Mark Gimein is a writer and editor in New York.
About six months ago, I threw myself into making videos using artificial intelligence. Yes, I know how that sounds. I have gotten disdainful comments from friends and knitted eyebrows from family. I have spent well over a thousand dollars on the hobby and been rejected from two AI film festivals — one with the humiliating consolation prize of discounted admission to see the movies that beat mine.
I know that the form is not in high repute. I also know that talking about AI video is at odds with the current mood. I know there is a chance — you can call it p(doom) if you insist — that chatbots will kill us all.
Until that day comes, however, I will persist with my AI moviemaking. Hear me out.
Four years ago, the prospect of AI video was greeted with amazement. Google’s brief clips, including an ice-skating teddy bear and a melting ice cream cone, received wall-to-wall coverage. Then came the Hollywood strike of 2023 and predictions that AI would destroy actors’ livelihoods.
And that was that. Producers went back to making new iterations of comic book franchises, and actors went back to beefing on Instagram. AI film was exiled to the slop mines of second-tier influencers.
For some of us, though, the dream didn’t die. Years ago, I spent nights and weekends working on scripts for TV pilots I was sure would take Hollywood by storm. They did not, and I figured that was probably the end of my entertainment ambitions.
But this year, I saw another path. Google was getting good reviews for its video model, Veo, so I figured I would give it a try.
I embarked on my first effort: a black-and-white faux German expressionist version of Franz Kafka’s “The Trial.” If you are thinking the commercial possibilities of this are limited, you are correct.
Still, I love “The Trial,” and I wanted to see what Google could do with it. The visuals of claustrophobic middle European streets were gorgeous. The AI actors looked good, if a little stiff. Unfortunately, Google’s video generator couldn’t keep track of which characters were supposed to deliver which lines, I could only make 10 seconds of video at a time, and the characters looked slightly different in each scene.
This is how I discovered the magic of Chinese technology. Yes, I understand how Chinese AI models could be laying the groundwork for a massive campaign of propaganda and potential cyberwarfare. When it comes to video, though, this is where the action is.
The current state-of-the-art offering is Seedance, owned by the company that created TikTok. (This is odd since TikTok already has an unlimited source of free content from humans, but I digress.) If you give Seedance reference images, it will make sure your characters look the same scene to scene. It will also make sure they sound the same, a detail most models leave out.
Most importantly, Seedance didn’t require me to painstakingly prompt it for every 10-second snippet of film. Instead, I could have Anthropic’s Claude prompt it to make dozens of segments at a time. Even better, I could have Claude manage the whole process, asking yet another AI (OpenAI’s ChatGPT Images) to create pictures of the characters and opening shots for each scene, fill in any parts of the script I was too lazy to write and stitch it all together into a mostly finished film. (The Washington Post has a content partnership with OpenAI.)
With Kafka out of the way (and failing to find a willing audience even within my own family), I decided to put all these discoveries together into lighter fare: a high-school-comedy version of “A Midsummer Night’s Dream.” I sketched out the general idea and left most of the script to AI, which I figured would be familiar with Shakespeare’s works and the oeuvre of John Hughes.
I waited about an hour and a half as Claude fed all my scenes into Seedance and watched them come out at a cost of about 16 cents a second. I then spent a week regenerating each scene half a dozen times to get the AI to pronounce “Titania” and “Lysander” correctly.
My reward was watching my 13-year-old suffer through the final product. He cracked a smile at Titania’s “forgeries of jealousy” speech, wonderfully condensed into one line. Then he rolled his eyes, muttered something about “dad’s AI stuff” and walked back into his room.
This was all the encouragement I needed. I turned to my latest masterpiece: a 7.5-minute James Cain-inspired noir, with a retro look and a philosophy-quoting con man. The Technicolor imitation looks beautifully authentic, though I did need to revise one scene 17 times to get the AI cigarette smoke right.
When friends are over, I like to make them watch my projects, prompting fierce grimaces from my wife. They always ask me what I plan to do with the movies. I never have a very good answer. If I make it into one of the AI film festivals, I’ll be happy. But the embarrassing truth is that making movies is the most fun thing I’ve ever found to do with a computer.
I am not pretentious enough to call my AI video-making art. But I have seen enough of the astonishing leaps in AI filmmaking technology over recent months to believe that soon enough, some other eager amateur will make something that is art, or at least prestige entertainment.
It’s common now to think of AI as a zero-sum project: Either people will keep doing something, or the computers will take over. I think the opposite. If anyone can make a movie sitting at their kitchen table, that’ll be a win for humans no matter how it’s made.
The post How I became a proud producer of slop appeared first on Washington Post.
