Imagination Garden

Inspiration

Let every kid have the childhood of their dreams. This is just the first step.

What it does

Every day, parents get a new illustrated bedtime story that is wholesome and educationally valuable. Stories are set in the same fictional universe. The parents can make requests for the lessons in the story, such as courage, friendship, or curiosity.

How we built it

GPT-3 (Davinci) is the language model, chained with multiple steps. The illustration is by Stable Diffusion with an aesthetic gradient for a pretty watercolor style. A lot of work happened in Notion and talking on the phone with potential users, educators, and child psychologists.

Challenges we ran into

Sometimes, GPT has strange ideas. Because we're dealing with content for children, wholesomeness needs to be 100% guaranteed. So we're adding a human QA step in the creation pipeline with Amazon A2I.

Accomplishments that we're proud of

According to the Internet, it's not trivial to generate stories with AI that have a meaningful beginning, arc, and ending. Ours do!

What we learned

When it comes to content for children, automatic generation is not appropriate because of occasional strange turns. We added human approval into the pipeline. This way, we'll have guaranteed safe content. Parents can request story lessons, such as courage, friendship, or curiosity. They get a new illustrated and vetted story every day. Children love immersing in a consistent fictional universe, so we're focusing on that!

What's next for Imagination Garden

Author Ken Liu (the Paper Menagerie, engl. Three Body Problem) is advising us on the structure of stories! Find me to chat, and GPT-3 adds: In the future, we hope to continue improving and expanding Imagination Garden to bring joy and learning to even more children around the world. We look forward to seeing the impact of our platform on the childhoods of kids everywhere.

Built With

• a2i
• gpt-3
• midjourney
• notion
• render
• stable-diffusion