Hollywood vs AI: My Journey Creating a Visual Drama With Artificial Intelligence

0:00 Over the past week, I've been able to experiment with AI content generation. And I've been able to create a mini episode, if you will, just by using some of the top models that are out. You'll be surprised at some of the things that I've learned. So stay tuned to figure out how I made this scene. I'm not wired for that kind of faith. You think the law is about guilt and innocence. It's about process. Protecting rights. It's about balancing in a world, tipping into chaos. In AI purely.

1:00 So go ahead and jump over to youtube.com slash jumpcutking or go to jumpcutking.com. Get old links and be redirected to YouTube and watch the full experience. But of course, if you're listening to this in the car, you might be able, might be lost a little bit. But you're going to get a lot of cool techniques. I had someone who called me for an emergency and was like, hey, can you generate these crazy AI baby videos? You saw me do one. And this was a direct response to someone saying, can you do it?

2:00 Because that's where I feel God has led me to be able to tell stories that are going to change people's hearts. And hopefully change the world. However, that's a multi-million dollar industry. We're not going to talk about how much money Apple spent recently on one of its largest nominated TV series, Severance. They paid a pretty penny for that. We still call it TV, but it's not TV anymore. Netflix, Paramount and all of them, they were TV stations, but now it's online streaming web series. I just don't want to call it a web series because you'll feel like it's been downplayed.

3:06 There are some things I don't agree with as being an Isaac Asimov reader, and I'm like, I wouldn't have done that, and you guys aren't doing this, this, and this. But it's pretty darn cool to take a multi, for lack of a better word, millions of years, I don't know if it's actually millions of years, or hundreds of years, a series of books spanning hundreds and hundreds of years, and make that into TV. That is a very, very creative way of doing it. So, I experimented a while back with artificial intelligence, and I generated, I took a scene from an old script of mine, and I mean real old, I mean so old that the amount of grammar issues on it's entertaining.

4:11 Here's the problem with Vivo. Vivo takes in either a start frame or generates its own frame, and it creates the scene. But if you ask it to continue the scene, it's not going to generate the same characters, it's not going to generate the same environment, and to be honest, what was really, really cool is kind of depressing. Not to mention the amount of cost. So, how AI in general works is they're all powered by, usually, NVIDIA core processors, unless you're Apple Intelligence, who has their own secure private cloud compute, which is specialized servers designed with Apple Silicon that have no, pretty much, no hard drives designed so that they operate, run the AI model, and send it to you, and then, you know, no information stored in cloud.

5:14 A CUDA core is a specialized processor unit that's designed to do massive mathematical calculations, used to help you get some of those brilliant lifelike gaming qualities that you see everywhere. But all of those have intensive computational components, so when NVIDIA started building this CUDA core, many, many years ago, it started to be a big thing for editing. So, before Apple Silicon, it was recommended that you get a Windows PC, and you get CUDA cores, NVIDIA graphics core, a decent one, and that's how you would render.

6:31 And you can get models that are 13 billion parameters, they're literally 36 gig compiled, if you will, trained models that will help produce the items. This is what Hydra actually uses, the Halo 3 model, Halo 2 and Halo 1, which were lip-syncing models designed to take a frame of video and render it. And the new Halo 3 model does all of this cool extra stuff, so, like, if someone's talking, their hands would move with them, or paper is open, as you can see from the demo that I created.

7:33 So, what cloud-based solutions do is they have server farms, whether they're rented through Amazon, usually, or whether they're rented through someone else, or they build their own. If you're ChatGPT, hopefully, you've built your own. All of these cloud models use these vast servers that are built with either, whether they use the new, it's not technically new, but the Google chip designed specifically for Tensor. It's Google Tensor 1, 2, whatever it is now. Or it's NVIDIA graphics cards, which is why NVIDIA stock is always screaming hot, and graphics cards are more and more expensive every time you look at them.

8:38 I know, it's crazy. And so, you have kind of an unlimited, if you pay for the ChatGPT Plus, or ChatGPT Pro, or ChatGPT Enterprise, ChatGPT Team, whatever it is, you get kind of unlimited image generation. It's great. But then when you go to use Google Veo, or use Sora, which, I'll be honest, Sora during my tests was just dreadful. I'll be honest, it doesn't have a lot of the computational power that some of the other models does. It doesn't do any lip syncing, and I will add, it does random stuff.

9:50 All of these things, and it should be, you know, enter these four commands, and you're good. But I've found that various different versions of updated Windows can conflict, and you can, it'll force you to have a version of Python that's above and beyond the specifications, and you should go in and change dependencies. And if you're not a coder, and you're a layman, you're never going to successfully run these models. Now, there are things called, like, ComfyUI, which is a huge improvement on the ability to generate some of these components.

10:51 It took 35 minutes, and to be honest, generated nothing usable. Just with one of the models I experimented with. So I decided that I was going to use the paid route, and I was going to pay for ChatGPT+. So I get an image, one image about every three to five minutes. It's not very great. So if you're multitasking, it took me four and a half hours to do this less than three-minute segment. Okay, it's less than a three-page script. Not great, if you ask me. I want to just be able to continuously create and not stop.

11:57 And my concept behind it is using AI kind of as a replacement for an animation studio, not to replace animators. I know my whole goal here is just to save money, because I don't have a multi-million dollar paycheck. I can't run around filming these glorious worlds and these stories I have in my mind. AI is going to let, is kind of going to bring storytelling to those that are willing to use it. But I will say, it does take a brain, and it does take a unique way of looking at it.

13:00 Did it just die? I think it died. I think it died spectacularly. There it goes. Okay, so it's going to go ahead, and it's going to generate an image. Now, here's a problem. I didn't specify the height and width of the image. Please note, I have memory turned off for two reasons. Because I have a vast amount of clients, and a vast amount of things that I do with it, and I don't want it to assume. And it always assumes, like, at one point in time, it assumed one of my clients' business.

14:03 It's more 16 by 10. It's a, or 4-3, that looks weird. And I go over here, okay, so it's rendering another one. So here's another group, right? 1, 2, 3, 4, this is the first image, and that's the second image. That's a problem. No two images are going to ever return the same result. A lot of AI models have a problem with hands and getting faces right. And when I'm animating multiple frames, if it's realistic, you might see hair kind of moving inside someone's face, or you might see them sing in a chair, and all of a sudden the chair is in front of them, and it's just a weird experience.

15:04 But you'll notice from the look as it's generating, it's a different color. It's just, it doesn't do consistence frame by frame. So as part of my experiment, I was like, okay, what can I do to make this a more consistent process? Let's think of an animation studio. Literally, you know, pulled a bunch of YouTube videos, pulled up a bunch of notes on what animation studios are, and thought of it kind of like a production studio. If I have a group of people, and you're doing this, you're doing this, you're doing this, I'm going to not tell you, hey, do me an entire series and give it to me tomorrow.

16:06 I don't understand why it's an animal, but these character cards are what animation studios will use to keep kind of characters consistent. So I found two, plugged them in, and asked it to generate me a character card. All right, so your front, back, close up. And I asked me, asked it to generate that in different wardrobes. So here is the person in pajamas. Then I asked it to do the same thing for another character. Here is the judge. Pretty much what I'm doing is I'm breaking it up in component parts.

17:07 Okay, great. I have a script, right? It's one, two, three pages long. So now that I have a script, I now have an environment, right? So I have my two characters, Ty and Judge Tormek, and then I have my environment. Think of this as you're building on top of one block and you're building on top of another block. Here's a courtroom. I generated a bio, generate the character cards. I took the bio and I created a scene, took information from the script, plugged it into the scene, and here is my wide shot.

18:10 So here is a tight shot, and then I noticed something, you know, it's 16 by 10, which is what I told it to. I should have said 16 by 9. I was stupid in my head in IMAX format or whatever it is. Then it stopped generating in 16 by 10. ChatGPT doesn't actually have memory. It has, it knows a certain amount of tokens. And so in image generation, it actually doesn't pass memory to the image generator. Sometimes it'll generate based off your previous image, but it will generate whatever it wants, whenever it wants.

19:13 Figured that way for the lightning before I ducked. The other character. They go back to ChatGPT, and here we are. Here's the judge, and we're generating different forms of emotion. Here's a close-up. Here's a close-up where he's happening. He's leaning in a little bit. Here's a close-up where he's leaning in. He's emphatic, and he's angry, and he's, you know, a judge. Here's my other character, right? This is him extremely angry, although in the clip, you don't actually feel that anger from the voice. And just going back and forth and generating these different shots to try to pull out what I can use.

20:22 But as you notice, it's changing the images on me, and it's changing the location of items on me, and they're not being consistent. And they have to kind of constantly conform it. The best experience I've found is if I provide my source material, ask it to generate a shot based off the source material, use the wide shot or an extreme wide shot as a source material to generate each individual frame. I'm kind of uploading, here's my characters, boom, boom, boom, here's my scene, here's what I want, here's the wide reference, here's what I want.

21:24 So I realized that I needed to replace this shot, and here's a close-up. Problem I have with this is he's not even looking at the right character. The character's over here. But, I can kind of get away with it, right? Because I have a little bit of screen direction here, a little bit, not much. I'm kind of at the mercy of it. Here are the top 5 things that I learned while doing this project. One, AI is like a 5th grader, and its attention span is less than 3 seconds.

22:28 So, in coding, if I have a function, and I give it variable, I want it to output exactly what else. So, I want to say, okay, I want you to multiply this number and provide you the function. I said 2, and it multiplies it by 2. Okay, it's always in return 4. So, you're kind of like narrowing down the infinite options that the neural network may generate, that AI might generate. And you're narrowing it down to these very specific, you're helping it out. Two, it is not good at batch image generation.

23:29 Put it into a database of successful prompts. Put it into a notepad. I don't care. Whatever you do, when it works, save it. And that will prevent you from having to do all this extra work in each time and trying to retype because you're going to get different responses. When it works, copy and paste it. And you can do what I do. Here's my prompt, and here's the custom information that I say in the prompt. I'm going to provide scene information. I'm going to provide X, Y, Z, and then I'm going to provide it underneath the prompt.

24:29 You can't give it too much information because it'll then output something completely outside of the prompt, or you're kind of constraining it too much. You want it to treat it like it's creative, even though it's not really. It's generative. But you want to give it enough information of a prompt so it knows what to do and enough information to help it. If you're asking yourself, okay, I want you to generate this, this, and this, and that's going to be your output. But instead, think, okay, let me generate this item first, this item second, this item third, then generate this item.

25:29 and even though you tell it and you have to constantly perform your prompt and constantly redo, and when they removed the edit button in some of their interfaces, you can't go back and change the prompt. If you know a prompt is successful, copy and paste it, start it in your chat. I always recommend starting a new chat and turning off memory. You're starting new and fresh, treating it like a new employee, a new fifth grader that you're walking into. Hey, do this. Hey, do this. Never reference another chat.

26:36 The moment I told it to cite references, it stopped generating or hallucinating content. In fact, with those references, I'm able to check and confirm everything that ChatGPT said. So I was able to be confident, 80-90% confident, in the content that I said in those podcasts, because I had it generated with AI, and we worked with it together. Five, artificial intelligence isn't bad. It's a time saver. But it's a tool more like CapCut, Photoshop, Microsoft Word. It's a part of your creative mindset. It's a part of your creative toolkit.

27:40 It's a friend in the respect of being able to help you think things differently, being able for you to help problem solve. But it is not your friend to do everything for you. Don't, ChatGPT isn't designed, or even if it was, it should not be a replacement for your brain. It's not a replacement for your marketing company. It's not a replacement for an animator. It's not a replacement for a voiceover artist. I found that AI writing level is less than a fifth grader, and its comprehension level is kind of pretend up here, but if you really read it has no substance, and it needs that human element.

28:47 You know, it's expensive. And so when I go to generate it, the responses I get are great, but are costly. And so for me to do an actual television series or an episode of a television series is still multi-thousands of dollars of credits going back and forth and not having the ability to conform the tool, not having the ability to experiment and try new things. It's because credits costs, they multiply, you know, it's like three to six seconds for a Hydro clip, but only 60% of Hydro clips I can use.

29:52 Generative AI is not bad. Yes, it may take a whole bunch of jobs, but it gives creative professionals like me that don't have multi-million dollars, don't know the people that are in Hollywood, would love to meet some of these executives and do experiments and to, you know, film some of the movies and television series that God has placed on my heart that are great stories that would make you cry and make you laugh and kind of change your perception on the world, which is the whole goal of the universe, is to get you to live this revolutionary life.

30:54 AI can elevate your toolset, AI can elevate your ability to accomplish something, but at the end of the day, if you're not careful, you will think you can't accomplish something because the amount of money something costs, and I will tell you, as a strategy consultant, if you will, I tell everyone, you can accomplish anything that you put your mind to. So, God can give you a creative solution, and I believe God gave me this creative solution. I thought I'd share it with you. I'm not trying to hide information.

31:54 It does have a little bit of strong language. Again, it was partially AI-generated, and in an effort to not have to do real deep character design, I kind of just kept some of that stuff in there. So, this is it. You're going to stand there all day like you wandered into the wrong room? I'm going to sit and act like the Lord, but my faith and trust it. Some men I know can handle these four damned cases. Figured I wafered the lightning before I ducked. You think I'm striking you down some?

32:55 Great. You still fight for his rights. Because of setting you down to let the system turn into a machine that crushes the innocent with the guilty. That's why I chose you. That's why I gave you these cases. You want me to fight for men and women who hurt people. Who admit they've hurt people. I want you to fight. Because someday you'll sit across from someone who can't. I want to fight for monster- I want you to fight so bad that they only do. I love you.

33:57 Protecting rights. About balancing and a world tipping into chaos. I didn't pull your name out of a hat, son. I saw what you are. Underachieving, maybe. But only because you're afraid of what happens if you do try. Son, the chosen one. Got it. You're not that special, Ty. People can change things if they stop hiding behind how hard it is. What if I fail? You will. But if you fail because you've hesitated, because you half-assed it, that's on you. It'll be your name they remember. You ever get tired of carrying all this weight?

34:57 the body camp footage that DA missed. That's a start. Don't be afraid of what you can be. As you can see, it shows great promise. There's limitations. There's some cool creative opportunities. Kind of works, kind of doesn't work. It's back and forth kind of thing. It's experiment. And it's proving that it's possible. I was limited to one character in a shot talking. That's it. That's in the four and a half hours. Now, if I had more time and a lot more credits and a lot more or maybe just upgrade my gaming machine to use an H100 or the 5070 Super that's $5,000, that gives you a lot of ability to use some of these open source models and builds your story without credits.

36:03 that anything is possible. That there's nothing that should be limiting you. I want you to know that nothing should limit you. That you can accomplish anything that God places on your heart. And you don't know when you're going to accomplish it, how you're going to accomplish it, or what tools you're going to accomplish it. All it requires is faith. So I want you to know to have faith today. I pray the Lord gives you a special measure of faith. We've all received a measure of faith when we were born.

37:08 thoughts in. So I rebuke any negative thoughts. Anything that says you can't do it. Because you can accomplish anything that you put your mind to. And God's proven that time and time again. And I pray right now that you understand the creativity and the brilliance that nothing holds you back. And God gives you the boldness to pursue today. In Jesus' name.