WEBVTT 00:00:00.000 --> 00:00:03.720 You know that feeling? That specific kind of 00:00:03.720 --> 00:00:06.660 panic. Oh, yeah. It's 11 .0 p .m. The house is 00:00:06.660 --> 00:00:09.119 totally quiet. You've got a video edit due, say, 00:00:09.240 --> 00:00:13.080 9 .00 a .m. sharp. The deadline. Right. And the 00:00:13.080 --> 00:00:14.679 pacing is tight. The color grade looks good. 00:00:14.740 --> 00:00:19.079 The audio is mixed. But then you hit a gap. You're 00:00:19.079 --> 00:00:21.760 missing the hero shot. Yeah. And not just the 00:00:21.760 --> 00:00:24.160 small D -roll clip, but the one transition that 00:00:24.160 --> 00:00:25.960 makes the whole thing work. The linchpin. So 00:00:25.960 --> 00:00:28.699 you check the stock libraries. Nothing. You check 00:00:28.699 --> 00:00:31.739 your hard drive. Still nothing. And suddenly 00:00:31.739 --> 00:00:33.820 you're calculating if you can hire a helicopter 00:00:33.820 --> 00:00:36.960 crew at midnight. Which, spoiler alert, you can't. 00:00:36.960 --> 00:00:38.780 That is usually the moment you just compromise. 00:00:38.820 --> 00:00:41.000 You slap in something mediocre and hope the client 00:00:41.000 --> 00:00:43.140 doesn't notice. Exactly. But we are not doing 00:00:43.140 --> 00:00:45.179 that today. We're looking at option four. Option 00:00:45.179 --> 00:00:47.939 number four. Welcome back to Deep Dive. Today 00:00:47.939 --> 00:00:50.079 we're looking at, well, the state of AI filmmaking 00:00:50.079 --> 00:00:53.659 in early 2026. Yeah. We've got a guide here by 00:00:53.659 --> 00:00:56.539 Max Ann titled, How to Generate Pro -Level B 00:00:56.539 --> 00:01:01.060 -Roll Footage on Demand in 2026. And what stands 00:01:01.060 --> 00:01:04.260 out to me right away is the shift in tone. We're 00:01:04.260 --> 00:01:07.120 not talking about those glitchy, you know, nightmare 00:01:07.120 --> 00:01:09.579 fuel videos from a couple of years ago. We are 00:01:09.579 --> 00:01:12.090 talking about precision. We really are. We're 00:01:12.090 --> 00:01:14.629 moving past the novelty phase as this isn't about 00:01:14.629 --> 00:01:17.590 typing cat riding a skateboard anymore. This 00:01:17.590 --> 00:01:19.950 is about directing. Directing. Yeah. The guide 00:01:19.950 --> 00:01:23.230 focuses on using the heavy hitters VO 3 .1, Sora 00:01:23.230 --> 00:01:26.629 2, Kling 2 .6 to bypass stock footage entirely. 00:01:27.450 --> 00:01:29.810 But the core argument, and I think this is the 00:01:29.810 --> 00:01:31.670 part that changes the game, is you have to stop 00:01:31.670 --> 00:01:34.849 guessing and start directing. Directing the algorithm. 00:01:35.209 --> 00:01:37.129 I like the sound of that, but I want to unpack 00:01:37.129 --> 00:01:39.150 what it actually means because it can sound a 00:01:39.150 --> 00:01:43.540 bit... For sure. So here's the roadmap. First, 00:01:43.680 --> 00:01:45.640 we need to understand this strategy shift, this 00:01:45.640 --> 00:01:47.480 thing called image -to -video orchestration. 00:01:47.859 --> 00:01:50.560 Then we're going to dissect the director's blueprint, 00:01:50.819 --> 00:01:53.900 the actual anatomy of a prompt that works. We've 00:01:53.900 --> 00:01:57.000 got three case studies that are frankly mind 00:01:57.000 --> 00:01:59.439 -blowing. A train in the Alps, a garden metaphor, 00:01:59.659 --> 00:02:01.959 and a zoom from space. Yeah, that last one is 00:02:01.959 --> 00:02:03.659 something else. And finally, we'll look at the 00:02:03.659 --> 00:02:06.200 personalities of these different AI models. It's 00:02:06.200 --> 00:02:08.740 a dense agenda, but a fun one. Let's start with 00:02:08.740 --> 00:02:11.590 the strategy. The source material makes a pretty 00:02:11.590 --> 00:02:13.969 bold claim right off the bat. It says the old 00:02:13.969 --> 00:02:17.550 way of doing this, text to video, is fundamentally 00:02:17.550 --> 00:02:20.610 broken. It is. It's broken because it relies 00:02:20.610 --> 00:02:23.750 on the machine guessing. Guessing. If you open 00:02:23.750 --> 00:02:27.530 Sora or Vio and type cinematic shot of a woman 00:02:27.530 --> 00:02:30.479 walking down a street. You're leaving a million 00:02:30.479 --> 00:02:33.060 variables just completely undefined. The lighting, 00:02:33.159 --> 00:02:35.680 the lens choice, the texture of the pavement, 00:02:35.879 --> 00:02:38.740 the era of the architecture. Right. The AI has 00:02:38.740 --> 00:02:43.180 to fill in those blanks. And usually it hallucinates. 00:02:43.500 --> 00:02:45.960 Okay. So when you say hallucinates in this context, 00:02:46.180 --> 00:02:48.300 what are we actually seeing on screen? we're 00:02:48.300 --> 00:02:50.240 seeing the melting watch effect the ai starts 00:02:50.240 --> 00:02:52.599 guessing and suddenly the woman has six fingers 00:02:52.599 --> 00:02:54.919 or the buildings in the background start to warp 00:02:54.919 --> 00:02:57.340 like they're made of liquid it's like a dream 00:02:57.340 --> 00:02:59.020 where you look at a clock and the numbers just 00:02:59.020 --> 00:03:02.060 slide off the ai creates a vibe but it fails 00:03:02.060 --> 00:03:03.780 at physics right it feels like a flop machine 00:03:03.780 --> 00:03:06.439 you pull the lever you get garbage you pull it 00:03:06.439 --> 00:03:08.639 again maybe you get something usable exactly 00:03:08.639 --> 00:03:11.159 and that is why the professional workflow of 00:03:11.159 --> 00:03:14.960 2026 is image to video orchestration it creates 00:03:14.960 --> 00:03:17.319 a firewall between the composition and the movement 00:03:17.719 --> 00:03:19.620 Walk me through the mechanics of that. It's a 00:03:19.620 --> 00:03:22.259 two -stage process. You never ask the video model 00:03:22.259 --> 00:03:24.759 to create the world. You ask an image model to 00:03:24.759 --> 00:03:27.419 create the world first. Okay. Stage one is generating 00:03:27.419 --> 00:03:30.539 a high -fidelity starting frame. Okay. The guide 00:03:30.539 --> 00:03:33.259 is very specific here. It recommends a tool called 00:03:33.259 --> 00:03:36.139 Nano Banana Pro. Nano Banana Pro. I know, the 00:03:36.139 --> 00:03:38.039 names in this industry are getting a little ridiculous, 00:03:38.060 --> 00:03:41.180 but under the hood, it's running Google's Gemini 00:03:41.180 --> 00:03:44.219 2 .5 or the 3 Flash model. And there's a reason 00:03:44.219 --> 00:03:46.500 for using that instead of, say, mid -journey 00:03:46.500 --> 00:03:48.719 for this step. There is. It comes down to character 00:03:48.719 --> 00:03:51.599 consistency and resolution. You need to lock 00:03:51.599 --> 00:03:55.439 in a 4K image with 300 dpi clarity. Got it. Gemini 00:03:55.439 --> 00:03:57.939 has shown this incredible ability to follow complex 00:03:57.939 --> 00:04:00.120 spatial instructions without adding artistic 00:04:00.120 --> 00:04:02.840 flair you didn't ask for. You want a blueprint, 00:04:03.060 --> 00:04:05.620 not a painting. Okay, so you've got this perfect 00:04:05.620 --> 00:04:10.099 static 4K image. Yeah. Stage two. Then... And 00:04:10.099 --> 00:04:13.020 only then do you feed that image into Veo or 00:04:13.020 --> 00:04:15.759 Kling to animate it. You are essentially telling 00:04:15.759 --> 00:04:19.060 the video AI, look at this picture. Don't change 00:04:19.060 --> 00:04:21.199 the lighting. Don't change the face. Just make 00:04:21.199 --> 00:04:24.100 the wind blow. You know, I have to admit, I still 00:04:24.100 --> 00:04:27.300 wrestle with this myself. Yeah. With prompt drift, 00:04:27.500 --> 00:04:30.240 as they call it. My early attempts trying to 00:04:30.240 --> 00:04:33.079 do it all in one go. Well, they look less like 00:04:33.079 --> 00:04:35.079 movies and more like those hallucinations we 00:04:35.079 --> 00:04:36.680 were talking about. Melting clocks. The melting 00:04:36.680 --> 00:04:39.180 clocks, exactly. So this two -stage process. 00:04:39.899 --> 00:04:42.680 It sounds like it takes longer to set up, but 00:04:42.680 --> 00:04:45.000 maybe saves hours of frustration. That's the 00:04:45.000 --> 00:04:47.040 idea. You would think it's doubling the work, 00:04:47.120 --> 00:04:49.180 but think about the slot machine problem. You 00:04:49.180 --> 00:04:51.199 might spend two hours re -rolling to get one 00:04:51.199 --> 00:04:53.860 good shot. With this, you spend 15 minutes getting 00:04:53.860 --> 00:04:56.180 the image right, and the video works on the first 00:04:56.180 --> 00:04:59.360 or second try. It's measure twice, cut once. 00:05:00.220 --> 00:05:02.379 Precisely. So just to make sure I fully grasped 00:05:02.379 --> 00:05:04.500 the mechanism here, why is generating the image 00:05:04.500 --> 00:05:07.660 before the video the critical unlock? It locks 00:05:07.660 --> 00:05:10.000 in the composition and resolution first. It just 00:05:10.000 --> 00:05:13.959 bypasses the video AI's tendency to hallucinate 00:05:13.959 --> 00:05:17.259 low -quality details when it's trying to calculate 00:05:17.259 --> 00:05:19.680 motion and pixels at the same time. You're reducing 00:05:19.680 --> 00:05:21.980 the cognitive load on the model. It doesn't have 00:05:21.980 --> 00:05:23.720 to invent the world and move it at the same time. 00:05:23.860 --> 00:05:26.240 Exactly. Okay, let's move to the director's blueprint. 00:05:26.620 --> 00:05:29.800 The source says AI responds to structure, not 00:05:29.800 --> 00:05:32.860 vibes. This is the biggest mistake people make. 00:05:33.079 --> 00:05:36.879 They use words like cool, epic. or emotional, 00:05:37.180 --> 00:05:39.639 the AI has no idea what emotional looks like 00:05:39.639 --> 00:05:42.519 in Kixels. Right. So Max Anne outlines a five 00:05:42.519 --> 00:05:44.800 -part structure that is mandatory if you want 00:05:44.800 --> 00:05:47.160 pro results. Let's run through them. Number one, 00:05:47.199 --> 00:05:49.079 camera specification. You have to tell it the 00:05:49.079 --> 00:05:51.699 angle and distance. Number two is visual composition, 00:05:52.000 --> 00:05:53.899 the subject, and environment. That's the easy 00:05:53.899 --> 00:05:55.899 part. Right, a dog in a park. But then you have 00:05:55.899 --> 00:05:58.240 number three. Technical details. This is where 00:05:58.240 --> 00:06:00.699 you separate the pros from the amateurs. Resolution, 00:06:00.980 --> 00:06:04.399 lens style, depth of field. Number four is motion 00:06:04.399 --> 00:06:07.899 description, what moves. And number five, mood 00:06:07.899 --> 00:06:10.139 and atmosphere. I want to deep dive on number 00:06:10.139 --> 00:06:12.819 three, the technical details. The guide mentions 00:06:12.819 --> 00:06:15.620 specific hardware, like actually naming the camera, 00:06:15.740 --> 00:06:19.019 shot on Sony Venice. Does the AI actually know 00:06:19.019 --> 00:06:21.519 what a Sony Venice is? Oh, it absolutely knows. 00:06:21.720 --> 00:06:23.600 You have to remember, these models were trained 00:06:23.600 --> 00:06:26.209 on the internet. They've analyzed millions of 00:06:26.209 --> 00:06:29.110 frames tagged with Sony Venice or Arri Alexa. 00:06:29.310 --> 00:06:31.709 So it's not just placebo tech. Not at all. When 00:06:31.709 --> 00:06:34.870 you type Sony Venice, you are triggering a specific 00:06:34.870 --> 00:06:37.230 set of weights in the neural network. You're 00:06:37.230 --> 00:06:40.029 telling it, I want high dynamic range. I want 00:06:40.029 --> 00:06:42.110 a specific color science where the highlights 00:06:42.110 --> 00:06:45.490 roll off smoothly rather than clipping into pure 00:06:45.490 --> 00:06:48.040 white. That is fascinating. It's like code switching. 00:06:48.199 --> 00:06:49.759 You're speaking the language of the training 00:06:49.759 --> 00:06:52.579 data. Exactly. If you just say cinematic, the 00:06:52.579 --> 00:06:55.660 AI gives you this generic, high contrast digital 00:06:55.660 --> 00:06:58.740 look, basically a video cut scene. But if you 00:06:58.740 --> 00:07:01.240 specify the camera, you get texture. So what 00:07:01.240 --> 00:07:03.379 happens if I have a great prompt, a train in 00:07:03.379 --> 00:07:05.480 the mountains, but I leave out those camera specs? 00:07:05.740 --> 00:07:08.959 Without the camera or lens specs, the AI defaults 00:07:08.959 --> 00:07:11.300 to a flat digital look. It feels like generic 00:07:11.300 --> 00:07:13.800 stock video instead of a rich cinematic film. 00:07:14.019 --> 00:07:16.379 Which brings us to the fun part. The case studies. 00:07:16.759 --> 00:07:19.360 I want to see this theory in action. Let's start 00:07:19.360 --> 00:07:21.620 with that classic travel shot, the Swiss Alps 00:07:21.620 --> 00:07:24.920 train. The peak wanderlust shot. We've all seen 00:07:24.920 --> 00:07:27.899 it. Red train, stone viaduct, snow -capped mountains. 00:07:28.540 --> 00:07:31.079 But the prompt here is so specific about the 00:07:31.079 --> 00:07:34.540 glass. It is. The prompt calls for a Sony Venice 00:07:34.540 --> 00:07:38.639 with an RE Signature Prime 24mm lens. Why that 00:07:38.639 --> 00:07:41.620 specific lens? Why 24mm? It's about the language 00:07:41.620 --> 00:07:44.879 of cinema. A 24mm lens is wide, but it's not 00:07:44.879 --> 00:07:47.939 a fisheye. It captures scale. It tells the AI 00:07:47.939 --> 00:07:49.980 we want the mountains to feel massive and the 00:07:49.980 --> 00:07:53.379 train to feel small. But more importantly, naming 00:07:53.379 --> 00:07:56.060 the signature prime lens tells the AI to keep 00:07:56.060 --> 00:07:58.560 the image sharp corner to corner, but with a 00:07:58.560 --> 00:08:00.819 slight organic softness so it doesn't look like 00:08:00.819 --> 00:08:03.060 CGI. And because you use the image -to -video 00:08:03.060 --> 00:08:05.079 workflow, you're not just getting a red blur. 00:08:05.300 --> 00:08:07.379 You get the texture of the stone on the bridge. 00:08:07.759 --> 00:08:09.279 The light hitting the paint. Right. And for the 00:08:09.279 --> 00:08:11.420 motion, the prompt is simple. Camera follows 00:08:11.420 --> 00:08:13.819 the train smoothly. Steady aerial tracking shot. 00:08:14.000 --> 00:08:16.040 You don't need to overcomplicate the movement 00:08:16.040 --> 00:08:18.699 if the image is perfect. Okay. Let's pivot to 00:08:18.699 --> 00:08:21.100 the second case study because this one feels 00:08:21.100 --> 00:08:23.379 very different emotionally. The growth metaphor. 00:08:23.660 --> 00:08:26.000 Right. This is for when you need to illustrate 00:08:26.000 --> 00:08:29.180 patience or progress. The visual is weathered 00:08:29.180 --> 00:08:32.000 hands pouring water and a seedling emerging from 00:08:32.000 --> 00:08:34.779 the soil. It feels much more intimate. And the 00:08:34.779 --> 00:08:37.399 hardware choice changes completely. Drastically. 00:08:37.580 --> 00:08:40.960 For this one, the guide recommends an ARI Alexa 00:08:40.960 --> 00:08:44.580 35 with a 50mm lens. Why the change to 50mm? 00:08:44.700 --> 00:08:47.519 The 50mm lens is often called the nifty 50 because 00:08:47.519 --> 00:08:49.779 it roughly mimics the human eye's perspective. 00:08:49.940 --> 00:08:52.899 It feels natural. It feels honest. But the key 00:08:52.899 --> 00:08:56.259 here is the aperture. With a 50mm, you get... 00:08:56.490 --> 00:08:58.669 Boca. Boca. That's the aesthetic blur in the 00:08:58.669 --> 00:09:00.409 background, right? Correct. You want the background, 00:09:00.590 --> 00:09:02.710 the garden, the fence to melt away into soft 00:09:02.710 --> 00:09:05.690 shapes. It isolates the subject. If you use the 00:09:05.690 --> 00:09:07.690 wide angle lens from the train shot here, it 00:09:07.690 --> 00:09:10.149 would just look weird and distorted. The 50 millimeter 00:09:10.149 --> 00:09:12.409 makes it feel like a documentary. There's a detail 00:09:12.409 --> 00:09:14.250 in the source here that I found really interesting. 00:09:14.490 --> 00:09:17.450 It emphasizes weathered hands and a watering 00:09:17.450 --> 00:09:21.350 can with aged patina. Why is that texture so 00:09:21.350 --> 00:09:24.070 important? Because perfection is the enemy of 00:09:24.070 --> 00:09:28.080 realism in AI. If you ask for hands, the AI gives 00:09:28.080 --> 00:09:32.820 you smooth, plastic, mannequin hands. By asking 00:09:32.820 --> 00:09:36.379 for weathered or aged patina, you are forcing 00:09:36.379 --> 00:09:39.840 the AI to render imperfections, scratches, wrinkles, 00:09:40.139 --> 00:09:43.620 dirt. That grit tricks our brain into thinking, 00:09:43.720 --> 00:09:46.379 oh, this is real footage. It's the uncanny valley 00:09:46.379 --> 00:09:48.620 concept. We reject things that are too perfect. 00:09:48.899 --> 00:09:51.700 Exactly. But, and this is a big but, there is 00:09:51.700 --> 00:09:53.960 a warning here regarding hands. Oh, right. AI 00:09:53.960 --> 00:09:56.120 and hands have a... A terrible relationship. 00:09:56.379 --> 00:09:58.940 The finger glitch problem. The guide advises 00:09:58.940 --> 00:10:01.139 keeping the hand movement minimal. The prompt 00:10:01.139 --> 00:10:03.779 says, hands stay mostly still, just tilt the 00:10:03.779 --> 00:10:06.700 watering can. If you ask for complex finger movements, 00:10:06.840 --> 00:10:09.200 the AI is likely to morph the fingers into spaghetti. 00:10:09.460 --> 00:10:11.419 So you have to design the shot around the limitations 00:10:11.419 --> 00:10:13.440 of the tech. You do. Keep it simple is the rule 00:10:13.440 --> 00:10:15.340 for motion. No, we have to talk about the third 00:10:15.340 --> 00:10:17.360 case study. This is the ambitious one, the impossible 00:10:17.360 --> 00:10:19.820 zoom. This one blew my mind. The shot starts 00:10:19.820 --> 00:10:22.440 from a 3D relief map in low -earth orbit, looking 00:10:22.440 --> 00:10:24.549 down at the U .S. Then it zooms off. all the 00:10:24.549 --> 00:10:26.750 way down to Lake Michigan and ends in downtown 00:10:26.750 --> 00:10:29.649 Chicago. Just think about the logistics of filming 00:10:29.649 --> 00:10:32.330 that for real. You'd need satellite imagery, 00:10:32.629 --> 00:10:35.549 high -altitude drone footage, helicopter footage, 00:10:35.809 --> 00:10:40.769 a camera on a crane. It is a $50 ,000 shot, minimum. 00:10:41.090 --> 00:10:42.470 The kind of thing you see in a Marvel movie. 00:10:42.690 --> 00:10:46.149 And here, it's a prompt, but a very complex one. 00:10:46.190 --> 00:10:48.570 It requires prompt logic. You can't just say, 00:10:48.590 --> 00:10:52.289 zoom in on Chicago from space. The AI will get 00:10:52.289 --> 00:10:54.490 lost. I don't know how you do it. You describe 00:10:54.490 --> 00:10:57.710 the layers, hyper -realistic, low Earth orbit 00:10:57.710 --> 00:11:00.529 at dusk. You describe the lighting changes, the 00:11:00.529 --> 00:11:02.990 golden city lights, and the deep blue atmospheric 00:11:02.990 --> 00:11:06.029 haze. You're essentially guiding the AI through 00:11:06.029 --> 00:11:07.690 the layers of the atmosphere. But the source 00:11:07.690 --> 00:11:09.789 mentions this one is hard to pull off. It usually 00:11:09.789 --> 00:11:12.190 fails on the first try. Maxan is very honest 00:11:12.190 --> 00:11:14.169 about that, which I appreciate. You might get 00:11:14.169 --> 00:11:16.269 weird light flashes. The transition from space 00:11:16.269 --> 00:11:18.769 to atmosphere might warp. The buildings might 00:11:18.769 --> 00:11:21.049 look like they are vibrating. So what's the recommended 00:11:21.049 --> 00:11:24.240 fix? If the AI glitches on that transition. You 00:11:24.240 --> 00:11:26.500 iterate. You can either split the shot into stages, 00:11:26.759 --> 00:11:29.159 generate the space part, then the descent, then 00:11:29.159 --> 00:11:31.940 the city, and stitch them together. Or you generate 00:11:31.940 --> 00:11:34.879 multiple versions, pick the best 80 % of each, 00:11:35.039 --> 00:11:37.519 and trim the bad parts. So it's less about one 00:11:37.519 --> 00:11:40.059 perfect generation and more about gathering raw 00:11:40.059 --> 00:11:43.100 material to sculpt. You're still an editor. Exactly. 00:11:43.139 --> 00:11:45.679 You're still a director. The AI is just the camera. 00:11:46.200 --> 00:11:48.419 I want to shift gears slightly to the tools themselves. 00:11:48.820 --> 00:11:52.639 We've mentioned VO, Sora, Kling. The guides suggest 00:11:52.639 --> 00:11:54.700 they have distinct personalities. They really 00:11:54.700 --> 00:11:56.740 do. Just like you'd choose a different cinematographer 00:11:56.740 --> 00:11:59.500 for an action movie versus a period drama, you 00:11:59.500 --> 00:12:01.340 choose your model based on the shot. Let's break 00:12:01.340 --> 00:12:04.960 them down. Start with Google VO 3 .1. Vio is 00:12:04.960 --> 00:12:08.100 the grounded realist. It's the benchmark for 00:12:08.100 --> 00:12:11.059 photorealism and lighting physics. If you're 00:12:11.059 --> 00:12:13.840 doing nature shots, architecture, or that director's 00:12:13.840 --> 00:12:16.039 blueprint environmental stuff where shadows need 00:12:16.039 --> 00:12:19.019 to fall correctly, Vio is the go -to. Okay. What 00:12:19.019 --> 00:12:22.720 about OpenAI's Sora 2? Sora is the artist. Best 00:12:22.720 --> 00:12:25.120 for narrative, surrealism, connecting the dots 00:12:25.120 --> 00:12:28.360 visually. If you have a shot that requires a 00:12:28.360 --> 00:12:30.860 bit of dream logic, like one object morphing 00:12:30.860 --> 00:12:33.759 into another, Sora shines there. It's less rigid 00:12:33.759 --> 00:12:37.460 about physics. And Kling 2 .6. Kling is the action 00:12:37.460 --> 00:12:40.000 star. It's the leader for fast motion. If you're 00:12:40.000 --> 00:12:42.080 making a car chase or something with rapid movement, 00:12:42.379 --> 00:12:44.620 Veo might get blurry, but Kling holds it together. 00:12:44.659 --> 00:12:46.940 It keeps the edges sharp. The guide also has 00:12:46.940 --> 00:12:48.879 a hardware cheat sheet, which I think is worth 00:12:48.879 --> 00:12:51.500 listing out. Definitely. This is the secret sauce. 00:12:51.559 --> 00:12:53.379 So if you're taking notes, write this down. For 00:12:53.379 --> 00:12:56.100 a wide shot, your prompt should say Sony Vaness 00:12:56.100 --> 00:12:59.399 plus 24 millimeter lens. For a portrait, ARI 00:12:59.399 --> 00:13:02.740 Alexa 35 plus a Cooke S4 50 millimeter lens. 00:13:03.039 --> 00:13:05.879 For an intimate close -up, use a red Komodo plus 00:13:05.879 --> 00:13:08.080 an 85 millimeter lens. And for a documentary 00:13:08.080 --> 00:13:12.120 look, Canon C300 plus a 35 millimeter lens. That's 00:13:12.120 --> 00:13:14.899 incredibly specific. Cooke S4 lens. I love that 00:13:14.899 --> 00:13:17.220 the AI knows what that glass looks like. It does. 00:13:17.360 --> 00:13:19.779 A Cooke lens has what cinematographers call... 00:13:19.820 --> 00:13:22.220 a cook look. It's warm, forgiving on skin tones, 00:13:22.379 --> 00:13:25.220 a vintage feel. The AI replicates that warmth. 00:13:25.360 --> 00:13:27.299 If you ask for a Zeiss lens, it would give you 00:13:27.299 --> 00:13:29.580 something cooler and sharper. So just to clarify 00:13:29.580 --> 00:13:32.980 on the model choice, why does Vio get the nod 00:13:32.980 --> 00:13:36.080 for those specific environmental shots in the 00:13:36.080 --> 00:13:39.080 director's blueprint over the others? Vio excels 00:13:39.080 --> 00:13:41.080 at environmental coherence and lighting physics. 00:13:41.379 --> 00:13:44.360 It's just less likely to produce that dream logic 00:13:44.360 --> 00:13:46.659 or weird artifacts in scenes that are supposed 00:13:46.659 --> 00:13:48.860 to look like the real world. It keeps you grounded. 00:13:49.319 --> 00:13:52.279 Exactly. We are back. We have covered the strategy, 00:13:52.500 --> 00:13:55.259 the blueprint, the case studies, and the tools. 00:13:55.399 --> 00:13:57.940 I want to try to synthesize this into a big idea. 00:13:58.159 --> 00:14:00.559 Let's do it. It seems to me that the core philosophy 00:14:00.559 --> 00:14:02.820 here, the unfair advantage that Maxanne talks 00:14:02.820 --> 00:14:05.769 about, isn't actually the tool itself. Everyone 00:14:05.769 --> 00:14:08.490 has access to VO or cling. Right. The tool is 00:14:08.490 --> 00:14:11.029 democratized. The advantage is the move from 00:14:11.029 --> 00:14:13.629 describing a scene to directing it. That is it. 00:14:13.710 --> 00:14:15.769 That is the whole ballgame. It is the difference 00:14:15.769 --> 00:14:18.309 between saying, I want a burger, and telling 00:14:18.309 --> 00:14:20.269 the chef exactly how you want the meat ground, 00:14:20.529 --> 00:14:23.309 the bun toasted, and the sauce layered. And that's 00:14:23.309 --> 00:14:26.129 why the workflow is so rigid. Image first using 00:14:26.129 --> 00:14:27.990 nano banana per, I still can't believe I'm saying 00:14:27.990 --> 00:14:30.750 that name seriously, to lock in the pixels. Then 00:14:30.750 --> 00:14:35.639 motion using VO or cling. And always, always 00:14:35.639 --> 00:14:38.639 use specific camera references to force the AI 00:14:38.639 --> 00:14:41.679 out of generic mode. It's interesting. Usually 00:14:41.679 --> 00:14:44.360 we think of AI as this thing that gives us infinite 00:14:44.360 --> 00:14:47.240 freedom. But here, the argument is that freedom 00:14:47.240 --> 00:14:51.080 leads to mediocrity. Constraints lead to excellence. 00:14:51.519 --> 00:14:54.419 AI video tools close the gap between imagination 00:14:54.419 --> 00:14:57.039 and execution, but they are only as good as the 00:14:57.039 --> 00:14:59.179 structure you provide. That's the quote from 00:14:59.179 --> 00:15:01.840 the guy that sticks with me. For the listener, 00:15:01.919 --> 00:15:04.179 the person who maybe has a presentation next 00:15:04.179 --> 00:15:06.960 week or just wants to play around, what is the 00:15:06.960 --> 00:15:09.440 immediate next step? I would encourage you to 00:15:09.440 --> 00:15:12.360 try one impossible shot this week. Don't just 00:15:12.360 --> 00:15:14.899 make a cat video. Open up Gemini. Type in Sony 00:15:14.899 --> 00:15:17.240 Vaness. Try the growth metaphor. Write down the 00:15:17.240 --> 00:15:20.100 prompt. Camera, composition, tech details, motion, 00:15:20.299 --> 00:15:22.820 mood. Exactly. Save that cheat sheet. See if 00:15:22.820 --> 00:15:24.360 you can create something that looks like it costs 00:15:24.360 --> 00:15:26.580 $10 ,000 while sitting on your couch just to 00:15:26.580 --> 00:15:29.240 see if you can. I love that. We're entering an 00:15:29.240 --> 00:15:32.330 era. where budget is no longer a barrier to visual 00:15:32.330 --> 00:15:35.470 storytelling. The only barrier left is the clarity 00:15:35.470 --> 00:15:38.730 of your vision. Can you see it clearly enough 00:15:38.730 --> 00:15:41.070 to describe it in the language the machine understands? 00:15:41.350 --> 00:15:43.330 That is the question. Thank you for diving deep 00:15:43.330 --> 00:15:45.129 with us today. Go direct some algorithms. See 00:15:45.129 --> 00:15:45.909 you next time. Take care.