Ellenor Argyropoulos Builds AI Workflows for Studios, and Says the Hard Part Is the Human Effort

Ellenor Argyropoulos says she used to get death threats over her AI filmmaking. She says the reaction has flipped in the space of about a year and a half, and now studios and well-known directors are the ones who bring her in when they want to use AI.

Much of the job, she says, is talking clients out of the idea that a prompt does the work. "I don't think I've actually ever worked harder than I ever have with AI," she told us.

In this episode of Inside the AI Studio, we talked with Ellenor Argyropoulos, a director who builds project-specific AI workflows for studios and directors across film and advertising, about why hybrid pipelines and close human review beat the one-prompt fantasy.

• AI filmmaking runs like a normal production. Argyropoulos says her AI projects still move through storyboarding, shot-level generations, selects, organizing, and editing, the same linear stages as any shoot.
• Studios rarely ask for a fully AI look. She says requests almost always land on hybrid workflows that keep live-action footage and pair AI with VFX.
• The ask is more creativity at the same budget. By her account, clients bring her in to expand what a fixed budget can do, not to spend less.
• Real-time backdrop swaps hold up on simple shots. She says teams can iterate a generated background on an LED wall for one or two clean shots, but scenes needing continuity get hard fast.
• She uses language models as a creative collaborator. Argyropoulos works through ideas, prompts, and failed outputs with tools like Claude and ChatGPT, including custom GPTs built for specific apps.

Argyropoulos came up directing commercials and wanted to move into features, she says. The strike pushed her toward AI work, and she now directs while advising production teams, from small groups to large ones, on using AI ethically and building workflows tailored to each project.

Studios hire her to get more creativity from the same budget

She says client conversations tend to split into two kinds. Some come in with a production problem to fix; others come in with a fixed budget and a push for more ambition.

Argyropoulos says the budget framing matters more than cost-cutting:

It can either go two ways. We've got this, can we fix it with AI? Or, look, this is our budget, we want to be as creative as possible. We're not doing this to lower a budget, we're trying to keep the same budget, but we want to have maximum creativity.

These conversations commonly land on hybrid AI and VFX workflows, she says. On bigger productions, Argyropoulos says she is often added specifically to bridge departments that do not always talk to each other.

Previs turns an art-department drawing into a playable 3D scene within a day

Argyropoulos says she usually joins productions very early, working in previs with DPs, art departments, and VFX before any move into post. She also helps filmmakers package existing assets into trailers and decks to raise financing for projects that might not get made otherwise.

The previs stage has pulled art departments and VFX closer together, she says. A drawing can become a rough 3D model dropped into a scene fast, sometimes within a day, so a director can see a version of it before principal photography.

That lets teams test camera angles, blocking, and whether shots cut together before the money goes out. She says it also surfaces problems early, like working out how a car should move through a scene from one angle versus another, in what she calls a very low-stakes setting.

Her hybrid workflow keeps the live-action camera move and layers AI over a rough 3D background

Few clients ask for an entirely AI-generated result, Argyropoulos says. Her preferred approach sits alongside VFX, close to the hybrid pipelines other AI-focused teams are building.

She points to a live-action green-screen shot where the team kept the camera move but needed a complicated army background. Argyropoulos walks through how she rebuilt it:

So I had the 3D team give me the model of what that would look like, and then I would take that very basic model of the demo background and texture it and add all the layers in, figure out which direction and lighting would marry up, and then use that to style transfer it with the actual footage so that you can keep the mocap and the camera move.

She also looks for places where AI can take a transition or roto pass that would otherwise eat a VFX artist's time, freeing them for more creative work.

Iterating a generated backdrop on an LED wall works for simple shots and gets hard on busy ones

She describes working with an AI background upscaled to 7K on a large LED wall, with actors in front. The scale exposed defects a smaller monitor hid, and a director could call for a new version, review it with the team, and put it back on the wall quickly.

Traditional virtual production renders a 3D environment in a game engine such as Unreal Engine and keeps it consistent as the camera moves. Argyropoulos says generated backgrounds are different, and the limits show up with complexity:

I would stay away from very big, busy, complicated scenes because anything busy, AI has more anomalies and there's more things for you to look for under a microscope. So if it's relatively simple, if it's just I'm going to change the beach that I'm on to this beach or this time of day instead, fine, no problem. But the second you have elements that need consistency, then that's going to take you some time and you want to be really careful of that.

Both gray-box shoots and traditional sets will keep their place, she says, depending on the project, the budget, and what a filmmaker wants.

She treats language models like Claude and ChatGPT as a creative collaborator

Argyropoulos says she leans on language models throughout a project, treating the exchange as a working partnership:

AI is my creative collaborator. It's fun. It's like an extension of you that's helping you get to where you need to go.

When an idea hits, she says she will test it with Claude or ChatGPT, feed in reference images, and work out how to talk to a given model. When an output fails, she shows the model where it broke, works through why, and tries again. She has also built custom GPTs tailored to specific apps.

The "magic button" idea ignores the production work behind an AI film

The misconception she runs into most is that AI erases jobs and that a single prompt spits out a finished film. On the jobs point, Argyropoulos pushes back:

The main misconception is that it's taking away jobs. And I think people need to understand that it's actually evolving jobs. And I think that's the biggest difference is to, again, to your rotoscope artist who doesn't necessarily want to be doing that, now they can be more creative and use it for other things.

She adds that AI gets "lumped under this one giant banner," so people miss that language models, video models, and image models are different things.

She points to a four-minute film she made while testing Google's Veo 3, which drew questions about whether one enormous prompt produced the whole thing. It did not. There were rushes for every shot, storyboarding, categorizing, and editing, the same parts of any production. "It's actually a lot of human effort that's going into that," she says, a point that runs against research on audience bias against AI involvement.

For studios weighing where AI fits, Argyropoulos's account points toward hybrid pipelines, early previs, and close human review rather than wholesale replacement, an argument that lines up with how platform builders describe adoption across the industry.

The full conversation with Argyropoulos is in this episode of Inside the AI Studio, shot at AI on the Lot.