AI in Physical Production

October 9, 2025

AI in Physical Production

AI in Physical Production: What’s Actually Happening On Set

I attended a PGA Physical Production Committee discussion yesterday about AI tools in production. The conversation was way more grounded than the panic I’ve been seeing online.

The panel had some serious credibility. Leland Krane (DP with 100+ credits), Alex DiGerlando (Emmy-nominated production designer), Jacquelyn Ryan (Disney’s manager of production innovation and one of the committee co-chairs), and Deirdra Govan (costume designer and VP of Local 829). These aren’t tech bros hyping vaporware - they’re people actually using this stuff day-to-day.

The Stigma is Real But Fading

Everyone agreed there’s still a stigma around admitting you use AI tools. The higher up in production you are, the less likely you’ll cop to it. But here’s the thing - that’s changing fast.

Jackie from Disney said studios are getting more interested in experimentation. They want to see people using innovative resources, but they’re being careful about it. Every use case at Disney goes through a panel of subject matter experts to make sure it’s not eliminating union positions or violating ethical guidelines.

Alex made a great point though - literally everyone uses AI in their normal life now. His mother-in-law uses it. The initial paranoia from the strikes is fading because people are understanding what these tools actually do.

What AI Actually Does (And Doesn’t Do)

This was the most important part. AI tools in production are mostly workflow improvements, not creative replacements.

Midjourney makes pretty images, but you can’t build a set from them. The physics don’t work. Alex said working with a concept illustrator who uses AI gets better results than fishing through Midjourney prompts yourself.

Leland pointed out the distinction between generative AI and process-oriented AI. The process stuff is what’s actually useful - pre-production planning, shot lists, research, organizing footage.

Deirdra emphasized this hard - it’s a tool that helps visualize and be more efficient. The problem is when people try to use it to eliminate roles and save money. That’s where the issues start.

The Tools People Are Actually Using

Here’s what the panelists mentioned they’re actually working with:

Image/Video Generation:

Workflow/Process:

Costume/Design:

The Studio Approval Process

Jackie broke down how Disney handles this - every single AI use case gets reviewed. They check if it adheres to responsible and ethical uses, if it’s human-centric, if it eliminates jobs.

Netflix is public about their approval process. Most studios aren’t. But they all have guardrails.

Independent productions? Way more flexible. Jackie uses tools like Rivet AI on her indie projects without needing panel approval. But she still maintains human oversight on everything.

Where It’s Actually Saving Time

Alex told a great story about using Unreal Engine to create a virtual model of a set while they were still building it. The director could do shot listing before the physical space was ready. A year later, they can now turn a photo of a sofa into a 3D model in seconds and drop it into the virtual set to see if it works.

That used to require collage boards and imagination and hoping it worked when you finally got it on set.

Deirdra uses AI for costume visualization when actors won’t come in until two days before shooting. Pre-fits without the actor. Test different looks before spending money.

Leland uses it as a “spellcheck for ideas” - bouncing concepts off ChatGPT to get another perspective. Tone-checking emails before sending them.

Jackie uses OpenAI to help craft better prompts for other AI tools - using AI to use AI better. That’s actually smart.

What Hasn’t Changed

This was important - none of the panelists have seen crew sizes shrink because of AI.

Leland said camera departments are still an operator per camera, AC per camera, second AC, loaders. Nothing has changed on set day-to-day.

Alex hasn’t reduced his crew. He doesn’t trust AI to do construction drawings without human verification. A measurement could be off and you’re screwed.

The pressure to work faster is coming from tightening budgets and schedules, not from AI expectations.

The Trust Issue

Can you trust AI? Everyone said the same thing - no, not by itself.

It’s like Wikipedia. Good starting point, but you have to verify. Jackie cross-references between ChatGPT and Claude. Alex said getting something 80-90% right is better than 100% right if it takes away your creative agency.

Leland made a great point about low stakes versus high stakes. Using AI to generate lookbook images to communicate with a director? Low stakes, works great. Using it for final pixel? Different story.

The Learning Curve

Deirdra’s advice was simple - just jump in and start using it. Most tools have tutorials. You learn by doing.

Jackie suggested starting with entry-level stuff like Microsoft Co-Pilot. Once you learn one tool, the others work similarly.

The stakes are low for most production uses. You’re not letting AI diagnose cancer or drive a car. You’re using it to organize your inbox or generate a concept image.

What They Want Next

Leland wants photometric planning tools - compare tungsten versus LED lights, see power consumption, figure out what’s actually more efficient.

Alex wants better integration between production design and VFX. Bring VFX into the art department earlier instead of rolling the dice in post months later. AI could generate temp elements on set so you can see what you’re actually framing for.

Deirdra wants efficiency tools for sustainability - virtual fittings, pre-visualization that saves time and money without compromising the creative process.

The Big Picture

The most interesting insight came from Alex - AI is like computers. We don’t have conversations about whether computers will disrupt filmmaking. They did. Some changes were good, some were bad. Photoshop changed everything. Movie Magic scheduling changed everything.

AI is just the next iteration of that. It’s not one monolithic thing - it’s hundreds of different tools that do different things.

Leland said the tools made by filmmakers work better than tools made by technologists. Artemis, Shot Deck, Shotcraft - these came from people who had pain points and solved them.

The improvements are leveling off now. We had a fever of new activity when ChatGPT launched, but it’s settling into how people actually use it. The tools are hitting their physical limits. Progress will be more incremental now.

My Take

What struck me listening to these folks was how practical and unglamorous the actual uses are. Nobody’s replacing cinematographers or production designers. They’re using AI to organize their inbox, search through footage faster, test furniture placement before ordering it, do research more efficiently.

The stigma exists because people don’t understand what these tools do. Once you get past the hype and the fear, it’s just… workflow improvements. Helpful, but not revolutionary. Not job-killing.

The ethical concerns are real - training data, copyright, job elimination through cost-cutting. But the people actually using these tools in production are being thoughtful about it. Studios are vetting everything. Below-the-line folks are being transparent about their use.

Young people aren’t intimidated by this. The established professionals are the ones resisting. That tracks with everything I’ve seen.

The future isn’t AI replacing creatives. It’s creatives using AI to work more efficiently, communicate better, and spend less time on tedious shit so they can focus on the actual creative work.

That seems… fine? Maybe even good?

The key is making sure these tools serve the people doing the work, not the people cutting the budgets. That’s where the vigilance needs to be.