Introduction and Camera Techniques

00:00 AI video models have gotten so good over the past year that a great AI movie doesn't really depend on the model anymore. It depends on how you write the prompt. And every shot you're watching right now is proof [music] of that. So, in this video, I'll show you exactly how I make cinematic AI scenes with the right settings, tricks, and prompts. And by the end, you'll be able to make your own today. >> [music]

00:36 [music] [music] >> Now, the tool that I'm going to be using for this is called Higsfield, which is also the sponsor of today's video. If you want to follow along, I'll leave a link to Higsfield down in the description below. Now, the way almost

00:59 every video prompt gets written is just one big block of text. So, the person, the place, the action, [music] the camera, and the lighting all get thrown in together. And that is exactly why those videos never feel like a movie. Because when [music] you describe everything at once, the model gives everything the same attention and nothing in the shot ends up looking important. So, the way I write now splits the [music] prompt into four parts, which are the camera, the acting,

01:22 the lighting, and the way it was filmed. And each part gets written its own way. So, I'll pick Sea Dance [music] 2.5, set it to 1080p and 21x9. Turn the audio on and set [music] the duration to 12 seconds. It's really important to set the resolution to 1080p yourself because Seance defaults to 720p [music] and that gives your video a soft, blurry feel, and a blurry AI movie really spoils the whole thing. So, the first

01:44 shot I'll try is a woman on a river ferry in the 1880s. And the moment the ferry pulls away, her cargo chest rolls off the deck and falls into the river. In the prompt, the camera gets four separate parts instead of one line at the end. So, there's the lens where the camera starts, how the camera [music] moves, and where it ends up. Every prompt I'm using is in the workflow document linked in the description, so you can copy the whole structure

02:06 straight in. >> [panting] >> Oh, for God's sake. >> And honestly, that looks like a shot from an actual movie. The camera starts down low next to a wheel, [music] pulls back along the deck as the chest rolls toward it, and swings up to her face the second the [music] chest drops. Now,

02:29 I'll try a shot where the camera runs alongside the action. And this one is a woman catching a departing train and missing the bottom step. So, it's the same settings again. And this time, I'll set the lens to 18 mm, which is a very wide lens. And I'll tell you why that matters once it's [music] generated. And this one came out really good. Her

02:57 boots stay huge in the foreground [music] while the train wheels turn right along the edge of the frame. And the camera only rises up to her face once she's actually made it onto the train. [music] And the reason I picked such a wide lens is that it makes everything close to the camera look huge and [music] everything far away look tiny. So the last camera move I'll try is the hardest one, which is where the camera pushes in toward the person while the lens zooms out at the same time. In

03:20 the prompt, the lens part says zoom out from 85 mm to 28, while the move part pushes the camera in from [music] 6 m to two, so his head stays the same size in the frame while the whole canyon opens up behind him. And honestly, this one looks incredible.

03:46 The canyon wall shrinks behind him while he stays exactly where he is. And that only happens because the prompt [music] says the camera has to physically move and the lens has to zoom. So, three different camera moves came out of the same four [music] parts. And all I changed was the numbers. But none of those three has two people talking to each other. And the second you put two people [music] in a shot and give them lines, the prompt has to change completely. So the second biggest part