Raimi's AI Art Lab

Visitors compose a prompt step by step with the museum character Raimi, choosing what they want to draw, and an AI generates the artwork. Designed so young visitors learn the principles of prompt engineering through play, with a 10-step (+5 advanced) choice-based prompt builder that shows how a prompt is put together.

The off-the-shelf GPT service used before would stop responding for no identifiable reason while visitors were using it. This is a hands-on exhibit, so a frozen screen meant the station sat unusable until staff intervened. After moving to a custom web app and serverless backend, it has not frozen once; the backend calls the OpenAI image-generation API, auto-watermarks every image with the museum logo, stores images on Cloudflare R2, and keeps generation metadata and stats in Supabase. Visitors take their creations home instantly via QR code, and because it runs as a web app, the exhibition kiosk could be locked down to a single full-screen browser.

It ran on the museum floor through June and July 2026: 3,049 images in June, rising above 8,000 in July, the museum's peak season, with generation volume measured in the app's own backend.