How We Built an AI Video System That Costs $0 Per Video

Most AI video tools charge per render. HeyGen bills you per avatar minute. Runway charges credits for every generation. Midjourney costs money every time you need an image. If you are running a studio and producing content at volume, those costs stack up fast. We sat with that problem for a while before deciding to build something different. The result is Vidz, the video AI inside Balay ni Bruno & Co., and it runs entirely on local hardware with zero per-render fees.

The Stack We Built

Vidz runs on four tools, all installed directly on the machine. Remotion handles the core video creation. It is a React-based renderer that lets you write JavaScript code describing what a video should look like, then compiles that code into a broadcast-quality MP4. There is no drag-and-drop timeline. There is no subscription dashboard. You write components the same way you would write a web page, and Remotion renders each frame as a still image before stitching everything together. That means full control over typography, animations, layout, and timing, all in code. FFmpeg handles the editing and compositing work. It trims clips, merges segments, applies color correction, and handles the final export. Auto-Editor scans any recorded footage and removes silence automatically. You record, run Auto-Editor, and get back a tight cut without touching a timeline. Whisper handles transcription locally, converting spoken audio to text that can be used for captions or burn-in subtitles.

Every single one of those tools runs on our machine. None of them phone home to a cloud API during a render. That is the core of why the per-render cost is zero.

How a Video Actually Gets Made

We built a three-segment reel workflow that covers most of the content Balay ni Bruno & Co. produces. The video opens with a talking head segment, a direct-to-camera intro recorded by a team member. Auto-Editor cleans the silence out of that recording before anything else happens. The middle segment is a split-screen teaching section, where the talking head stays in a smaller frame on one side while screen recordings, graphics, or motion text play on the other side. Remotion generates those programmatic graphics from code, so changing a number in a script updates the entire visual rather than requiring someone to manually redesign a slide. The video closes with another talking head segment, a short direct-camera close. FFmpeg stitches all three segments together into one export. The whole pipeline, from raw recording to finished MP4, runs locally and produces a file ready for any platform.

Why This Matters for the Business

We produce content consistently. Consistency is what builds an audience and what demonstrates capability to clients. If every piece of content carries a variable cost tied to a third-party subscription, production decisions get distorted. You start thinking about whether a video is worth making based on what it costs to render instead of whether it serves the audience. A local stack removes that distortion entirely. The only cost is the machine it runs on, which we already own. When Balay ni Bruno & Co. builds an AI video system for a client, the same logic applies. The client is not paying a monthly subscription to a video platform that could change its pricing or shut down. They are running a system they own, on infrastructure they control. That is a fundamentally different kind of stability than renting access to someone else's cloud.

We are not saying cloud tools have no place. HeyGen is genuinely good for avatar video. Runway produces impressive motion. There are things those platforms do that a local stack does not cover. But for the core workflow, recording, editing, captioning, compositing, and rendering polished MP4 content at volume, the local approach wins on cost, on speed, and on control. Vidz is proof that building the right system upfront is almost always better than renting a convenience you will pay for indefinitely.