significant potential across play, education, and family life, well beyond what a hackathon build can fully explore.

Your story. Your voice. Your book.