The critics absolutely savaged it. When Darren Aronofsky’s AI studio Primordial Soup dropped the first episodes of “On This Day… 1776” last week, the internet basically erupted with complaints about waxen faces, repetitive camera moves, and what multiple outlets branded as “AI slop.” The Guardian didn’t pull punches, calling the series from the once-lauded Black Swan director “embarrassing” and “ugly as sin.”
But here’s the thing nobody expected. A production source just revealed to Ars Technica that creating these supposedly quick-and-dirty AI videos actually takes weeks for each minutes-long episode. Weeks. For something critics assumed was pumped out by typing a prompt into a computer and hitting enter.
The Reality Behind AI Filmmaking
The year-long series features photorealistic scenes of George Washington, Benjamin Franklin, and Thomas Paine doing Revolutionary War things 250 years ago. Time magazine’s President Ben Bitonti pitched it as “a glimpse at what thoughtful, creative, artist-led use of AI can look like,” but the production reality sounds more like filmmaking purgatory.
According to the anonymous source, the AI video generator rarely nails it on the first try. Or the 12th. Or sometimes even the 40th. “You don’t know if you’re gonna get what you want on the first take or the 12th take or the 40th take,” they admitted. It’s weirdly similar to live action filmmaking, except instead of repositioning lights or adjusting an actor’s blocking, you’re feeding slightly tweaked instructions back into an algorithm and praying.
The team storyboards everything, finds visual references, sets up shots exactly how they want them, feeds all that into the AI model, then waits to see what comes out. Most of the time it’s not quite right. Maybe the lighting doesn’t land on a face properly. Maybe some weird hallucination creeps in. The Technology is impressive but far from magical.
Why Everything Still Needs Human Touch
Here’s what actually is human in this supposedly AI-powered project: the script (written by Aronofsky’s longtime collaborators), all the voice acting (recorded by SAG actors), the music, editing, sound mixing, visual effects, and color correction. The AI only generates the raw video shots, which then get stitched together and cleaned up by human editors.
The production source couldn’t even imagine replacing human editors with AI. “We actually desperately need an editor,” they said. Someone who knows when to cut out of a shot early to create urgency, when to linger for emotional impact. The kind of instinctive storytelling decisions that algorithms don’t understand.
They tested AI-generated voices for temp tracks and found them noticeably artificial. Not ready for professional Business production at all. Which raises an obvious question: if the voices aren’t good enough, why are the faces and bodies?
The Experiment Nobody Asked For
The team is billing this as a massive experiment, fully expecting to make mistakes and learn as they go. The source admitted they’re “more often than not, pushing deadlines” despite the theoretical time savings of not hiring actors or scouting locations. Yes, it’s cheaper than filming a historical docudrama on location. But it’s definitely not fast.
They chose short-form videos specifically because maintaining consistency gets exponentially harder with length. “It’s one thing to stay consistent within three minutes. It’s a lot harder and it takes a lot more work to stay consistent within two hours,” the source explained. Hallucinations and nonsensical images remain “still a problem.”
Even keeping individual shots short helps with control. A 20-second clip has way more opportunities for something to go wrong than an eight-second one. When something breaks in that 20-second generation, you start completely over.
The most telling moment came when the source was asked why they thought AI was ready to replace human actors specifically. Their answer? “I don’t know that we do know that, honestly.” They just want to play with new tools because new tools don’t come around often. We have to try things to know if they work.
Which is fine as an artistic experiment, but it’s hard not to notice they chose to experiment with the one part of filmmaking that employs the most vulnerable workers while keeping the prestigious writing and directing roles firmly in human hands. The news here isn’t that AI can make movies; it’s that even when it tries, it still needs an army of humans and weeks of iteration to produce something watchable.
