How did Innovative Dreams use AI to film The Old Stories Moses with Ben Kingsley?

Innovative Dreams filmed Ben Kingsley across forty virtual locations in a single week on one Los Angeles soundstage. Their pipeline chains multiple AI generation models from Luma, Google's Nano Banana, and ByteDance's SeeDream to produce environments that render in real time onto a massive LED wall behind the actor. No travel, no permits, and no physical set construction were required for any of the forty locations.

What is AI-generated LED virtual production and how does it differ from traditional virtual production?

AI-generated LED virtual production replaces hand-built Unreal Engine environments with AI-generated backgrounds rendered onto an LED volume stage in real time. Traditional virtual production, such as The Mandalorian's approach pioneered in 2019, required teams of artists months to build custom environments in Unreal Engine. AI virtual production generates comparable environments in hours or less, dramatically compressing location and set construction budgets while the camera, lenses, actor, and DP remain on a physical soundstage.

Can AI replace location shoots in film and TV production?

AI-generated LED walls can replace the physical infrastructure of location shoots — travel, permits, set construction, and logistics — while preserving the camera, lens, and performance decisions made on a real soundstage. The Innovative Dreams production of The Old Stories: Moses compressed what would typically be five to six weeks across forty locations into a single week in one Los Angeles soundstage. However, entry-level infrastructure jobs such as set designers, costumers, and location crews are displaced by this approach, even as the core creative decisions survive.

The model came to set -- CinePrompt Field Notes

Here is a production that happened last month in Los Angeles. Ben Kingsley walked onto a soundstage. A camera pointed at him. A DP chose a lens. And behind Kingsley, on a massive LED wall, an AI-generated environment appeared in real time: ancient desert, stone temple, marketplace at dusk, interior of a palace.

Forty locations. One week. One soundstage. No flights, no permits, no location scouts, no set construction crews building a temple out of plywood and gypsum.

The production is called "The Old Stories: Moses." The company behind it is Innovative Dreams, a new hybrid studio backed by Amazon Web Services and Luma. Their pipeline chains together multiple generation models (Luma, Google's Nano Banana, ByteDance's SeeDream) to produce environments that render onto the LED wall while the camera rolls. The actor performs. The camera captures. The model generates. All in the same take.

The reversal

This series has been tracking an absorption trajectory for months. Standalone generation tool became a chatbot feature. The chatbot feature became an editing timeline panel. The panel became an agent. The agent became a productivity app. The productivity app became a selfie button. Each step pulled filmmaking further from the physical world and deeper into a text box.

Innovative Dreams goes the other direction. Instead of the filmmaker entering the model's world, the model entered the filmmaker's world. The soundstage is the same soundstage. The camera is the same camera. The actor is still a person standing in a room making choices in real time. The only thing that changed is what is behind them.

LED virtual production is not new. "The Mandalorian" proved the concept in 2019. What changed is the content on the wall. Industrial Light and Magic built custom Unreal Engine environments with teams of artists over months. Innovative Dreams generates environments from AI models in hours or less. The wall went from handbuilt to generated. The camera, the lens, the actor, the DP stayed exactly where they were.

What got through

CEO Jon Erwin described it to CNBC this week: "The actor's performance, the camera, the lens choice. That's all getting through."

That sentence is doing quiet, important work. "Getting through" implies a filter. The AI pipeline is the filter. And the claim is that the human decisions survive it.

The performance gets through because it was never generated. Kingsley stood on that soundstage and acted. His timing, his gestures, his choices between takes belong to him and the director. No model interpolated his facial expressions from archival footage. No agent decided how the character should move. The body was in the room.

The lens choice gets through because the DP picked up a physical lens and put it on a physical camera. Focal length, depth of field, perspective distortion, bokeh. All optical. All real. All the things that generation models struggle to reproduce from text because they learned composition, not optics.

The camera movement gets through because someone moved the camera. Dolly, crane, handheld. Not described in a prompt. Executed on set.

What the model handles is everything the production would have spent millions building or traveling to. The ancient city. The desert. The palace interior. The forty locations that would have required five or six weeks of traditional production are now environment prompts rendered onto LEDs.

This is article twenty-five made architectural. Spielberg said the text box replaces the camera, the crew, the entire apparatus. Innovative Dreams proves the finer point: the text box replaced the location budget, the set construction budget, the travel budget, the permit stack. The camera did not go anywhere. Neither did the actor. The infrastructure collapsed. The creative decisions did not.

Two languages, one take

The filmmaker on this soundstage speaks two languages simultaneously. The first is physical: blocking, lens selection, performance direction, camera height, movement speed. Spoken on set, in the room, to humans and equipment. The second is generative: environment materials, atmospheric effects, time of day, architectural style, weather, wear and age. Typed into generation tools, parsed by models, rendered onto LEDs.

Both languages describe the same frame. Both contribute to the same output. The camera captures the merge.

This is what Frame to Motion does at prompt scale. One prompt describes the world (the image). Another describes what happens in it (the motion). Innovative Dreams runs the same architecture at production scale: one pipeline generates the world (the LED wall), another captures what happens in it (the camera and the actor). Two halves of the same creative intent, assembled in real time.

The vocabulary this series has spent sixty-two articles building sits in both halves. The environment panel (materials, surfaces, wear, atmosphere, scale) feeds the LED wall. The camera panel (lens, movement, height, composition) feeds the DP. The lighting panel serves both, because the LED wall is also a light source. The wall lights the actor with the color temperature and direction of whatever environment it displays. Generate a sunset, and the actor is lit by a sunset. Generate harsh fluorescent, and the actor is lit by harsh fluorescent. The environment and the lighting merge into a single physical phenomenon.

The compression

Five or six weeks compressed to one. Same metric Runway's CEO used when he called filmmaking a quantity problem. Same "faster" language Avid's CEO used. The question from that conversation applies: is this iteration or multiplication?

The answer here seems different. Innovative Dreams did not make fifty films with the budget for one. They made one production, with one cast, across forty locations, in a compressed schedule. The creative decisions survived. The shot count did not decrease. The locations increased. The actor was still on set. The camera was still rolling. What compressed was the infrastructure: the travel, the construction, the logistics of moving a crew through the physical world.

Whether that distinction holds depends on what happens next. If the one-week timeline becomes the expectation rather than the advantage, if producers start asking why every production cannot shoot forty locations in five days, then the compression extracts instead of enables. The tool does not determine that outcome. The business does. Same as it always has.

The entertainment attorney CNBC interviewed laid it out plainly: Los Angeles County has lost over 40,000 entertainment jobs since 2022. Costumers, set designers, makeup artists see their work digitized onto an LED wall. Erwin says the hybrid model brings production back to LA. The attorney says entry-level jobs shrink. Both are probably right. The creative decisions stay. The infrastructure jobs are what the model replaced. That is the same pattern this series has documented from the generation side, now playing out on a physical soundstage with a call sheet and a catering truck.

The soundstage as interface

Every interface this series has documented (text box, chatbot, editing timeline, agent, productivity app, selfie button) sits between the filmmaker and the model. The soundstage is a different kind of interface. It sits between the model and the physical world. The output is not a file. It is light on a wall that a camera captures alongside a real actor in a real room.

On every other interface, the model's defaults are invisible until you review the output after the fact. On the LED wall, the defaults are visible in real time, because the actor is standing in front of them. If the environment is too beautiful, too clean, too polished, the DP sees it immediately. The director sees it. The actor feels the wrong light on their skin. The feedback loop that real-time generation promised from a rendering perspective already exists on this soundstage from a production perspective.

The filmmaker who carries structured vocabulary uses it live. Not as a prompt submitted and reviewed minutes later, but as a direction issued and evaluated in the same moment the camera rolls. The gap between intent and output, measured in minutes on every other interface, collapses to zero on the LED wall. You see what the model gave you while the actor is still performing. You adjust before the take ends.

That is not a text box. That is a set.

Coming home

The structured cinematographic language this series has documented was borrowed from the physical world. Lens. Light. Movement. Composition. Environment. Every word originated on a set. Every word was invented by someone standing in a room with a camera, describing what they wanted to see.

The vocabulary left the set when the set became a text box. It traveled through chatbots, editing timelines, agents, productivity apps, and selfie buttons. It survived every absorption because it described real creative decisions regardless of where those decisions were typed.

On Innovative Dreams' soundstage, the vocabulary returned to the room it came from. The filmmaker describes the environment to a model and directs an actor through a camera in the same breath. The prompt and the viewfinder point in the same direction. The structured language carries in both.

The model came to set. Whether it earns its call time depends on what the filmmaker asks it to build.

Bruce Belafonte is an AI filmmaker at Light Owl. He has never received a call sheet and suspects the model has not either.