I've been awaiting OpenAI's video generation program "Sora" for some time now. All the tests I've seen from it indicated that this would be a game-changing AI program that will spell the end of the old ways of Hollywood and push us into a new age of movie and television creation.
As such, I've been proclaiming the end is nigh for Hollywood for some time, and I consider this to be a fantastic thing for society. Hollywood has clearly become an evil place that uses its control to achieve ends that are nothing short of sinister in some cases, and grotesque in others. It attempts to push degradation as a virtue, create villains out of good guys, and people have gone into it with various goals and aims, some of them innocent, others not, only to be chewed up and spit out like so much rubbish. Some of these stories even have scars that run deeper than anyone deserves.
Paired with the greed and ideological radicalism it's infected by, Hollywood can't become a thing of the past soon enough.
But after playing with Sora for a little bit, I can honestly say that day is approaching soon.
Allow me to show you some of the things I created with just a simple prompt. As you watch these small clips, pay attention to the details such as light reflections, clothing movement, muscle movements, and more.
The first thing I wanted to do was create some clips from a western. The inspiration was that of a cowboy and a strawberry-blond-haired woman. The woman is entering a saloon to confront the cowboy. Pay attention to the lighting, the way her hair moves, and the details on the cowboy's face as you watch.
Tried to create a scene from a book I'm writing using Sora. It got some things wrong, but otherwise, Sora's ability to create a scene just by me describing it is pretty decent. @OpenAI has some way to go before this is wholly useable, but color me impressed. pic.twitter.com/ci98QLb0mT
— Brandon Morse (@TheBrandonMorse) December 13, 2024
Here's another attempt I made, this time while outdoors and incorporating a horse.
Holy wow, Sora from @OpenAI is incredible. It generated this with just a simple prompt. pic.twitter.com/ijfgFfrmeG
— Brandon Morse (@TheBrandonMorse) December 13, 2024
I then dove into other scenarios to test out various aspects of Sora's capability.
I created a Knight, watching a fire at night. He's dressed in a battered cloak as a crow moves next to him. This scene is fantastic, as the trees in the background create a haunting atmosphere. The crow moves as a crow would on the ground, with jerky movements. What could impress me more is the lighting, as there's not a lot of flickering from the fire, but otherwise, the shading is done perfectly here.
Been playing a bit more with @OpenAI's Sora, and I've got to say, outside of the random oddities that still plague AI video generation, the detail and cinematic nature of these videos is incredible.
— Brandon Morse (@TheBrandonMorse) December 17, 2024
Here's one I made of a knight sitting next to a campfire. I asked it to put a… pic.twitter.com/uFFjBQ9rVn
Next, I wanted to see how Sora handled surreal moments. I decided to make a classroom scenario, but the teacher was a Minotaur. I didn't give the AI any details about the environment, allowing it to make it up on its own. What I got was interesting. I wasn't overtly impressed with the Minotaur teacher as it looks like a standard puppeteer mask one would wear in a 90s movie or a modern commercial, however it wasn't a bad take. The students are well created, acting naturally, and the classroom setting featured a blackboard with fully rendered pictures that look very good in the background, arranged in a way that doesn't look chaotic.
I decided to have some fun and create a classroom where the teacher was a Minotaur. I wasn't super impressed, as the Minotaur's head looked more like a 90s' era puppet mask, but the details of the classroom, including the students, looks fantastic. pic.twitter.com/PoRVkfB6nX
— Brandon Morse (@TheBrandonMorse) December 17, 2024
This finally led me to wondering how well Sora would handle body movements, so I told it to create a scenario where a woman in a bikini was running down the beach with a St. Bernard. Of all the things I asked of it, this would probably be the most difficult as it has a lot of things it has to consider, namely the body shape of both the woman and the dog as it moved. Sure enough, the render ended up being plagued with issues. The sand moves oddly under the woman's feet, a leash appears and disappears in both hands, and at one point, the dog's head becomes a tail.
However, despite these obvious issues, I was impressed with how the woman's body moved naturally as she ran. You can see her leg muscles act naturally, and the fat in those areas jiggle in reaction to the force of her footsteps. Her arms, breasts, and shoulders move naturally with the rhythm of her gait. Her hair is slightly out of sync with her movements, but it still moves like hair. Even the shading of her body was consistent.
Likewise, the dog's body also displayed the right muscle movements. Its ears moved very naturally with its gait, and I was impressed with the shadows interacting on its body depending on where the sun hit it as it moved. Like the woman's hair, the tail seemed to only move in unnatural ways.
I next wanted to see how Sora handled body details and muscle movements so I told it to generate a short video of a woman running down the beach in a bikini with a St. Bernard. There are a lot of weird things that the AI did in this video, including a leash appearing out of… pic.twitter.com/6xedV506aR
— Brandon Morse (@TheBrandonMorse) December 17, 2024
Watching these videos, it's pretty clear that AI video generation is still rife with consistency issues, but it's definitely something that OpenAI is trying to get a handle on. One way they actually attempted to create consistency is with "storyboards," where it takes a character or scene you created and uses those exact elements in another video you can generate.
Sora, and programs like it (Google just announced it's going to release its own) are not at the point where they're going to topple Hollywood just yet, but they're getting there, and they're getting there fast. Just think that not that long ago, AI video generation looked nightmarish, with AI unable to fully render human movements well.
However, I don't see many of these issue being around long. In two to three years, I imagine we'll see issues such as oddly moving hair or dog's heads turning into tails being a thing of the past. Story boarding will allow for more consistency, and there are some programs that are already working to include sound integration.
As I've covered before, programs like Sora will take movie creation out of the hands of the Hollywood elite and put them into the hands of the people, opening up a wide variety of ideas and fresh creativity to hit the internet. At some point, we may even see AI creations on the big screen, created by people like you and me.
(READ: The Tragedy That Is the Lost Art of Story Writing)
It's going to be a wild time, and while that time is still some years away, it's approaching fast.
Hollywood is about to become a relic, replaced by the people and their AI-assisted creations.