Why 'house Of David' Director Thinks Ai Can Save Hollywood Jobs

In 1926, head Cecil B. DeMille hired hundreds of workers to build a group of Jerusalem wrong nan DeMille Studios successful Culver City for nan classical silent movie “The King of Kings.”

A period later, Jon Erwin filmed his biblical epic ‘The Old Stories: Moses,’ starring Ben Kingsley, connected nan aforesaid workplace batch now owned by Amazon MGM Studios.

Except now, overmuch of nan architecture, godforsaken location, and supernatural parts of nan three-episode miniseries were generated done artificial intelligence. The prequel to ‘The House of David’ bid debuts connected Amazon Prime connected Thursday.

A accumulation that traditionally would person taken months to sprout and require aggregate locations was filmed wholly successful 1 week pinch a unit of conscionable 100 group — who ne'er near Los Angeles.

“We did this monolithic sword-and-sandal epic, and we ne'er near a soundstage, very akin to really James Cameron does Avatar aliases really Jon Favreau does ‘The Mandalorian,’” said Erwin, nan head of nan series. “When you sphere nan capacity and nan activity of nan crews and nan section heads, past you tin do things that are incredibly cost-effective for studios.”

As Hollywood grapples pinch accelerated technological change, a increasing number of filmmakers and companies successful Southern California are utilizing AI devices to radically rethink really films and TV shows are made.

“Some are still resisting, but galore are recognizing that, for amended aliases worse, AI is present and not going anyplace and it is important to reimagine what movie creation tin look for illustration successful ray of nan caller possibilities AI creates,” said Victoria Schwartz, head of nan entertainment, media, and sports rule programme astatine Pepperdine Caruso School of Law.

A surface of LED panels called “the Volume” is utilized to movie scenes for head Jon Erwin’s bid “The Old Stories: Moses.”

(Genaro Molina / Los Angeles Times)

Erwin is among nan first moving board astatine a awesome streaming level to afloat merge AI into a commercialized production.

Last month, he launched Innovative Dream, a Manhattan Beach accumulation services institution backed by Amazon. The institution will rent its virtual accumulation accommodation to different studios and create training programs for emerging filmmakers.

Although overmuch of Hollywood is bracing for AI to hollow out jobs, Erwin argues nan opposite: that AI, applied ethically astir quality performances, tin return astatine slightest immoderate accumulation jobs that person been outsourced moreover arsenic different positions are eliminated.

“I deliberation nan greater threat of occupation nonaccomplishment successful our manufacture is really conscionable really costly things person gotten and really agelong they return to make,” Erwin said. “If you tin make things quicker, and you tin make things astatine a value constituent that studios will opportunity ‘yes,’ you tin employment much group successful aggregate and create jobs.”

Although machine graphics person been basal to Hollywood since nan 1990s, they traditionally required hundreds of artists and months of post-production activity to spot actors aliases crowds successful integer worlds. Much of nan labor-intensive ocular effects activity known arsenic rotoscoping was outsourced to shops successful India and different countries pinch overmuch little labour costs than successful California.

By 2019, productions specified arsenic Disney’s “The Mandalorian” bid precocious this further by utilizing massive LED screens to task images of photorealistic integer worlds — “Star Wars” ships, forests, aliases deserts — arsenic actors’ performed successful costume successful beforehand of them. A virtual creation section spent months designing nan integer environments, and past loading them onto nan ample screen connected nan time of nan shoot.

AI takes nan process a measurement further.

Through “Moses,” Erwin is championing what he calls “hybrid” filmmaking: a workflow that marries live-action pinch AI-enhanced workflows successful virtual production. The process combines what utilized to beryllium abstracted phases — filming pinch actors and ocular effects — to hap almost simultaneously. Scenes changeable connected group is made disposable to aggregate editors and AI artists wrong minutes connected nan accumulation floor, arsenic they show near-finished sequences backmost to nan formed and director.

“You tin create assets successful 3 aliases 4 days, not 10 weeks. And that intends you tin really benignant of make nan situation while you’re shooting,” he said.

Erwin, 43, grew up successful Alabama and built his profession astir faith-based films specified arsenic ‘I Still Believe’ and ‘Jesus Revolution.’ He had spent years trying to show biblical stories astatine nan standard portrayed successful nan root material.

When he sounded “House of David,” a play astir nan life of King David, workplace executives were initially skeptical. “I was told to conscionable travel up pinch a smaller idea,” he said.

To represent Goliath’s root story, actors were filmed connected greenish screens and AI was utilized to make a mythical series involving acheronian sky, rain, mountains and angels pinch wings.

It marked 1 of nan first integrations of generative AI successful a awesome commercialized production. The series, which premiered past twelvemonth was viewed by 44 cardinal viewers worldwide and reached No. 1 connected Prime Video successful nan U.S.

By Season 2, nan squad utilized 30 different tools, some accepted and AI, to make images, sounds and video. They pivoted from shooting solely connected location successful Greece to filming immoderate parts successful L. A. successful beforehand of an LED wall.

AI was utilized to make battle scenes and grow nan inheritance crowd size to thousands of group successful a fraction of nan clip accepted CGI required. The usage of AI-generated scenes jumped from 70 successful Season 1 to 400 shots successful nan 2nd season.

Jeff Thomas, a generative AI filmmaker who directed 2 episodes of Season 2, said each section was made for little than $5 million, defying workplace statement that nan show required a “Game of Thrones”-level fund of $12 cardinal to $15 cardinal per episode. Erwin declined to disclose nan budgets for nan “House of David” bid aliases nan “Moses” prequel..

“The Bible describes that conflict arsenic location was 100,000 group connected each side. Well, it’s ne'er been portrayed for illustration that because we’ve ne'er had nan resources,” Erwin said. “We’re yet capable to show that scope and scale.”

Erwin conceived of nan thought of “Moses” complete Christmas, wrote nan book successful January and created a four-minute trailer wholly created by AI. Amazon greenlighted nan bid later that month.

Kingsley had a short model earlier his adjacent commitment, truthful Erwin prepared and changeable each 3 episodes connected a soundstage successful a week — a task that would person antecedently taken six months to prepare.

For nan pivotal Red Sea scene, Erwin generated nan h2o volumes and tidal waves successful little than hr utilizing AI models from Chinese institution Kling AI and Palo Alto-based Luma AI, which would person taken weeks successful nan accepted process. They wrote matter prompts that explored 18 different variations of nan oversea parting and discarded nan ones that didn’t work, enabling Kingsley to respond to a tidal activity projected onto a 360-degree LED wall screen.

“‘Moses’ really represented a full caller method of filmmaking for me,” Erwin said.

For “The Old Stories: Moses,” head Jon Erwin utilized AI for wide shots, stunt-heavy conflict sequences and to make ample crowds to showcase nan expansive scope of biblical stories. The reddish statement he said he wouldn’t transverse is utilizing it successful spot of actors.

(Genaro Molina / Los Angeles Times)

For important scenes portraying nan palace hallway successful Egypt, wherever Moses talks to nan Pharaoh, they built cardboard boxes arsenic nan columns successful nan palace, and “reskinned” them pinch intricate carvings utilizing AI. Although nan group could accommodate only 20 extras, they utilized AI to create hundreds of inheritance actors.

Erwin besides utilized generative AI to synthetically grow partially built sets featuring soil and rocks and to “de-age” Kingsely to look arsenic a young Moses.

But immoderate things were disconnected limits for AI, including Kingsley’s performance.

“I conscionable deliberation our faces are truthful intricate and nan micro expressions are truthful intricate, truthful that’s ever real,” he said.

Instead, AI was utilized to co-design nan character: Erwin primitively imagined a bald Moses, but based connected Kingsley’s feedback, they fine-tuned nan look pinch weathered hairsbreadth and mustache.

“The statement successful nan soil for maine is replacing an actor,” Erwin said. “I don’t want to beryllium successful nan manufacture if I can’t activity pinch actors.”

Jon Erwin’s “hybrid” accumulation involves generating a assortment of environments specified arsenic forests, deserts, aliases conflict sequences utilizing AI, and projecting them connected nan LED screen.

(Genaro Molina / Los Angeles Times)

When asked astir nan inheritance extras displaced by AI crowd generation, Erwin said that’s nan incorrect measurement to deliberation astir it.

“It’s not a comparison of what would “Moses” person costs otherwise. It’s a comparison of “Moses” would person ne'er been made otherwise, and that’s nan measurement you person to deliberation astir it,” he said.

Overall contraction successful Hollywood has led to less films being changeable connected location successful Los Angeles, and a 30% drop successful intermezo manufacture jobs since its 2022 peak.

“I deliberation you tin do those things 3 to 5 times faster, astatine little than 30% nan cost,” he said. “I really spot this instrumentality group arsenic an antidote to nan occupation nonaccomplishment problem successful our industry.”