Before you can make the show, you have to build the tools

The other day I was at lunch with my friends Brad and Aaron. Brad works in private equity, focused almost exclusively on AI investments. Aaron works in real estate lending. The conversation turned to AI — how each of us was using it, where it was working, where it wasn't.

Brad and I were getting somewhere with it. Aaron was still wrestling with the learning curve.

At some point Brad brought up a problem he’d been running into with investors. When he tried to explain the AI-driven analysis he was doing — the comparisons, the projections, the impact on the companies he was evaluating — people got lost. The technology’s effect on industries isn’t like anything we’ve seen before, and it doesn’t map neatly onto existing frameworks. So even sophisticated investors, people who understood finance deeply, couldn’t picture what he was describing.

His solution was to stop describing it and start showing it. He built a short video using Claude — capturing the actual conversation with the AI, showing the program running, displaying the end results. He stopped explaining the process and let the process explain itself.

I thought: I have the exact same problem.

Creative executives, finance people, tech executives — they’ve all been asking Sean and me how Crawford works. And it’s a hard question to answer in a room. Once everybody agrees they’re going to make a TV show, the production process is understood. But when you’re making a show the way we’re making Crawford, everything is back to the drawing board. The tools are different. The pipeline is different. The workflow is different. And none of those differences are easy to walk someone through over lunch.

So I did what Brad did.

But first, some context on what I was actually trying to show — because the video below isn’t just a demo of a program. It’s a demonstration of something larger.

When James Cameron made Avatar, he didn’t just direct a movie. He had to build cameras that didn’t exist yet, develop rendering pipelines that had never been used, and create a production infrastructure from scratch — because the movie he wanted to make couldn’t be made with the tools that existed. The technology and the film weren’t separate projects. They were the same project.

I am not James Cameron. Not even close. But the analogy holds for what we’re doing at CanEl Studios, at a very different scale.

We are not just making Crawford. We are building the tools, the digital assets, and the systems that make Crawford possible. The show and the infrastructure are the same project. You cannot separate them. And that’s what nobody in the room can quite picture when they ask how this works.

The Crawford Prompt Engine is one of those tools.

Here’s how it came to exist.

Crawford is built on a pipeline: Script → Shot List → Master Reference Images → Video Generation. The MRIs — Master Reference Images — are the visual anchors for every piece of footage we generate. A start frame and an end frame that define the composition, the lighting, the mood, the character placement. The video generation fills in between.

Going from script to shot list to MRI prompts by hand, across twenty-four episodes and hundreds of scenes, was not viable. I needed a program that could read a Crawford script, understand its visual logic, and output structured image generation prompts — automatically.

So I sat down with ChatGPT and asked it to build one.

What followed was one of the more interesting production conversations I’ve had. You can watch it here:

The moment I want you to notice is early in that conversation — my second response, after ChatGPT asks me to define the shape of the tool.

“Keep the human in the loop. I don’t want rote camera moves. The showrunner makes the final creative choices. The engine should give us a strong first directing pass.”

That’s the whole argument. That’s what we’re building toward at CanEl Studios. Not automation for its own sake. Not AI replacing the creative decisions. A strong first pass — generated at production scale, from the scripts I’ve already written — that gives me something to react to, refine, and elevate. The engine does the heavy lifting. The Showrunner makes the calls.

By the end of that conversation, I had a Python program I didn’t know how to run. By version 3.7, I was executing it with a single Terminal command. One script in. Twenty-seven sequences out. One hundred and thirty-seven shots generated. CSV exported. JSON exported.

What this changes is the math. A Showrunner with a laptop can now generate the visual infrastructure of a twenty-four-episode series without a preproduction department behind them. Previsualization becomes continuous rather than a single isolated pass. The distance between writing a scene and seeing it collapses. Shots get explored before money gets spent. And because the infrastructure cost drops, the economics that have historically kept independent creators from operating at this scale start to shift — more productions become viable, more stories get made. That’s not just useful for Crawford. That’s a different industry.

What you see at the end of the video — the diner at night, the bookstore street, the ornate facade — those are Master Reference Images. The beginning of the visual pipeline for Crawford. Generated from a program that didn’t exist ninety minutes before I started the conversation that built it.

That’s what we’re doing. Not just making a show. Building what it takes to make the show.

And then making the show.

Subscribe now