We're presenting a full feature film in can in less than a month. 80 minutes long, fully AI generated. Now, the issue is that we don't have the film yet.
What we do have is 10 million credits, a team of 15 people, 14 days start to finish, and no clue if we'll make it. Welcome to episode 1. Hi, I'm Adil.
I'm one of the people generating this film and I'm telling you right now, this is either the future of cinema or the worst decision we've ever made on camera. Here's what's happening. We're making an 80-minute film, fully AI generated.
Again, no cameras, no sets, no actors. Well, except for me. It premieres at Khan in less than a month, and we only have 14 days to finish the whole film.
And with this film, we want to prove that AI can actually produce cinema. What does that mean? A viewer watches 80 minutes and forgets they're watching AI.
The film holds together as a story. Beginning, middle, end, payoffs that work. Characters feel like characters, not generator figures.
The emotional moments actually land. And people feel something they didn't expect to feel. If those things happen, we proved it.
If they don't, we learned what's still missing. We have 15 people working on this, including me. The same people that worked on our Kickfield original series like Arena Zero, Hell Grind, Zephyr and some of whom you've already met previously.
14 professional directors, do obs, editors, and one seli are joining to make this happen. Needless to say, this has never been done before. And honestly, I don't know if we can, but we're doing it anyway.
Every couple of days, I'm going to drop a new episode. Same place, same time, same chaos. You'll see the breeze, you'll see the prompts, and you'll see the credits burn.
And most importantly, you'll see what breaks. That'll probably include me. This is Hell Grind, the continuation of episode 1, and it's the first AI feature film in history, if we make it.
So, quick context, we didn't start from zero. Hell Grind episode 1, the one that's already out, is 22 minutes long. You can check it out right here.
Here's what it actually took. 253 videos made the final cut. Out of 16,181 generations on CS 2.
0, 107 images made it in out of 10,71. We generated total credit spend, brace yourself, 1,152,95 credits. That's about $69,000 for 22 minutes.
And that's not waste. That's the craft. Every generation that didn't make it taught us the prompts for the next one.
That's what I want you to understand before we start. Masterpieces aren't single prompts. They're the 26,000 you don't see.
So, here's where it gets interesting. 22 minutes, almost 27,000 generations. Now, we have to make 60 more minutes in 14 days with 10 million credits.
Now, in traditional production, to make the first episode, it would have cost millions of dollars, month of pre-production, and years from script to screen. But don't take just my word for it. This is Chuck Russell.
He directed The Mask, The Racer, The Scorpion King. He's been making blockbusters in Hollywood for over 30 years. If anyone's going to give me a real number on what episode 1 would have cost in the traditional system, it's him.
Hi, I'm Chuck Russell. I'm a writer, producer, and director. And I was just asked for my first reaction uh watching Higsfield new production of Hellgrind.
And uh I was pleasantly surprised because I was entertained. That was my first reaction. While I would imagining looking at any AI production with a more critical eye, I got into the characters.
I got into the story. It's about a bunch of edgy street kids who get superpowers. So it's a traditional genre.
Very ambitious for an AI production. About a 20-minute production. Uh, and it was fun.
So, this was full AI. What we're doing in mainstream is liveaction actors with AI enhancements for sets, visual effects. It's a new tool in the visual effects toolbox.
For a full AI production like this, though, it's evolving quickly. Uh, congratulations to the filmmakers. You made me care about your characters.
Something I almost never see in AI. Uh, so I thought it was a great job. You have to understand this is a micro group of people, probably a handful of people.
It's a micro budget for sure. And so if this was a traditional studio workflow with the same kind of material, you'd be talking about maybe a minimum of $5 million depending on cast for this 25minute piece. So it's a whole new opportunity for independent filmmakers and for people that are making films in the mainstream like myself.
We're still developing techniques that use our live action actors. This all started back. It's not that new in 2019.
My in my impression, the the Mandalorian creating endless landscapes on volume screens was one of the first forms of AI, which to me is only smart CGI. So, fear no AI. It's a wonderful tool.
It reminds me a lot of when we uh did CGI in the mask where no one knew it was brand new, no one knew where it was going to go. So, I think it's always going to be about the human heart. Uh there's authenticity in these these mini productions.
Uh that's one of the biggest advantages and I saw that in Hellgride. So congratulations folks. Now you did the math whether it's worth it.
So here's the process. Before any of the generation starts, we read the script. I know it's hard to believe, but it actually took only 3 days to finalize the entire script for the film.
And I'll show you in a second how we streamlined that entire process. As for the script itself, I'm not going to spoil it, but I'll tell you this. There's a moment in this film where the room went completely silent.
I even cried a little bit. So now our mission is to convey those emotions to the viewers through the generations. So how did we write the script?
We built it using a custom claw skill we trained for screenwriting. Instead of asking an LLM to write a scene, we built a system that knows our characters, our tone, our pacing rules, our world. We feed it context and it outputs scenes that actually fit the film.
That's the difference between AI written content that sounds like AI and AI written content that sounds like us. I'll leave a link to it in the description. Just make sure to adjust it to fit your writing style and voice first.
The easiest way to do that is to run it once with your idea and then any parts you don't like, you can tell Claw to change it in the skill itself. If you want a deeper dive on this, let me know in the comments. But let's move to the most interesting part of this video.
So once the script was logged, every asset went into one shared canvas. Now this feature is still in development and the directors and I actually had a lot of input on how to make it the most useful and easy to use for you and well for us. It'll be live on the website by the time you watch this video hopefully.
So definitely go check it out. Here's what we used it for. So first is the style prefer.
This will be hard baked into every single prompt when we make the shot list in the next video. We got the style, lighting, color, camera. I mean, very detailed so that shot one and shot 80 stay consistent.
We got some old iterations here, but this is the final one we locked. Next up, we got assets. So, we have locations, characters, props, and as you can see, it's not just the names, but a detailed description so that we give the model as much context as possible.
Next up, we have the characters and a bunch of different forms. Uh in the base form, as you can see, we have the prop sheets baked into the character sheets. That way, we don't have to upload multiple images at once.
And Cance actually does a pretty good job at rendering such small objects throughout all shots. And generating them in different forms is very important, especially when working in a team. Say I'm generating shot 78.
And in shot 13, Frogo received a lot of injuries. If I just use his base form, the viewers might not necessarily notice, but they will know something is off subconsciously. We've locked everything we could as of right now.
As we work on some of these scenes and create more assets, they will be immediately added to this canvas. Actually, I'll quickly show you how to create the characters just so you have an idea. For this, I'm heading over to Cinema Studio 3.
5. Here, I have AI cast selected, and I want a Tibetan monk. Uh, let's do 40 years old.
Uh, and let's make him bold. Now, I'm going to generate four images, but I'm going to do one at a time. That way, every single one of them is completely different.
All right, we got four nice options. I like him right here the most. I think he would fit our film perfectly.
This one's not bad, but he's not bold like we wanted to. So, let's switch to GPT image 2. Drg him in.
And let's make him completely bold. And there we go. The same actor, completely bold.
We got four options. All of these look nice. I think this one is the best one for comparison.
Let's see what it look like before. All right. And the after.
I mean, same face, same clothing. Perfect. Now the same workflow was used to create all these characters.
But the key here is iteration. We used Soul Cinema to create the characters and then GPD image to edit them. So for Jack for example, we had nearly 600 generations in Soul Cinema and then a couple hundred more in GPD image to get the perfect outfit.
This step is very important because these images will be used for all of the CE 2. 0 generations and will affect the output a lot. So make sure to spend your time here.
And lastly, we have the locations also from a bunch of different angles to keep everything consistent across shots. And this canvas is what holds the whole world together. We split the film into chunks.
Beginning for sequence, the battle, etc. Each chunk goes to a different team. Each team has its own director.
We're going to use this to create a short list with a prompts in the next video. Now, the plan is that every team generates 2 and 1/2 minutes of usable footage every day. At the end of week one, we need to have the full 80 minutes completed so that we can spend week two on refining key moments.
So, at the end of each day, we submit edited 2 and 1/2 minutes to the lead editor who put everything together and give us feedback on what to refine for week two. And it's crucial for us to iterate fast. If anyone falls behind by a minute on say day three, they have to make up for it on day four.
That's the plan anyway. So, here's where we are. Script done, canvas built, shot list split, teams assigned.
Day one of 14 starting now. And in the next episode, I'm going to take you inside the first generation date. I'll show you how to create a short list with prompts for the entire film using Claw.
You'll see me try to land my first shot, and I'm going to show you the breakdown of where we are in terms of credit spend. We've got the script. We've got the assets.
The only thing we don't have yet is the film. Day one starts now. Let me know in the comments whether you think we'll make it in time, how many credits we'll spend in total, and what your expectations are from this film.
As always, if you found this video helpful, hit the like button, subscribe, and I'll see you guys in the next one.