StoryLoom

An AI-powered storybook that crafts magical tales, brings them to life with narration, and lets kids color their imagination.

Comment

Inspiration

Every child deserves to see themselves as the hero of their own story. But creating a personalized children's book has traditionally been expensive, time-consuming, and required artistic skills most parents don't have. I wanted to democratize storytelling—making it possible for anyone, regardless of technical or artistic ability, to create a professional-quality, fully illustrated children's book in minutes. The inspiration came from watching parents struggle to find books that truly resonated with their children's unique interests, backgrounds, and imagination. Storyloom bridges that gap by putting the power of AI-driven creativity directly into the hands of storytellers everywhere.

What it does

Storyloom is an end-to-end AI-powered children's storybook creation platform that transforms a simple idea into a complete, illustrated, narrated storybook. Here's what users can do:

Story Creation Journey: Start with either a "Surprise Me" option for randomly generated stories or customize every detail through an intuitive multi-step form Define character names, gender, custom elements (like favorite animals or hobbies), and story settings Choose the story length (1-10 pages) Watch as AI generates a complete narrative with consistent characters and engaging plot

*Voice Narration: * Smart text-to-speech integration with ElevenLabs that only generates audio when users click play, Interactive Reading Experience: Beautiful page-by-page layout with play/pause controls next to each page number for narration PDF Export: Download the complete storybook as a high-quality PDF for printing or sharing

How we built it

Tech Stack: Frontend: Next.js 15 + React 19 + TypeScript with custom SCSS styling AI Models: Google Vertex AI (Imagen 3), Google Gemini 2.5 Flash, ElevenLabs

Multi-Agent AI Orchestration: "Storyteller" Agent (Gemini): Generates structured story blueprints with detailed character profiles—including exact clothing colors, hair style, facial features, and physical descriptions using Zod schemas "Prompt Engineer" Agent (Gemini): Creates page-specific image prompts that programmatically inject the exact character descriptions from the blueprint into every single image generation request "Illustrator" Agent (Imagen 3): Receives these meticulously crafted prompts and generates watercolor illustrations with consistent character appearance across all pages "Narrator" Agent (ElevenLabs): Converts story text to natural speech with smart lazy-loading—only generates audio when users click play, then batch-processes all pages

Challenges we ran into

Maintaining Visual Character Consistency Across AI-Generated Images AI image models typically generate each image independently, leading to characters that look different on every page. We solved this by creating a structured character schema that captures exact physical details (clothing colors, hair style, features) and programmatically injecting these descriptions into every single image prompt, ensuring the "Illustrator" agent receives identical character briefings for each page. Vertex AI Authentication and Model Compatibility Initial attempts with simple API keys failed, requiring a migration to Google Cloud's Application Default Credentials (ADC) system

What's next for StoryLoom

Performance & Scalability: Optimize story generation for longer books (50-100 pages) by implementing parallel processing, streaming responses, and database caching to reduce generation time from minutes to seconds for large-scale stories. Deployment & Market Reach: Acquire a custom domain and deploy to production on Google Cloud Run, then launch targeted marketing campaigns to reach parents, educators, and children's content creators through social media, parenting forums, and educational platforms.

Built With

Share this project:

Updates