AI video production built entirely on Cloudflare

What it does OpenStory ingests a script and uses AI to generate a sequence of images, motion clips, and audio with a consistent visual style across scenes. It handles scene breakdowns, continuity tracking, and prompt engineering automatically. The whole stack runs on Cloudflare Workers at the edge.

The interesting bit The architecture is deeply tied to Cloudflare’s serverless primitives: Workflows manage durable generation jobs, Durable Objects stream real-time progress via SSE, and D1 with R2 handle data and storage. For local development, the entire stack runs inside Miniflare without Docker or a Cloudflare account.

Key highlights

• Script analysis via OpenRouter LLMs yields automatic scene breakdowns with camera angles, mood treatments, and continuity tracking.
• Fal.ai generates images and motion video, with enforced style consistency for characters, locations, color palettes, and lighting.
• Cloudflare Durable Objects push real-time generation progress to the frontend over server-sent events.
• Passwordless sign-in via Better Auth passkeys.
• The codebase uses oxlint, tsgo, and Knip for quality, plus Vitest and Playwright for testing.

Caveats

• Team workspaces and shared asset libraries are marked “coming soon,” so multi-user collaboration is not yet implemented.
• AI generation depends on external third-party APIs (Fal.ai and OpenRouter); there is no local or self-hosted fallback.

Verdict A solid study for developers building media workflows on Cloudflare or evaluating TanStack Start at the edge. Not yet the right choice if you need fully self-hosted AI or immediate team collaboration.