MakeMyStory started from a simple observation: my daughter loved stories where she was the main character, but creating those stories took enormous effort. The first version was a weekend hack — a script that stitched together GPT outputs with placeholder images. The response from other parents was immediate and enthusiastic.
How It Works
The product flow is intentionally simple. A parent enters their child's name, age, a few interests (dinosaurs, space, ballet — whatever the child loves), and optionally a moral theme (kindness, courage, sharing). The system generates a 10-12 page illustrated story within minutes.
Under the hood, the architecture is more involved:
Story generation uses a multi-step prompting pipeline. Rather than generating the entire story in one pass, the system first creates a story outline with plot beats, then generates each page's text sequentially while maintaining narrative coherence. This approach produces dramatically better stories than single-shot generation.
Illustration generation is the hardest technical challenge. Each story needs 10-12 illustrations that are visually consistent — the main character should look the same across all pages. The system uses a combination of detailed prompt engineering and style-locked generation to maintain consistency.
Quality control includes automated checks for age-appropriateness, narrative coherence, and illustration quality, with graceful fallbacks when generation quality drops below threshold.
Technical Decisions
Building MakeMyStory involved several interesting technical tradeoffs:
Why Next.js? Server-side rendering for SEO, API routes for the generation pipeline, and React Server Components for the story reading experience. The framework handles a lot of the infrastructure complexity that would otherwise require separate services.
Streaming generation UX. Rather than making users wait for the entire story, the interface progressively reveals pages as they're generated. This turns a 2-minute wait into an engaging experience.
Cost management. AI generation costs are significant at scale. The system uses intelligent caching, generation batching, and model selection based on the specific task to keep per-story costs viable.
What I Learned
Building a consumer AI product taught me lessons that don't come from enterprise analytics work. Speed of iteration matters more than architectural perfection. User perception of quality is as important as actual quality. And the gap between "demo quality" and "production quality" in AI applications is much larger than most people realize.