WEBVTT 00:00:00.000 --> 00:00:02.000 I was reading this guy this morning that's a 00:00:02.000 --> 00:00:04.160 director's manual for AI filmmaking just out 00:00:04.160 --> 00:00:08.019 this week, January 2026. And it opens with this 00:00:08.019 --> 00:00:11.599 analogy that honestly it made me just put my 00:00:11.599 --> 00:00:13.500 tablet down. I think I know exactly which one 00:00:13.500 --> 00:00:15.980 you mean. The genius child. The genius child. 00:00:16.199 --> 00:00:19.059 Such a powerful image. The whole idea is that 00:00:19.059 --> 00:00:21.980 AI isn't a robot. It's not really a tool. It's 00:00:21.980 --> 00:00:24.500 a prodigy. It has this infinite imagination, 00:00:25.280 --> 00:00:28.550 but absolutely no discipline. Right. Zero impulse 00:00:28.550 --> 00:00:30.530 control you leave it alone in a room without 00:00:30.530 --> 00:00:32.490 clear instructions, and it just you know draws 00:00:32.490 --> 00:00:35.750 on the walls. It makes total chaos Welcome to 00:00:35.750 --> 00:00:37.649 the deep dive today. We're exploring the state 00:00:37.649 --> 00:00:40.869 of cinematic AI filmmaking It's early 2026 and 00:00:40.869 --> 00:00:42.570 we're trying to figure out why so much of the 00:00:42.570 --> 00:00:45.969 AI video We all see still feels a little off. 00:00:46.109 --> 00:00:48.350 Yeah a little floaty exactly We have these incredible 00:00:48.350 --> 00:00:51.810 tools now cling AI Google vo3 but the output 00:00:51.810 --> 00:00:54.890 is so often this like shimmering morphing mess 00:00:54.890 --> 00:00:57.359 and then Every once in a while you see a clip 00:00:57.359 --> 00:00:58.840 that looks like it was shot on a professional 00:00:58.840 --> 00:01:00.859 camera. Yeah, it has real weight to it. It has 00:01:00.859 --> 00:01:03.299 intention. And the premise we're unpacking today 00:01:03.299 --> 00:01:06.079 is that the difference isn't the software. It's 00:01:06.079 --> 00:01:08.840 the direction. So we're going to dig into the 00:01:08.840 --> 00:01:10.519 craft. We'll talk about the physics of lenses, 00:01:11.260 --> 00:01:14.480 the virtual actor workflow and the tool landscape 00:01:14.480 --> 00:01:16.760 right now in twenty twenty six. We're going to 00:01:16.760 --> 00:01:20.280 go from just typing a cool robot into a box to 00:01:20.280 --> 00:01:22.579 actually directing a scene that feels, you know, 00:01:22.579 --> 00:01:25.109 human. Let's start with the very first principle 00:01:25.109 --> 00:01:28.329 in this guide. It's something that feels completely 00:01:28.329 --> 00:01:31.250 counterintuitive for video. The guide says if 00:01:31.250 --> 00:01:33.390 you want your video to look professional, the 00:01:33.390 --> 00:01:35.969 first thing to do is stop moving the camera. 00:01:36.109 --> 00:01:38.549 The power of stillness. But why? I mean, we're 00:01:38.549 --> 00:01:40.250 making moving pictures. Why would I lock the 00:01:40.250 --> 00:01:42.430 camera down? It is the absolute hardest thing 00:01:42.430 --> 00:01:44.930 for beginners to accept. You pay your subscription 00:01:44.930 --> 00:01:47.810 to Kling or Veo, and your first instinct is, 00:01:47.989 --> 00:01:50.489 I paid for motion. I want motion. I want the 00:01:50.489 --> 00:01:52.730 camera flying through a city. More is always 00:01:52.730 --> 00:01:54.829 more. But it's usually just too much. It gives 00:01:54.829 --> 00:01:57.310 you that kind of seasick feeling. If you think 00:01:57.310 --> 00:01:59.650 about the greatest films, like No Country for 00:01:59.650 --> 00:02:02.650 Old Men or The Godfather, the most powerful shots 00:02:02.650 --> 00:02:04.769 are often totally locked off. The camera's on 00:02:04.769 --> 00:02:07.170 a tripod. It doesn't move at all. Because it 00:02:07.170 --> 00:02:09.669 forces you to actually look at the subject. Precisely. 00:02:10.120 --> 00:02:13.479 The guide uses this beautiful example of a portrait 00:02:13.479 --> 00:02:16.680 of an old fisherman. If you tell an AI model, 00:02:16.939 --> 00:02:19.099 make a cool video of a fisherman, it's probably 00:02:19.099 --> 00:02:21.039 going to have him waving, or the camera will 00:02:21.039 --> 00:02:23.500 do this crazy swoop around his head. It looks 00:02:23.500 --> 00:02:26.180 like a video game. It feels weightless. Right. 00:02:26.560 --> 00:02:29.180 But the professional approach is to lock the 00:02:29.180 --> 00:02:32.060 camera. You specify a 50 millimeter lens, maybe 00:02:32.060 --> 00:02:35.219 an f1 .8 aperture to get that nice blurry background. 00:02:35.560 --> 00:02:39.199 And you focus on the micro movements. What do 00:02:39.199 --> 00:02:41.340 we mean by micro? movements exactly. It's the 00:02:41.340 --> 00:02:42.979 little stuff that proves something is alive. 00:02:43.360 --> 00:02:46.319 The smoke curling up from his pipe, the slow 00:02:46.319 --> 00:02:49.159 heavy blink of his eyes, the mist floating in 00:02:49.159 --> 00:02:51.199 the harbor behind him. You see the texture of 00:02:51.199 --> 00:02:53.620 their wrinkles on his face. When the camera is 00:02:53.620 --> 00:02:55.620 still, the viewer isn't just distracted by a 00:02:55.620 --> 00:02:57.960 big whoosh. They have to connect with the person. 00:02:58.159 --> 00:03:01.539 So why do we equate motion with quality when 00:03:01.539 --> 00:03:04.479 stillness often holds the emotion? Because stillness 00:03:04.479 --> 00:03:06.879 forces us to confront the character's humanity. 00:03:07.099 --> 00:03:11.009 Okay, so stillness is the foundation. But movies 00:03:11.009 --> 00:03:13.629 do move, obviously. So the guide moves on to 00:03:13.629 --> 00:03:15.550 directing the eye, and it brings up a technique 00:03:15.550 --> 00:03:18.430 called rack focus. I know what that looks like, 00:03:18.430 --> 00:03:20.969 but how do you translate that to a prompt? So 00:03:20.969 --> 00:03:23.009 rack focus is basically the director pointing 00:03:23.009 --> 00:03:25.289 and saying, look here. OK, now look there. You're 00:03:25.289 --> 00:03:27.469 just shifting the sharpness from the foreground 00:03:27.469 --> 00:03:30.550 to the background. And in AI, that's a huge flex, 00:03:30.569 --> 00:03:33.330 because it implies real 3D depth. Can you give 00:03:33.330 --> 00:03:35.789 us the visual from the guide? Yeah. So picture 00:03:35.789 --> 00:03:40.599 a jazz bar. Moody, kind of noir. Vibe. In the 00:03:40.599 --> 00:03:42.620 front, you've got a glass of whiskey. The ice 00:03:42.620 --> 00:03:44.800 is melting. It's perfectly sharp. In the background, 00:03:44.860 --> 00:03:47.080 all blurry. There's a woman in a red dress looking 00:03:47.080 --> 00:03:49.719 out a window. A classic Noir setup. Total classic. 00:03:50.099 --> 00:03:52.319 But the movement isn't the camera flying at her. 00:03:52.460 --> 00:03:54.840 The movement is the lens itself changing focus. 00:03:55.360 --> 00:03:58.520 You tell Kling AI to rack focus to background. 00:03:58.960 --> 00:04:01.120 The glass goes blurry, and suddenly the woman 00:04:01.120 --> 00:04:03.680 is sharp. It creates a story with no dialogue 00:04:03.680 --> 00:04:07.319 at all. Here's the drink, and there she is, leaving. 00:04:07.710 --> 00:04:10.250 So is the camera merely recording or is it acting 00:04:10.250 --> 00:04:12.689 as a narrator? It's a narrator physically pointing 00:04:12.689 --> 00:04:16.170 at what matters most. This idea of control brings 00:04:16.170 --> 00:04:18.709 us to what might be the biggest frustration with 00:04:18.709 --> 00:04:21.370 AI video. I think everyone listening has felt 00:04:21.370 --> 00:04:24.889 this. You generate a character, say a cyberpunk 00:04:24.889 --> 00:04:27.509 mercenary. She looks great in shot one. You generate 00:04:27.509 --> 00:04:29.949 shot two and suddenly she has a different nose 00:04:29.949 --> 00:04:32.370 or her hair is shorter. Oh, the identity drift. 00:04:32.370 --> 00:04:34.550 It just breaks the illusion instantly. Right. 00:04:34.709 --> 00:04:37.899 The face changes and the audience is gone. The 00:04:37.899 --> 00:04:39.500 guide says we've been doing this all backward. 00:04:39.980 --> 00:04:41.879 We usually try to generate the character inside 00:04:41.879 --> 00:04:44.100 the video tool. Which is a huge mistake. The 00:04:44.100 --> 00:04:46.160 video generator is already trying to calculate 00:04:46.160 --> 00:04:50.199 motion and lighting and physics. It's just too 00:04:50.199 --> 00:04:52.480 much for the genius child to handle. You need 00:04:52.480 --> 00:04:54.939 a virtual actor. And this is where that specific 00:04:54.939 --> 00:04:58.360 tool comes in. Nanobanana. I know the names in 00:04:58.360 --> 00:05:01.259 2026 are just ridiculous, but nanobanana is kind 00:05:01.259 --> 00:05:03.540 of the industry standard right now for good reason 00:05:03.540 --> 00:05:05.939 It just yeah, it listens better The strategy 00:05:05.939 --> 00:05:08.319 is you don't start with a video you start with 00:05:08.319 --> 00:05:11.180 a character sheet like a casting photo Exactly 00:05:11.180 --> 00:05:14.220 like a casting photo you go into nanobanana not 00:05:14.220 --> 00:05:18.139 cling not vo and you create one master image 00:05:18.139 --> 00:05:20.660 Front view full body. Let's stick with that cyberpunk 00:05:20.660 --> 00:05:24.759 mercenary silver hair purple streaks, jacket 00:05:24.759 --> 00:05:27.480 with neon circuits. You get that image perfect. 00:05:27.639 --> 00:05:29.240 And then you take that image somewhere else. 00:05:29.420 --> 00:05:31.899 Then and only then do you go to your video tool. 00:05:32.060 --> 00:05:33.980 You upload that image and you say, this is my 00:05:33.980 --> 00:05:36.319 reference, do not change her. You're anchoring 00:05:36.319 --> 00:05:40.079 her identity. Does creating a static soul for 00:05:40.079 --> 00:05:42.939 the character make the AI -generated movement 00:05:42.939 --> 00:05:46.199 more believable? Yes, because consistency creates 00:05:46.199 --> 00:05:49.220 the illusion of a continuous life. OK, character 00:05:49.220 --> 00:05:52.279 locked. focus directed, now we actually need 00:05:52.279 --> 00:05:54.639 to move the camera. The guide makes this huge 00:05:54.639 --> 00:05:57.540 distinction between zoom and dolly. They might 00:05:57.540 --> 00:05:59.439 sound the same, but they feel completely different, 00:05:59.660 --> 00:06:01.060 don't they? Oh, completely different physics. 00:06:01.279 --> 00:06:03.439 A zoom is just optical. You're making the image 00:06:03.439 --> 00:06:05.699 bigger. It kind of flattens everything. A dolly 00:06:05.699 --> 00:06:07.720 is when you physically move the camera body through 00:06:07.720 --> 00:06:09.379 the space. So you're actually walking forward. 00:06:09.519 --> 00:06:11.519 You're walking forward. And because the camera 00:06:11.519 --> 00:06:14.480 itself is moving, the relationship between the 00:06:14.480 --> 00:06:16.860 foreground and background changes. That's what 00:06:16.860 --> 00:06:19.540 our brains read as real. The guide also mentions 00:06:19.560 --> 00:06:22.500 The vertigo effect, the dolly zoom, that classic 00:06:22.500 --> 00:06:25.600 Hitchcock shot. The Jaws shot. It's when you 00:06:25.600 --> 00:06:28.139 dolly the camera backwards while you zoom the 00:06:28.139 --> 00:06:31.420 lens in. That sounds incredibly complicated. 00:06:31.680 --> 00:06:34.259 It looks like a headache. It creates this warping 00:06:34.259 --> 00:06:36.160 effect. If you do it on a character in an alley, 00:06:36.220 --> 00:06:38.339 they stay the same size, but the background behind 00:06:38.339 --> 00:06:40.639 them just stretches and gets all weird. It's 00:06:40.639 --> 00:06:43.160 instant anxiety. How does the physics of the 00:06:43.160 --> 00:06:45.839 camera lens manipulate the viewer's psychological 00:06:45.839 --> 00:06:48.319 state? It changes our spatial relationship to 00:06:48.319 --> 00:06:51.939 the subject, creating unease or closeness. It's 00:06:51.939 --> 00:06:54.579 just wild that we have to explain these physical 00:06:54.579 --> 00:06:57.480 ideas to a digital brain. We have to simulate 00:06:57.480 --> 00:06:59.899 the glass. We do. And that also applies to moving 00:06:59.899 --> 00:07:02.439 sideways. The guide brings up pan versus truck. 00:07:02.620 --> 00:07:04.839 Trucking, another one of those industrial sounding 00:07:04.839 --> 00:07:07.379 terms. Well, it comes from putting the camera 00:07:07.379 --> 00:07:10.839 on a truck or a track. A pan is just standing 00:07:10.839 --> 00:07:14.079 still and turning your head, left, right. The 00:07:14.079 --> 00:07:16.779 problem is, in AI video, that often makes the 00:07:16.779 --> 00:07:18.959 background look like flat wallpaper. Right, there's 00:07:18.959 --> 00:07:22.240 no depth. And trucking fixes that. Trucking is 00:07:22.240 --> 00:07:25.430 sliding the camera sideways. Imagine you're running 00:07:25.430 --> 00:07:27.750 right alongside a horse. Okay, I'm thinking of 00:07:27.750 --> 00:07:30.129 the example from The Samurai riding through a 00:07:30.129 --> 00:07:33.089 bamboo forest. Perfect example. If you pan that 00:07:33.089 --> 00:07:35.709 scene, it looks flat. But if you truck rate, 00:07:35.949 --> 00:07:38.490 moving parallel to the horse, you get the parallax 00:07:38.490 --> 00:07:40.990 effect. Parallax. Can you break that down? It's 00:07:40.990 --> 00:07:43.250 simple physics, really. The things that are close 00:07:43.250 --> 00:07:46.089 to the camera seem to move fast. So the bamboo 00:07:46.089 --> 00:07:48.230 trees in the foreground are just whizzing by, 00:07:48.389 --> 00:07:50.850 all blurred out. But the mountains way in the 00:07:50.850 --> 00:07:53.810 background are moving really slowly. That difference 00:07:53.810 --> 00:07:56.389 in speed tells your brain, OK, this is a 3D world. 00:07:56.410 --> 00:07:58.949 This is real. Why is the parallax effect the 00:07:58.949 --> 00:08:00.949 dividing line between amateur and professional 00:08:00.949 --> 00:08:03.470 visuals? It proves the scene exists in three 00:08:03.470 --> 00:08:05.860 dimensions, not just two. I want to touch on 00:08:05.860 --> 00:08:08.379 something that usually just destroys AI video, 00:08:08.579 --> 00:08:11.220 even if you're trucking correctly. Complex actions. 00:08:11.319 --> 00:08:13.199 We've all seen it. Someone tried to generate 00:08:13.199 --> 00:08:15.860 a guy jumping off a cliff and halfway down his 00:08:15.860 --> 00:08:18.079 legs turned into spaghetti. The hallucination 00:08:18.079 --> 00:08:21.860 phase, yeah. The AI just loses track of anatomy. 00:08:22.139 --> 00:08:24.379 So how do you direct something that complex? 00:08:24.670 --> 00:08:27.870 The guide suggests a multi -frame strategy. It's 00:08:27.870 --> 00:08:30.389 brilliant because it stops treating the AI like 00:08:30.389 --> 00:08:32.809 a magician and starts treating it like an animator. 00:08:33.269 --> 00:08:35.190 Instead of just saying man jumps off cliff, which 00:08:35.190 --> 00:08:38.409 is way too big, you give it guardrails. A starting 00:08:38.409 --> 00:08:40.750 point and an end point. Exactly. You generate 00:08:40.750 --> 00:08:43.549 keyframe A, the man standing on the edge. Then 00:08:43.549 --> 00:08:46.909 you generate keyframe B, the man mid -dive, arms 00:08:46.909 --> 00:08:49.450 spread out, jacket flapping in the wind. You 00:08:49.450 --> 00:08:51.649 feed both of those images into the video generator. 00:08:51.950 --> 00:08:55.330 And the AI just... fills in the middle. It interpolates 00:08:55.330 --> 00:08:57.870 the path, but because it knows where it has to 00:08:57.870 --> 00:09:00.169 end up, it doesn't get lost. It calculates the 00:09:00.169 --> 00:09:02.330 gravity of the flapping jacket, all of it. It's 00:09:02.330 --> 00:09:04.269 connecting the docks instead of just guessing 00:09:04.269 --> 00:09:06.769 where to go. So are we directing the AI or are 00:09:06.769 --> 00:09:09.909 we simply setting boundaries to contain its hallucinations? 00:09:10.309 --> 00:09:12.470 We are building guardrails so its imagination 00:09:12.470 --> 00:09:14.610 doesn't braid physics. Speaking of boundaries, 00:09:14.990 --> 00:09:17.649 there's one more glitch in the guide that I found 00:09:17.649 --> 00:09:23.000 both hilarious and kind of terrifying. The second 00:09:23.000 --> 00:09:26.000 face problem. Oh, the nightmare fuel. When the 00:09:26.000 --> 00:09:28.000 camera does an orbit shot, it circles around 00:09:28.000 --> 00:09:30.299 a character, and when it gets to their back, 00:09:31.539 --> 00:09:33.120 there's another face on the back of their head. 00:09:33.299 --> 00:09:35.799 It happens because the AI gets confused. It just 00:09:35.799 --> 00:09:38.440 knows character equals face. So when it runs 00:09:38.440 --> 00:09:40.679 out of face to show, it panics and just puts 00:09:40.679 --> 00:09:42.500 another one on the back of their skull. How on 00:09:42.500 --> 00:09:45.549 earth do you stop that? You need landmarks. You 00:09:45.549 --> 00:09:48.190 have to give the AI something specific to look 00:09:48.190 --> 00:09:50.750 at on the character's back. The example is a 00:09:50.750 --> 00:09:52.789 wizard on a mountain. If you just say, orbit 00:09:52.789 --> 00:09:55.789 the wizard, you might get the second face. But 00:09:55.789 --> 00:09:58.350 if you specify, the wizard wears a cape with 00:09:58.350 --> 00:10:01.450 a golden dragon emblem on the back, now the AI 00:10:01.450 --> 00:10:04.120 has a mission. It has a target. Right. It thinks, 00:10:04.299 --> 00:10:06.419 OK, I'm circling. I need to find that dragon 00:10:06.419 --> 00:10:09.139 emblem. It anchors the whole image on that object, 00:10:09.360 --> 00:10:11.659 not the anatomy. That's such a clever workaround. 00:10:11.740 --> 00:10:13.600 You're just guiding its attention so it doesn't 00:10:13.600 --> 00:10:15.960 have to guess. I want to take a quick pause here. 00:10:16.559 --> 00:10:18.779 We've been talking about the how, the trucking, 00:10:18.879 --> 00:10:21.200 the landmarks, the static shots. But in a moment, 00:10:21.480 --> 00:10:24.700 I want to shift to the what. The actual tools 00:10:24.700 --> 00:10:27.220 available to us in 2026, because the landscape 00:10:27.220 --> 00:10:30.039 has really changed. We're back. We're deep diving 00:10:30.039 --> 00:10:33.460 into cinematic AI filmmaking in January, 2026. 00:10:33.879 --> 00:10:35.720 We've covered the techniques, but let's get practical. 00:10:36.259 --> 00:10:38.399 There are a million tools out there. Subscription 00:10:38.399 --> 00:10:40.480 fatigue is a real thing. I don't want to pay 00:10:40.480 --> 00:10:42.740 for five different services if I don't have to. 00:10:42.860 --> 00:10:45.620 It's very real. And the guide is pretty blunt 00:10:45.620 --> 00:10:49.600 about this. The era of the generalist AI is kind 00:10:49.600 --> 00:10:52.039 of fading. So the one app to do it all dream 00:10:52.039 --> 00:10:54.720 is over. For high end work, yeah. We have a specialized 00:10:54.720 --> 00:10:57.789 tool chain now. Just like a real film crew has 00:10:57.789 --> 00:10:59.389 different departments, you need different AIs 00:10:59.389 --> 00:11:02.090 for different jobs. The guide breaks down four 00:11:02.090 --> 00:11:04.409 main players. Okay, let's run through them. We've 00:11:04.409 --> 00:11:06.110 already mentioned the first one. Nano Banana. 00:11:06.379 --> 00:11:08.720 The foundation. It's on the open art platform. 00:11:08.940 --> 00:11:12.320 This is not for making video. It is purely for 00:11:12.320 --> 00:11:15.279 creating your source images, your actors. Its 00:11:15.279 --> 00:11:18.220 whole superpower is realistic skin and just listening 00:11:18.220 --> 00:11:21.000 to your prompt. If you skip this step, your virtual 00:11:21.000 --> 00:11:23.360 actor will look like plastic. Garbage in, garbage 00:11:23.360 --> 00:11:26.320 out. Exactly. Okay, next up is the budget king, 00:11:26.759 --> 00:11:30.539 Google VO3. VO3. That's the landscape engine. 00:11:30.860 --> 00:11:33.779 Yeah, it's amazing for B -roll. 4K landscapes, 00:11:34.139 --> 00:11:36.940 drone shots, establishing shots of forests. It 00:11:36.940 --> 00:11:39.059 understands physics really well, and, you know, 00:11:39.059 --> 00:11:41.379 it's usually free or very low cost. Right. But 00:11:41.379 --> 00:11:43.580 it struggles with complex human acting. It's 00:11:43.580 --> 00:11:45.299 a little stiff, so you use it for the world, 00:11:45.379 --> 00:11:47.019 not for the people in it. OK, so for the people, 00:11:47.039 --> 00:11:49.320 for the acting, where do we go? You go to the 00:11:49.320 --> 00:11:51.720 storyteller, Kling AI. This is the paid tool, 00:11:51.980 --> 00:11:53.639 the heavy hitter. It's the king of movement. 00:11:53.899 --> 00:11:56.960 When you type truck right or dolly in, Kling 00:11:56.960 --> 00:11:58.840 actually knows what those words mean physically. 00:11:59.100 --> 00:12:01.259 It's for the serious filmmaker. And the last 00:12:01.259 --> 00:12:05.539 one sounds a bit chaotic. Yeah. Seance 1 .5 Pro. 00:12:05.860 --> 00:12:08.860 The action star. This thing is huge on TikTok 00:12:08.860 --> 00:12:12.279 for anime styles. If you want explosions, magic 00:12:12.279 --> 00:12:15.600 spells, fast cuts, wild stuff, Seance is your 00:12:15.600 --> 00:12:18.330 tool. It's just pure high energy. But you probably 00:12:18.330 --> 00:12:20.309 wouldn't use it for our old fisherman portrait. 00:12:20.509 --> 00:12:22.950 Oh, god, no. If you ask sedents for a fisherman 00:12:22.950 --> 00:12:24.909 smoking a pipe, the pipe would probably turn 00:12:24.909 --> 00:12:27.409 into a dragon and just fly away. It makes ADHD 00:12:27.409 --> 00:12:29.450 video. It's too caffeinated for real emotion. 00:12:29.870 --> 00:12:32.669 So does the specialization of these tools suggest 00:12:32.669 --> 00:12:36.309 we're moving away from all -in -one AI models? 00:12:36.750 --> 00:12:38.909 Absolutely. Specialized tools are now required 00:12:38.909 --> 00:12:42.009 for specific cinematic tasks. That brings us 00:12:42.009 --> 00:12:44.809 to the final lesson from the guide, the golden 00:12:44.809 --> 00:12:47.360 rule. We touched on it. But it's really about 00:12:47.360 --> 00:12:49.840 the language we use. The megaprompt. Right. The 00:12:49.840 --> 00:12:52.720 days of typing a girl walking are just over. 00:12:52.879 --> 00:12:54.860 That gets you nowhere now. If you type a girl 00:12:54.860 --> 00:12:57.320 walking, you get the average of every video on 00:12:57.320 --> 00:13:00.340 the internet, which is boring. The guide gives 00:13:00.340 --> 00:13:03.139 us a specific formula. The prompt is now your 00:13:03.139 --> 00:13:05.039 script, your set design, your lighting crew, 00:13:05.179 --> 00:13:07.159 all of it. Now what's the formula? It's movement 00:13:07.159 --> 00:13:09.679 type plus a detailed subject plus the environment 00:13:09.679 --> 00:13:12.320 plus camera gear plus lighting. Give us an example. 00:13:12.440 --> 00:13:14.259 The bad version versus the good version. OK. 00:13:14.570 --> 00:13:17.789 Bad version. Cyberpunk girl walking in rain. 00:13:17.909 --> 00:13:20.669 And that gets ya. A generic, kinda boring cartoon. 00:13:20.929 --> 00:13:23.669 Now, the good version. Tracking shot. Camera 00:13:23.669 --> 00:13:26.970 trucks right. A cyberpunk girl with neon glowing 00:13:26.970 --> 00:13:29.809 skin walking through a rainy alleyway. Reflections 00:13:29.809 --> 00:13:33.289 on wet pavement. Transparent plastic coat. Shot 00:13:33.289 --> 00:13:37.549 on Sony Venice 2. 35mm prime lens. F1 .4 aperture. 00:13:38.029 --> 00:13:40.960 Teal and orange palette. Wow. You're literally 00:13:40.960 --> 00:13:43.500 specifying the aperture, f1 .4. You have to. 00:13:43.840 --> 00:13:47.759 Telling the AI f1 .4 is code for I want a really 00:13:47.759 --> 00:13:49.779 blurry background and a sharp subject. If you 00:13:49.779 --> 00:13:52.200 don't say that, it defaults to a flat image where 00:13:52.200 --> 00:13:54.539 everything is in focus, like a cheap TV show. 00:13:54.919 --> 00:13:57.100 AI always guesses boring unless you force it 00:13:57.100 --> 00:13:59.039 to be interesting. Is the role of the director 00:13:59.039 --> 00:14:01.840 shifting from managing people to managing language? 00:14:01.940 --> 00:14:04.740 Yes. The prompt is now the script, the set, and 00:14:04.740 --> 00:14:06.860 the crew. I still kind of wrestle with that idea 00:14:06.860 --> 00:14:08.940 myself, that the words themselves are the lens. 00:14:09.610 --> 00:14:11.110 Let's bring this all back. We started with the 00:14:11.110 --> 00:14:13.350 genius child. We moved through stillness, depth, 00:14:13.669 --> 00:14:15.789 consistency. What's the big takeaway for someone 00:14:15.789 --> 00:14:17.789 who's about to open their laptop and try this? 00:14:18.070 --> 00:14:20.450 The big idea is just intentionality. Stop letting 00:14:20.450 --> 00:14:23.950 the AI drive. There is no make cool movie button. 00:14:24.269 --> 00:14:27.429 One, start with a static shot. Master stillness 00:14:27.429 --> 00:14:31.269 first. Two, use physical camera terms like truck 00:14:31.269 --> 00:14:33.950 and dolla to create that 3D space. And three, 00:14:34.570 --> 00:14:36.389 lock your character's identity with a reference 00:14:36.389 --> 00:14:39.679 image. It's really about craftsmanship. Even 00:14:39.679 --> 00:14:42.360 in this automated world, craft still matters. 00:14:42.679 --> 00:14:45.480 Maybe more than ever. Because when everyone can 00:14:45.480 --> 00:14:47.059 generate a video, the only thing that's going 00:14:47.059 --> 00:14:49.299 to stand out is taste. Can you tell a good story? 00:14:49.419 --> 00:14:51.600 Taste is the differentiator. I like that. And 00:14:51.600 --> 00:14:53.200 so here's my challenge for everyone listening. 00:14:53.740 --> 00:14:56.220 Open NanoBanana. Open whatever tool you use. 00:14:56.379 --> 00:14:58.559 Don't try to make an epic battle. Try to make 00:14:58.559 --> 00:15:00.659 that fisherman. Try to make a portrait that's 00:15:00.659 --> 00:15:03.139 perfectly still, except for one tiny thing, a 00:15:03.139 --> 00:15:05.779 puff of smoke, a blink. Just master the stillness 00:15:05.779 --> 00:15:07.840 first. That's a great place to start. And I'll 00:15:07.840 --> 00:15:09.600 add one final thought from the guide that I found 00:15:09.600 --> 00:15:12.059 really comforting. It said, don't feel bad if 00:15:12.059 --> 00:15:14.659 your first videos are not perfect. Amen to that. 00:15:14.840 --> 00:15:18.139 It's 2026. This tech is moving so fast, but it 00:15:18.139 --> 00:15:20.480 is still an art form. And you have to be vulnerable 00:15:20.480 --> 00:15:23.000 enough to make bad art before you can make good 00:15:23.000 --> 00:15:25.549 art. The first draft of anything is garbage. 00:15:25.710 --> 00:15:28.529 Just keep prompting. Thanks for diving in with 00:15:28.529 --> 00:15:30.990 us today. Go direct some masterpieces. We'll 00:15:30.990 --> 00:15:31.490 see you in the next one.