Turning daily walks into creature collection adventures

WorldWisp turns familiar subjects such as puddles on your commute, streetlights, or the sofa in your room into encounters with collectible creatures. The goal is not to escape daily life, but to make daily life itself feel worth noticing again.

The loop is intentionally simple: take a photo, get an encounter, add it to the collection, and feel like trying again tomorrow. The product is designed around making that loop light enough to repeat while still keeping a sense of discovery.

What makes it different

One-off image generation can be fun in the moment, but it rarely leaves anything behind. In WorldWisp, the same motif can grow as the same species over time, while each encounter still changes its mood or appearance. That turns the experience from a single output into a longer loop of discovering, collecting, and rediscovering.

The other key point is placement quality. The system uses scene context to decide how a creature should appear in the original photo, so the result feels less pasted on and more like an encounter that belongs in that scene.

How it works

The current implementation is a serverless web app. The frontend uses React, while the backend runs on AWS Lambda and API Gateway. Data lives in DynamoDB, images are stored in S3, and the pipeline combines Bedrock Nova models for understanding and text generation with gemini-2.5-flash-image for encounter rendering.

The flow is organized into three stages:

1. Capture The app moderates the input, understands the image, and extracts a core_word that represents the motif
2. Discover It decides whether the subject should connect to an existing species or become a new one
3. Encounter It generates a creature image that fits the photo context and records the result in the collection

This structure is what lets the product keep species continuity while still making repeated encounters feel fresh.

What the demo shows

The current demo lets you upload one photo, see the generated encounter, and then inspect how that result accumulates inside the collection.

1. Upload one photo
2. Review the generated encounter image, character name, and description
3. Open the collection to revisit the species history and recent encounters

That is enough to understand the full loop quickly, even on a first visit.

What I am improving now

The current focus is on making character designs more appealing and helping encounter images blend more naturally into the source photo. Longer term, I want to move the experience toward a more mobile-first UX where capture, encounter, and collection browsing feel even lighter.

The full submission write-up is published on AWS Builder.

AWS Builder article