Video Machine

Project Overview

Video Machine turns written slides into finished videos. You type a title, pick a theme, shape, and pace, and add slides as headline plus detail lines while a preview rebuilds itself as you type. Point it at a source instead and it drafts the slides for you, either a client-ready SEO audit from a URL or a market recap from a handful of ticker symbols. Renders run in the cloud and land in one to two minutes as either vertical 1080x1920 for social or widescreen 1920x1080, with an email when the file is ready. The platform runs on React Router 7 with Remotion for rendering, Trigger.dev for background jobs, Railway S3 for storage, and Polar as merchant of record for credit purchases.

Knowledge Gained

Shipping this to paying customers taught me more than building it did. Remotion made the rendering tractable but the hard part was everything around it: long-running jobs that outlive a request, retries that do not double-charge a credit, and progress states that keep someone informed while a render takes ninety seconds. Trigger.dev handled orchestration and gave me a place to reason about failure instead of scattering try/catch through route handlers. Wiring credits through Polar webhooks meant treating balance as something that must reconcile, not a number to decrement optimistically, since a failed render has to return the credit it took. Drafting free and charging only on render turned out to be both a pricing decision and an architecture one, because the split forced a clean boundary between the editor and the render pipeline.

The Impact

Video Machine is a product with a pricing page, not a demo. It covers the full path from an idea to money in an account: rendering pipeline, background job orchestration, object storage, transactional email, authentication, a free trial that converts, and a checkout flow through a merchant of record. Owning every layer meant making the tradeoffs myself, including where to spend on reliability and where a simpler answer was good enough. For anyone evaluating whether I can take something from empty repository to running service that strangers pay for, this is the clearest example in the portfolio.

Links