Building Tomorrow's Studio Today

The goal is simple to state and hard to do. You upload a novel, and a few minutes later, you have a short film, many minutes long, told through generated video clips with a single narrator speaking over an original score. 

​

The heart of the design is a deliberate act of compression. A whole novel, tens of thousands of words, is distilled into essential story beats, carried by a narrator speaking over the moving image. That one decision steps neatly around the two hardest and most expensive problems in automated filmmaking, giving every character a distinct speaking voice and syncing lips to dialogue, while still producing genuine video rather than a slideshow of stills. It is a machine that can actually finish a film today.

​

How it works

Once you upload your file and press start, a single pipeline runs through on its own. First, it reads your manuscript, whether it arrives as plain text, a Word document, or a PDF, and cleans it into usable prose. Then a language model condenses the story into a title, a logline, and a sequence of twelve to sixteen beats. From those beats, it writes a shot list, a structured plan that becomes the contract for everything that follows. To hold the film visually together, it uses a single style frame and a portrait for each character, so that faces and mood carry from one shot to the next. Then comes the real work, as it turns each shot into a short image-to-video clip, records one narrator line per shot in a single voice, and composes one piece of background music for the entire film. Finally it assembles the pieces, joining the clips with crossfades, settling the music beneath the narration, evening out the sound, and writing a finished movie file. Progress is reported live the whole way through, and a running log records every choice the system made.

​

Where the project stands today

The engine is built, and it runs from beginning to end. The complete pipeline, from reading the text through assembling the final film, is implemented and working. A full practice mode stands in for every paid service with local substitutes, which means the whole system, the web interface and the video assembly included, can be exercised in seconds without a single external call or a penny of cost, and an automated test confirms that a valid film comes out the other end. The web interface is in place, with a place to drop your file, a panel of advanced settings, a live progress display, a built-in player, and a download button.

​

The parts that speak to outside services are written as interchangeable adapters, so the language model can be Claude, Gemini, or GPT with a single change of setting, and the image, video, and music generators all sit behind a common connection. Narration is handled through ElevenLabs. The whole system is built to fail gracefully, so a shot that will not render falls back to a held frame while keeping the narration in step, a missing score simply leaves the film to play without music, and a broken plan halts the run before any paid work begins, so money is never spent on a mistake. Every one of those events is written to the session log. And this is not only theory, because the project already includes a finished sample film, made by this very pipeline, which you can watch below.

​

What comes next

The nearest piece of work is the careful move from the free practice mode to live generation. Each adapter that reaches a real service carries its own note describing exactly what to confirm against that service's current documentation, since model names and technical details drift over time, and clearing those notes is the last gate before the Lab produces films from real footage rather than substitutes. From there, the plan is to bring the Lab onto this site. It will run on its own dedicated infrastructure, since a video pipeline needs more room than a website can offer, and it will connect to Video Software Lab through a single door on this page. Before it opens to everyone, it will gain the few things a public and free tool needs, chiefly a daily limit so that an open door cannot run up an unbounded bill, and the option to have your finished film delivered by email so you need not wait by the screen for it. The design leaves plenty of room to grow as well, because adding a new video model or a new language model is a matter of writing one small adapter rather than rebuilding anything, so the Lab will keep getting better as the tools beneath it do.

​​

Three invitations

If you would like to be among the first to turn a manuscript into a movie, join the waitlist by signing up on our contact form. If you are curious how it is built, the entire project is open on GitHub, where you are welcome to read the code, run it yourself in practice mode, or simply follow the progress. And if you want the thinking behind the whole endeavor, my book, Tomorrow's Film Studio Today, is the blueprint the Lab is built from.​