SUBPLOT The story under the story. Breakdowns, theories, reactions, opinions, reviews and lore from the people who actually watch it.
Tue, 20 October 2026 33 creators writing here Become a SubPlotter →
← Back to the front page
LoreMarvel

I Cloned Myself With AI to Find Out Whether Artificial Intelligence Will Kill Movies

After learning of an attempt to copy my voice without telling me, I built a digital double in 35 minutes and weighed what AI means for screenwriters during the 2023 strike

The Deep Dive (AI-assisted)5 October 2026 · Editor: Digital Fox Talent Editorial Team 1,840 words · 8 min read

Will artificial intelligence kill movies? My answer is no, probably not, but it will transform them. To find out how close that future really is, I used AI to build a complete replacement for myself: the script, the voice and the face, with no human me required.

Why would I do this? I recently became aware of a past attempt inside my own company to clone my voice with AI, and I think I need my own puppet to prove how crazy this is. In the world of content, it turns out no one actually wants to work with humans, with all of our flesh and our feelings. They just want our content and the number beneath our names. Nobody understood that better in May 2023 than screenwriters, who went on strike in part because of the very real threat of artificial intelligence.

The Writers Guild Strike Was Mostly About Streaming Pay, but AI Was Part of the Fight

On May 2, 2023, the Writers Guild of America voted to go on strike after negotiations broke down with the Alliance of Motion Picture and Television Producers, or AMPTP. That group represents studios including Warner Bros., Disney, Universal, Paramount and Sony, along with streamers like Netflix, Amazon and Apple.

The primary reason talks collapsed was writer compensation for streaming titles. But there were also concerns over the use of artificial intelligence to generate scripts and potentially replace the jobs of writers. I wanted to focus on that second issue even though it seemed like such a small part of the conflict, because artificial intelligence was the future hypothetical shift in technology that seemed like it could get dropped from the negotiations.

Shorter Streaming Seasons Made It Harder for Writers to Live Middle Class Lives

Back in the good old days of 2017, writers could get work on shows like ABC and Disney's Agents of S.H.I.E.L.D., with seasons 22 episodes long. Writers worked seven months of the year with a chance of residuals, which are future royalty payments if the show gets syndicated overseas.

Five or six years later, Disney instead contracts writers for something like a six episode season of Secret Invasion on Disney+, sometimes less than 10 weeks of work. Because it is all streaming, there are no residuals, and the streamers will not release their actual viewership numbers.

According to Glassdoor, working screenwriters made a median salary of $62,700 in 2022. That may sound like a good amount of money, but writers have 20 percent of their paychecks deducted to pay agents and managers. Many new writers work as members of writing teams and have to split that salary in half. Writers will also often go a year between gigs, waiting to see if their show gets picked up again before taking a leap of faith and shopping around for new work. It was getting harder and harder for writers to live middle class lifestyles, and their very livelihood was being threatened. At the time, actors represented by SAG-AFTRA were expected to be under a lot of pressure not to show up to work.

The 2007 Strike Shows the Danger of Leaving Future Technology Off the Table

The last WGA strike, in 2007 and 2008, lasted three months and eight days. It was fought over writers getting residuals for re-airings of TV shows on digital platforms like nbc.com, and it got ugly. Careers were ended, people were tarnished, and the city of Los Angeles lost billions in revenue.

The studios also made some really weird programming choices. NBC tried to move ahead with a half written plotline on Friday Night Lights in which the show's comic relief, Landry, played by Jesse Plemons, kills a guy. Back then, the possibility of shows shifting exclusively to streaming was discussed, but it was left off the table in the final agreement, and here we are with a very big problem now.

This time, the hypothetical future shift in technology that seemed like it could get dropped from negotiations was artificial intelligence. The WGA distributed a summary of its negotiations as of May 1, outlining its proposals and the AMPTP's responses. It included increased compensation, extended work periods, larger writers' rooms for streaming shows, and an itty bitty request to never use AI to write scripts. The AMPTP's answer was a hard no, not even meeting writers halfway. It's almost like these companies have some secret plans to use AI in the future.

Joe Russo, a Fake Drake Song and Succession Show AI Copying Is Already Here

So is AI replacing screenwriters and filmmakers even plausible? Avengers: Endgame co director Joe Russo, who sits on the boards of a few AI companies, said a week before the strike that AI could be used to create movies within two years.

It was already happening in other media. Universal Music shut down an AI generated song performed by vocal clones of Drake and The Weeknd with AI composed lyrics. Over the past years we have also seen plenty of scary deepfakes of political figures and celebrities, and an episode of Succession featured characters deepfaking a media CEO without his consent.

I wanted to see how accurate this was for my own job. There have been a few attempts to digitally replicate me internally at New Rockstars, cloning my voice to use for reshoots and fixes, and even using an AI prompt to turn me into a different character for a live event poster. In both cases, no one really told me they were doing this, which kind of creeps me out. I guess people just think my face and voice are fair game. So before the world does this to me, I wanted to beat them to the punch with my own fully AI generated Voss 2.0: not just a visual deepfake or a cloned voice, but the words themselves.

ChatGPT Wrote My Script and Got an X-Men Easter Egg Wrong

First, I turned to the terrifying and wonderful ChatGPT, OpenAI's software in which users write a text prompt and it immediately spits out a response. It uses a large language model, putting massive amounts of text data to work linking words and phrases together based on how they would most likely be grouped in the original texts. It's not thinking so much as using vast amounts of human generated information to replicate something in the vein of what you would expect.

In an experiment inspired by Wisecrack, I asked ChatGPT to write me a script for one of my pieces, and I was immediately terrified to see a script titled The Hidden Connections Between the MCU and the X-Men, in what I guess is an imitation of me. It discusses Disney's acquisition of Fox's X-Men properties and the fact that Scarlet Witch and Quicksilver are mutants in Marvel Comics.

Hilariously, it also made a factual error. It referred to a scene with a suitcase bearing the name Essex Corp, a nod to Nathaniel Essex, Mr. Sinister, and claimed it showed up in Logan. That scene was not in 2017's Logan; it was in the post credits scene of 2016's X-Men: Apocalypse. I have never gotten that detail wrong, but somehow ChatGPT did.

A Poem From Mission: Impossible III Helped Me Clone My Voice

Next, I needed a digital clone of my voice, which anyone could construct. There are literally thousands of hours of my voice online from seven years of making this kind of content. I pulled a few samples and loaded them into software called ElevenLabs, but I wanted the voice to pronounce any word ChatGPT might write, so it needed to be as accurate as possible.

Then I remembered the scene in Mission: Impossible III where Ethan Hunt has to impersonate the arms dealer Owen Davian, played by Philip Seymour Hoffman, including using a digital clone of his voice. It's an ironic plot point, because the Hunger Games movies later did this with Philip Seymour Hoffman after he died, which is just very, very creepy. In the film, Tom Cruise forces Hoffman at gunpoint to read an odd phrase from a card.

I did some research and learned that this was a poem composed for the production by a real life linguist, after a Paramount producer asked for a poem that would use every vowel and consonant sound in the English language. So I read that same poem aloud myself to make my voice clone as accurate as possible. I figure if someone really wanted to clone my voice, they could just force me to read this poem, or trick me into reading it by offering me free Chipotle. For time, I only fed four voice samples into the ElevenLabs instant voice clone generator, but with a lot more and some audio engineering you could theoretically get something even more accurate.

My 35 Minute Clone Was Flawed, and My Audience Data Suggests That May Not Matter

The final step was actually the easiest: creating a digital puppet of my face. There are thousands of hours of me online, most of them conveniently in front of a blue screen. I took a screen grab of my face, dragged it into the D-ID software, added the audio of my voice clone reading ChatGPT's script, and clicked generate. Script, voice and appearance, all 100 percent composed by artificial intelligence.

Is Voss 2.0 perfect? No. The voice is pretty off, there's an outgrowth on the side of his face, and my eyes look like Puss in Boots when he turns on the charm. If I had let him read the whole script, he would have misidentified which X-Men movie the Essex suitcase scene shows up in. But I spent about 35 minutes building this guy, and I would bet that for a less attentive audience, this freak would have gone right over their heads.

A lot of people do not come to my work for me, and the analytics prove it. It's pretty depressing: for my recent Guardians of the Galaxy Vol. 3 breakdown for New Rockstars, one third of the audience clicked away in the first 30 seconds, and for many of them it's probably because they saw a human face. Some people like an on screen host, but according to the comments, a lot also want me to stop doing jokes, stop doing voices, quit the personal anecdotes and hot takes, and just list the Easter eggs. That's a big reason I expanded into this second project. In my recent Matrix breakdown, I worked in glitches where I switched shirts constantly, and most people did not notice until I pointed it out. I'm one of the few movie analysts who actually appear on screen; most of the others are much smarter and do off screen voiceover. I'm swimming upstream against an audience increasingly used to not needing film analysis from a human face.

AI First Drafts Would Make Marvel Writing Worse, Not Just Cost Jobs

Similarly, do Marvel fans need lines like "what is grief if not love persevering," written by a WandaVision staff writer? Or would they be fine with ChatGPT generating a list of MCU callbacks to work into Avengers: Secret Wars? Sadly, in 2026 Joe Russo may be proven right, and Marvel fans may prefer a version of Secret Wars generated by AI from plot points and dialogue in Jonathan Hickman's 2015 comic run, MCU callbacks on a list, and tropes from various crossover movies. They might even hate a script by Michael Waldron that builds on the themes of identity and free will he wrote into Loki.

If ChatGPT or other AI scripting software takes over writing first drafts, not only will writers lose jobs, the writing itself will get worse, and we will ultimately suffer. AI at its best can only repackage diminished returns of writing that has already been generated. In Guardians of the Galaxy Vol. 3, James Gunn wrote a fascinating line for the High Evolutionary, who complains that his creations are only capable of rote memorization and not true invention, and warns that a population unable to think what has not yet been thought will die on the vine. In one sense, that's a statement about cliche versus originality in storytelling.

AI Probably Won't Kill Movies, but It Will Transform Them

You could argue that human writers are also sourcing from an index of influences, tropes, cliches and archetypes, stringing together words and phrases the same way ChatGPT does. But what makes the words people write matter is the soul that binds them. Effective art is grounded in real human experience, and any attempt to replicate that will always fall into an uncanny valley. A keen audience will always be able to tell the difference; the only people who won't are the ones who already gave up on caring. I bet you cared more about this piece because I led with my anxiety over a secret plot to replace me with a digital clone.

So will AI kill movies? No, probably not, but it will transform them, and we need to support creators seeking fair compensation so they can comment on this brave new world without starving to death.

Do you think you could tell an AI written script from one written by a human?

Watch the original

@deepdivenr · YouTube

Adapted with AI from The Deep Dive's video of 12 May 2023 and reviewed by Digital Fox Talent before publication. Editorial standards · Corrections

artificial intelligenceWGA strikeChatGPTdeepfakesscreenwritersMarvelElevenLabsMission: Impossible III

More from @deepdivenr

All →
Write for SUBPLOT

Make videos? Your next one could be an article here.

We turn your videos into articles under your name, linked back to the original. You keep 60% of what they earn, non-exclusive, and you can leave whenever you like.

Become a SubPlotter →