Self-host openflipbook

OpenFlipbook reimagines the flipbook.page concept as a self-hosted infinite canvas. Every page is a single AI-generated image; you start with a prompt or uploaded picture, then click anywhere to go deeper. A vision model interprets the click, a planner writes the next image request, and the cycle repeats. The stack runs locally via Docker Compose: a Next.js 15 frontend, a FastAPI backend (normally hosted on Modal but runnable locally), MongoDB for metadata, and RustFS as an S3-compatible blob store. All model calls use your own API keys — fal.ai for images, OpenRouter for the planner and click-resolution vision models. Features include swappable image providers with quality tiers, optional animation (one-shot MP4 or LTX streaming video), persistent permalinks that load from Mongo and object storage without re-generating, embeddable read-only worlds served over oEmbed or a simple script tag, a place inspector for renaming and locking locations, citations that surface source URLs, freehand region selection, a timeline scrubber, progressive rendering with a fast draft model, in-place editing with automated scoring, and a World Mode that treats pages as places on a persistent map. No built-in authentication exists; public deployments should be protected by a reverse proxy with TLS and auth. The codebase is MIT-licensed and under active development.