## Inspiration AI images are slow and expensive, even when a useful image already exists. Also sometimes you don't need a exact fit e.g. a stock image. We wanted to create a library that has millions of images that can be returned instantly and that grows depending on what people search.
What it does
wagmi.photos searches 500,000+ open images and returns the closest match instantly. Missing prompts are queued and generated based on demand.
How we built it
Cloudflare handles search and metadata at the edge. Genblaze runs background generation, while Backblaze B2 stores images, sizes, and provenance. We use the open dataset called PD12M for the first 500,000 images (there are 12+ million we can use)
Challenges we ran into
Keeping embeddings consistent across two runtimes was tricky. Scaling semantic search and background jobs reliably also took several iterations.
Accomplishments that we're proud of
We built a live, OpenAI-compatible API with fast cache hits, durable B2 storage, and a library that improves as people use it.
What we learned
A cache can do more than save money—it can show what users actually want. We also learned how important provenance and retry-safe workflows are.
What's next for wagmi.photos
Grow the library, support more generation providers, improve matching, and make the service easier to use from apps and coding agents.
Built With
- backblaze
- cloudflare
- genblaze
- gmicloud
- python
Log in or sign up for Devpost to join the conversation.