WEBVTT 00:00:00.000 --> 00:00:03.220 It is truly mind -bending to think about. Beat. 00:00:03.899 --> 00:00:07.320 Just months ago, cinematic AI video was brutally 00:00:07.320 --> 00:00:09.759 difficult. Oh, absolutely. You faced endless 00:00:09.759 --> 00:00:11.800 rendering failures back then. Right. I mean, 00:00:12.019 --> 00:00:14.640 it required extremely complex software. Rendering 00:00:14.640 --> 00:00:16.980 just failed constantly. Yeah, that process of 00:00:16.980 --> 00:00:19.519 building the final video files was brutal. But 00:00:19.519 --> 00:00:23.179 today, everything has completely changed. Generating 00:00:23.179 --> 00:00:25.780 a scene simply requires typing a few words into 00:00:25.780 --> 00:00:30.160 a chat box. Two secs silence. Welcome to this 00:00:30.160 --> 00:00:33.119 deep dive. Today we are figuring out how to master 00:00:33.119 --> 00:00:36.200 Gemini Omni. It is a massive shift. We are going 00:00:36.200 --> 00:00:38.259 to walk you through setting up your digital avatar. 00:00:38.399 --> 00:00:40.259 Yeah, without it looking creepy, we will also 00:00:40.259 --> 00:00:42.880 figure out how to construct the perfect prompt. 00:00:43.039 --> 00:00:44.840 Right, and we will explore chat -based scene 00:00:44.840 --> 00:00:47.079 editing too. We will look at blending your real 00:00:47.079 --> 00:00:49.280 iPhone footage with Hollywood -level effects. 00:00:49.539 --> 00:00:51.799 Finally, we will uncover the structural secrets 00:00:51.799 --> 00:00:53.939 to avoiding those frustrating rendering errors. 00:00:54.140 --> 00:00:56.539 It is a totally new way to approach video creation. 00:00:56.780 --> 00:00:59.579 What is fascinating here is the core technology. 00:01:00.039 --> 00:01:02.960 Gemini Omni is a multi -input model. OK. Let 00:01:02.960 --> 00:01:07.180 us unpack this quickly. Multi -input model, an 00:01:07.180 --> 00:01:10.840 AI that processes text, video, and audio together. 00:01:11.120 --> 00:01:14.060 Exactly. That completely changes the entire production 00:01:14.060 --> 00:01:16.140 workflow. You are not just feeding it text anymore. 00:01:16.299 --> 00:01:18.980 Right. You are collaborating with it across multiple 00:01:18.980 --> 00:01:21.879 formats simultaneously. Yeah. But before we can 00:01:21.879 --> 00:01:24.000 generate a movie, we have to put ourselves in 00:01:24.000 --> 00:01:28.340 it. Yes. But rushing the initial scan ruins the 00:01:28.340 --> 00:01:31.359 entire output. Many people make this mistake 00:01:31.359 --> 00:01:33.680 right out of the gate. They really do. They end 00:01:33.680 --> 00:01:36.299 up with a weird fake -looking digital character. 00:01:36.560 --> 00:01:39.200 The initial setup decides the quality of every 00:01:39.200 --> 00:01:41.760 future video you make. Right. If you give the 00:01:41.760 --> 00:01:45.060 system bad data on day one, you get bad videos 00:01:45.060 --> 00:01:47.140 forever. There is a catch with the setup, though. 00:01:47.200 --> 00:01:48.859 You actually have to use your mobile phone for 00:01:48.859 --> 00:01:50.900 this. You cannot do this on a desktop computer. 00:01:51.019 --> 00:01:52.859 Yeah, your phone has the necessary hardware built 00:01:52.859 --> 00:01:55.239 right in. It requires the front -facing camera 00:01:55.239 --> 00:01:57.459 and the depth sensors to work. Just to clarify 00:01:57.459 --> 00:02:00.400 for everyone, depth sensors. Phone cameras that 00:02:00.400 --> 00:02:03.439 measure how far physical objects are. They analyze 00:02:03.439 --> 00:02:06.480 your facial data accurately. Exactly. Desktop 00:02:06.480 --> 00:02:09.539 webcams just do not have that 3D mapping capability. 00:02:10.039 --> 00:02:12.460 They only see a flat, two -dimensional image. 00:02:12.659 --> 00:02:15.259 So you open the app on your phone, you see numbers 00:02:15.259 --> 00:02:17.219 on the screen, and you have to speak them out 00:02:17.219 --> 00:02:21.099 loud. That step is actually crucial for the audiovisual 00:02:21.099 --> 00:02:24.319 think. Because it teaches the AI how your specific 00:02:24.319 --> 00:02:27.240 mouth moves. Right. It matches your physical 00:02:27.240 --> 00:02:29.400 mouth movement directly to your unique voice. 00:02:29.939 --> 00:02:32.840 Then you move your head slowly in a circle. And 00:02:32.840 --> 00:02:35.120 you also move it side to side. Yeah, but you 00:02:35.120 --> 00:02:37.879 must keep your shoulders completely still during 00:02:37.879 --> 00:02:40.740 this process. It feels like setting up FaceEyed, 00:02:40.759 --> 00:02:44.280 but for a Hollywood stand -in. Beat. Environmental 00:02:44.280 --> 00:02:46.659 lighting is also incredibly crucial here. Oh, 00:02:46.659 --> 00:02:49.039 for sure. You need to sit facing a natural window. 00:02:49.659 --> 00:02:52.560 Or you must use a strong, direct -room light. 00:02:52.740 --> 00:02:54.580 Because you can never have light shining from 00:02:54.580 --> 00:02:56.680 behind you. Right. It creates dark, muddy shadows 00:02:56.680 --> 00:02:59.360 across your features. The AI completely fails 00:02:59.360 --> 00:03:02.099 to read your face clearly. Dark edges confuse 00:03:02.099 --> 00:03:04.360 the system's depth perception, right? Yeah. It 00:03:04.360 --> 00:03:06.319 ruins the borders around your hair and clothes. 00:03:06.500 --> 00:03:09.129 That causes weird visual glitches later. This 00:03:09.129 --> 00:03:12.090 system is also incredibly strict about your expression 00:03:12.090 --> 00:03:14.909 during the scan. You must maintain a strictly 00:03:14.909 --> 00:03:18.409 neutral expression. A small, subtle smile is 00:03:18.409 --> 00:03:21.210 perfectly fine. But why is a completely neutral 00:03:21.210 --> 00:03:24.969 expression so strongly required here? Well, exaggerated 00:03:24.969 --> 00:03:27.650 expressions bake permanent distortion into the 00:03:27.650 --> 00:03:30.810 baseline model. Oh, really? Yeah. If you scan 00:03:30.810 --> 00:03:33.969 your face with a huge, goofy grin, the AI assumes 00:03:33.969 --> 00:03:36.310 that is your standard resting face. So it just 00:03:36.310 --> 00:03:38.789 constantly looks like that. Exactly. Every single 00:03:38.789 --> 00:03:42.030 video you generate will feature that exact stretched 00:03:42.030 --> 00:03:45.009 expression. Got it. Big smiles confuse the AI 00:03:45.009 --> 00:03:47.770 and warp your future video outputs. You really 00:03:47.770 --> 00:03:50.009 want to give this system a clean, relaxed canvas. 00:03:50.629 --> 00:03:52.969 From that baseline, it can artificially generate 00:03:52.969 --> 00:03:55.629 sadness, anger, or joy much more accurately. 00:03:55.770 --> 00:03:58.250 OK. Now that our digital face is completely calibrated. 00:03:58.539 --> 00:04:00.879 we move to the fun part. How do we actually get 00:04:00.879 --> 00:04:02.840 on screen without burning through our wallet? 00:04:03.020 --> 00:04:06.340 That is a very real, practical concern for creators. 00:04:07.020 --> 00:04:09.460 Video generation takes massive amounts of serious 00:04:09.460 --> 00:04:11.759 computing power. And processing video commands 00:04:11.759 --> 00:04:14.620 directly on the main Gemini interface is expensive. 00:04:14.939 --> 00:04:17.160 Very expensive. It drains your account credits 00:04:17.160 --> 00:04:19.600 incredibly rapidly. You will run out of rendering 00:04:19.600 --> 00:04:21.680 attempts before you even finish your first scene. 00:04:21.959 --> 00:04:24.839 Yeah, to avoid that trap, you should use Google 00:04:24.839 --> 00:04:28.240 Labs. Specifically, you want to use their Flow 00:04:28.240 --> 00:04:31.360 workspace. Flow is an experimental testing environment, 00:04:31.480 --> 00:04:33.980 right? Built specifically for pro users, yeah. 00:04:34.600 --> 00:04:37.500 It lets you test custom prompts freely without 00:04:37.500 --> 00:04:39.759 those strict credit limits. What about the digital 00:04:39.759 --> 00:04:42.399 avatar you just spent time making? It syncs instantly 00:04:42.399 --> 00:04:45.079 between both workspaces under one single login. 00:04:45.319 --> 00:04:48.199 Oh, nice! It creates a seamless transition for 00:04:48.199 --> 00:04:50.860 the user. You build the avatar once, and it lives 00:04:50.860 --> 00:04:53.930 across the entire ecosystem. Once we are inside 00:04:53.930 --> 00:04:56.129 the flow workspace, we need to write commands. 00:04:56.810 --> 00:04:59.670 For total beginners, diving right into complex 00:04:59.670 --> 00:05:02.589 prompting is overwhelming. It really is. So you 00:05:02.589 --> 00:05:04.230 should probably start with the ready templates. 00:05:04.649 --> 00:05:06.670 Templates are a fantastic starting point for 00:05:06.670 --> 00:05:08.889 learning the visual language. Yeah, you can select 00:05:08.889 --> 00:05:11.730 pre -built visual styles like pastel film or 00:05:11.730 --> 00:05:14.709 Japanese anime. Or even something heavily stylized 00:05:14.709 --> 00:05:17.439 like comic book. The system automatically handles 00:05:17.439 --> 00:05:19.680 all the complex lighting and color math behind 00:05:19.680 --> 00:05:22.740 the scenes. You do not need to think about complicated 00:05:22.740 --> 00:05:25.879 camera terminology yet. But eventually you will 00:05:25.879 --> 00:05:28.139 definitely want more creative control. You will 00:05:28.139 --> 00:05:30.740 want to write your own custom prompts from scratch. 00:05:31.220 --> 00:05:33.300 Definitely. And writing custom prompts requires 00:05:33.300 --> 00:05:36.800 a very specific five -part structure. You need 00:05:36.800 --> 00:05:41.220 a subject, an action, a location, specific lighting, 00:05:41.540 --> 00:05:43.939 and a camera angle. Right. If we connect this 00:05:43.939 --> 00:05:46.579 to the bigger picture, it is like stacking Lego 00:05:46.579 --> 00:05:49.360 blocks of data. You build the scene piece by 00:05:49.360 --> 00:05:52.319 piece. Yeah, ensuring the AI understands every 00:05:52.319 --> 00:05:54.740 single layer. Let us look at a great science 00:05:54.740 --> 00:05:57.459 fiction example. Imagine asking the system for 00:05:57.459 --> 00:06:00.519 a 10 second video. The subject is a lone astronaut. 00:06:00.740 --> 00:06:02.699 The action is standing still and looking up. 00:06:02.980 --> 00:06:05.699 The location is the dusty red surface of Mars 00:06:05.699 --> 00:06:08.740 facing a massive glowing crystal monument. The 00:06:08.740 --> 00:06:11.620 lighting block would be neon blue. ambient lighting. 00:06:12.019 --> 00:06:14.740 And the final camera angle is a cinematic medium 00:06:14.740 --> 00:06:18.779 shot. It is incredibly specific. Beat. You know, 00:06:18.800 --> 00:06:20.879 I still wrestle with prompt drift myself. It 00:06:20.879 --> 00:06:22.779 happens to everyone at first. Just to define 00:06:22.779 --> 00:06:26.600 that prompt drift, when AI outputs slowly change 00:06:26.600 --> 00:06:29.019 away from your idea. Exactly. But let me push 00:06:29.019 --> 00:06:31.519 back on this strict five part structure for a 00:06:31.519 --> 00:06:34.819 second. If we give the AI slightly less information, 00:06:35.199 --> 00:06:37.920 might it result in more creative surprises? Not 00:06:37.920 --> 00:06:41.459 at all. Vague prompts simply force the AI engine 00:06:41.459 --> 00:06:45.019 to guess wildly. Oh, really? Yeah. When it guesses, 00:06:45.579 --> 00:06:48.220 it pulls from completely unrelated visual training 00:06:48.220 --> 00:06:51.339 data that almost always results in messy, highly 00:06:51.339 --> 00:06:54.379 inconsistent visuals. So fewer details don't 00:06:54.379 --> 00:06:56.879 equal more creativity, they just equal more rendering 00:06:56.879 --> 00:07:00.000 mistakes. Precisely. It makes the system incredibly 00:07:00.000 --> 00:07:03.199 noisy with conflicting visual information. A 00:07:03.199 --> 00:07:05.519 strict structure acts like guardrails for the 00:07:05.519 --> 00:07:07.720 rendering engine. OK, once you finally generate 00:07:07.720 --> 00:07:10.220 that very first 10 -second clip, you realize 00:07:10.220 --> 00:07:12.660 something immediately. It is rarely completely 00:07:12.660 --> 00:07:14.779 perfect on the first try. And this is exactly 00:07:14.779 --> 00:07:16.920 where Omni completely separates itself from older 00:07:16.920 --> 00:07:19.240 tools. Right. With older generation tools, you 00:07:19.240 --> 00:07:20.939 just had to start completely over. You crossed 00:07:20.939 --> 00:07:22.779 your fingers, changed a word, and tried again. 00:07:22.939 --> 00:07:25.199 Yeah, it was super frustrating. But here, you 00:07:25.199 --> 00:07:27.600 literally converse with the editor. Omni allows 00:07:27.600 --> 00:07:30.300 step -by -step scene editing directly inside 00:07:30.300 --> 00:07:33.660 the chat interface. It feels exactly like working 00:07:33.660 --> 00:07:36.639 alongside a human video editor. It really does. 00:07:37.480 --> 00:07:40.000 Let us look at the first major editing feature 00:07:40.000 --> 00:07:43.120 available. You can continue scenes directly from 00:07:43.120 --> 00:07:45.959 the last exact frame. This fundamentally fixes 00:07:45.959 --> 00:07:48.920 the old problem of AI videos being painfully 00:07:48.920 --> 00:07:52.560 short. Yes. The AI catches the absolute last 00:07:52.560 --> 00:07:56.029 frame of your 10 -second clip. It uses that exact 00:07:56.029 --> 00:07:58.689 frozen frame as the starting point for the next 00:07:58.689 --> 00:08:01.129 clip. So your character can realistically stand 00:08:01.129 --> 00:08:03.730 up from a chair. Then in the next prompt, they 00:08:03.730 --> 00:08:06.389 can walk slowly to the door. And the story flows 00:08:06.389 --> 00:08:08.930 smoothly forward. The complex room lighting stays 00:08:08.930 --> 00:08:11.970 identical across both clips too. The second feature 00:08:11.970 --> 00:08:15.329 is editing one very specific detail inside the 00:08:15.329 --> 00:08:18.189 frame. This represents a massive technical achievement 00:08:18.189 --> 00:08:20.389 in spatial computing. You can tell the chat editor 00:08:20.389 --> 00:08:22.389 to change the color of your shirt to light blue. 00:08:22.550 --> 00:08:24.129 And the engine leaves everything else in the 00:08:24.129 --> 00:08:26.550 scene perfectly intact. The rainy weather, your 00:08:26.550 --> 00:08:28.709 facial expression, and the background do not 00:08:28.709 --> 00:08:31.370 change at all. It is amazing. The third feature 00:08:31.370 --> 00:08:33.870 focuses heavily on generating silent cinematic 00:08:33.870 --> 00:08:36.529 clips. Because we do not always need people talking 00:08:36.529 --> 00:08:39.230 directly to the camera in a video. Right. Sometimes 00:08:39.230 --> 00:08:42.299 quiet slow -mo scenes carry much more emotional 00:08:42.299 --> 00:08:45.279 weight. for the viewer. You can focus on small 00:08:45.279 --> 00:08:48.659 highly detailed environmental elements. Imagine 00:08:48.659 --> 00:08:51.080 generating a close -up of heavy raindrops falling 00:08:51.080 --> 00:08:53.940 on a car window at night. With blurry colorful 00:08:53.940 --> 00:08:57.120 city neon lights glowing in the deep background. 00:08:57.320 --> 00:09:00.159 Focusing on small tight details creates highly 00:09:00.159 --> 00:09:02.879 stable incredibly artistic footage. It really 00:09:02.879 --> 00:09:05.580 does. But I have a practical question about this 00:09:05.580 --> 00:09:09.039 step -by -step chat editing process. Let us say 00:09:09.039 --> 00:09:11.519 I want a blue shirt and I also want rainy weather. 00:09:11.840 --> 00:09:15.149 Does asking for multiple major changes at once 00:09:15.149 --> 00:09:18.669 save me processing time? No, actually. That is 00:09:18.669 --> 00:09:21.029 a very common mistake new users make. Oh, really? 00:09:21.129 --> 00:09:23.490 Yeah. Combining multiple structural edits at 00:09:23.490 --> 00:09:26.990 the exact same time deeply confuses the AI. Yeah. 00:09:27.149 --> 00:09:28.950 It usually ruins the parts of the clip that were 00:09:28.950 --> 00:09:30.990 already beautiful. Makes sense. Changing too 00:09:30.990 --> 00:09:33.090 much at once breaks the system's focus. Right, 00:09:33.210 --> 00:09:35.169 and you end up wasting your valuable rendering 00:09:35.169 --> 00:09:37.490 credits. You should only change one specific 00:09:37.490 --> 00:09:40.490 detail per chat turn to keep the AI stable. Okay, 00:09:40.769 --> 00:09:42.870 we're going to take a brief pause for a message 00:09:42.870 --> 00:09:45.889 from our sponsors. Mid -roll sponsor, break placeholder. 00:09:46.129 --> 00:09:49.500 And we are back. We have spent time talking about 00:09:49.500 --> 00:09:51.919 building totally synthetic worlds so far. Yeah, 00:09:51.980 --> 00:09:55.440 we have. But the real magic happens when we inject 00:09:55.440 --> 00:09:58.600 actual reality into the Omni -Engine. This is 00:09:58.600 --> 00:10:00.820 where the tool becomes incredibly practical for 00:10:00.820 --> 00:10:03.559 everyday smartphone users. We can actually edit 00:10:03.720 --> 00:10:06.759 real physical videos from our own camera rolls. 00:10:06.940 --> 00:10:08.820 You can just use your phone to record a normal 00:10:08.820 --> 00:10:12.220 real video outside. You might simply film a quiet 00:10:12.220 --> 00:10:15.220 empty city street, or you might record a snowy 00:10:15.220 --> 00:10:17.879 mountain out the window of a moving car. Then 00:10:17.879 --> 00:10:20.379 you upload that real video clip directly into 00:10:20.379 --> 00:10:23.320 the chat interface, and you ask the AI to add 00:10:23.320 --> 00:10:25.519 entirely magical elements to it. Like you could 00:10:25.519 --> 00:10:28.100 ask it to add a massive flowing waterfall pouring 00:10:28.100 --> 00:10:30.740 down that real mountain. Exactly. The system 00:10:30.740 --> 00:10:33.379 automatically analyzes exactly how your physical 00:10:33.379 --> 00:10:36.000 camera was moving. It perfectly tracks the original 00:10:36.000 --> 00:10:38.919 camera movement. The virtual, computer -generated 00:10:38.919 --> 00:10:41.519 water blends naturally with the real pine trees 00:10:41.519 --> 00:10:44.360 beat. You can also use the system to create highly 00:10:44.360 --> 00:10:47.299 emotional photo montages. You just upload your 00:10:47.299 --> 00:10:50.200 regular, everyday photos directly to the Uploads 00:10:50.200 --> 00:10:52.720 menu. You should definitely avoid using any photos 00:10:52.720 --> 00:10:54.519 of famous people, though. Yeah, the Flow Platform 00:10:54.519 --> 00:10:57.179 actively blocks those requests due to strict 00:10:57.179 --> 00:10:59.840 safety guardrails. But with your own family photos, 00:10:59.899 --> 00:11:03.659 you just ask the system for a story. The AI intelligently 00:11:03.659 --> 00:11:06.559 adds very smooth transition effects between the 00:11:06.559 --> 00:11:09.559 images. It adds gentle cinematic camera zooms 00:11:09.559 --> 00:11:12.980 to create a lively moving memory video. It is 00:11:12.980 --> 00:11:14.980 super cool. You can even add stunning effects 00:11:14.980 --> 00:11:17.700 to a single static Pirkert photo. You just upload 00:11:17.700 --> 00:11:20.159 a standard close -up face photo into the chat. 00:11:20.320 --> 00:11:22.679 The system mathematically calculates the space 00:11:22.679 --> 00:11:26.320 layers of the hair and eyes. Space layers? Invisible 00:11:26.320 --> 00:11:29.860 3D slices the AI builds from a flat photo? Right. 00:11:30.090 --> 00:11:32.649 It uses these invisible layers to add natural 00:11:32.649 --> 00:11:35.450 micro -movements to the still image. It creates 00:11:35.450 --> 00:11:37.990 the hyper -realistic effect of wind blowing gently 00:11:37.990 --> 00:11:40.450 through the subject's hair. Or it makes the character's 00:11:40.450 --> 00:11:42.370 eyes blink naturally looking around the room. 00:11:42.690 --> 00:11:44.990 And it does all this without artificially altering 00:11:44.990 --> 00:11:49.169 the original face details. Whoa. Beat. Imagine 00:11:49.169 --> 00:11:52.320 scaling to a billion queries. The processing 00:11:52.320 --> 00:11:54.879 power required to do that instantly across the 00:11:54.879 --> 00:11:58.100 globe is staggering. It is exactly like having 00:11:58.100 --> 00:12:00.960 an entire Hollywood VFX department sitting inside 00:12:00.960 --> 00:12:03.580 a text box. It really is. But how exactly does 00:12:03.580 --> 00:12:06.740 it calculate something complex like wind in a 00:12:06.740 --> 00:12:09.700 flat 2D photo? It calculates those spatial layers 00:12:09.700 --> 00:12:12.000 between the individual strands of hair and the 00:12:12.000 --> 00:12:14.360 background. So it actually maps out the depth. 00:12:14.460 --> 00:12:17.600 Yeah. This allows the AI engine to simulate realistic 00:12:17.600 --> 00:12:21.679 3D depth and physical movement. Ah, it cuts the 00:12:21.679 --> 00:12:24.100 photo into 3D layers to create natural movement. 00:12:24.259 --> 00:12:25.980 Yes, exactly. It separates the human subject 00:12:25.980 --> 00:12:27.820 from the background mathematically, allowing 00:12:27.820 --> 00:12:30.440 independent motion. To truly master this entire 00:12:30.440 --> 00:12:32.320 system, we need to understand the engine under 00:12:32.320 --> 00:12:34.700 the hood. Right. We need to optimize our daily 00:12:34.700 --> 00:12:37.679 workflow to perfectly match its technical capabilities. 00:12:38.000 --> 00:12:40.139 The underlying technology driving this incredible 00:12:40.139 --> 00:12:42.700 speed is called the flash model. The flash model, 00:12:43.159 --> 00:12:45.879 a lightweight AI designed for extremely fast 00:12:45.879 --> 00:12:47.990 video processing. Thanks to this streamlined 00:12:47.990 --> 00:12:50.610 model, a complex 10 -second scene generates in 00:12:50.610 --> 00:12:53.490 under two minutes. That rapid turnaround speed 00:12:53.490 --> 00:12:56.190 completely changes how digital creators actually 00:12:56.190 --> 00:12:59.149 work. In the past, you painfully waited 30 minutes 00:12:59.149 --> 00:13:01.889 just for a short, blurry clip? Yeah, it was agonizing. 00:13:02.250 --> 00:13:04.570 It is helpful to compare Gemini Omni against 00:13:04.570 --> 00:13:06.990 the other major AI video tools out there. For 00:13:06.990 --> 00:13:10.330 instance, we can look at Sora from OpenAI. Sora 00:13:10.330 --> 00:13:14.019 is highly realistic. but it is known to be very 00:13:14.019 --> 00:13:17.080 slow. It is also quite hard to prompt effectively 00:13:17.080 --> 00:13:20.120 for specific repeatable camera angles. You can 00:13:20.120 --> 00:13:22.419 also compare Omni directly to a platform like 00:13:22.419 --> 00:13:24.759 Runway. Runway is definitely faster, but the 00:13:24.759 --> 00:13:27.259 outputs often look much more like computer graphics. 00:13:27.460 --> 00:13:30.940 Gemini Omni seems to hit a very specific, highly 00:13:30.940 --> 00:13:33.639 productive sweet spot. It offers rapid speed, 00:13:34.039 --> 00:13:36.419 chat -based ease of use, and stunning cinematic 00:13:36.419 --> 00:13:38.559 quality. You just talk to it normally, refining 00:13:38.559 --> 00:13:40.899 the shot over a coffee. To guarantee that cinematic 00:13:40.899 --> 00:13:43.539 quality, you must use very specific professional 00:13:43.539 --> 00:13:45.679 vocabulary. Yeah, you must intentionally use 00:13:45.679 --> 00:13:48.500 specific camera lens and lighting terms in your 00:13:48.500 --> 00:13:52.059 prompt. For example, typing 35mm anamorphic lens 00:13:52.059 --> 00:13:55.080 forces a wide, highly cinematic focal length. 00:13:55.279 --> 00:13:58.120 It perfectly simulates the optical characteristics 00:13:58.120 --> 00:14:01.220 of a special movie camera lens. Another incredibly 00:14:01.220 --> 00:14:04.519 great phrase to use is shallow depth of field. 00:14:04.700 --> 00:14:07.379 Shallow depth of field. Oh! Blurring the background 00:14:07.379 --> 00:14:10.019 so your main subject stands out. It mathematically 00:14:10.019 --> 00:14:13.039 separates the extremely sharp subject from the 00:14:13.039 --> 00:14:15.059 blurry background behind them. You should also 00:14:15.059 --> 00:14:18.549 actively avoid using vague. poetic adjectives 00:14:18.549 --> 00:14:21.850 like surreal. Because they just confuse the rendering 00:14:21.850 --> 00:14:25.429 process and create messy visual noise. Beat. 00:14:26.190 --> 00:14:28.970 There is one final critical insight regarding 00:14:28.970 --> 00:14:31.789 your daily workflow. You must treat the 10 second 00:14:31.789 --> 00:14:34.289 output limit as a structural requirement. Do 00:14:34.289 --> 00:14:36.789 not view it as a software flaw or a frustrating 00:14:36.789 --> 00:14:39.129 limitation. Right. But honestly, isn't a strict 00:14:39.129 --> 00:14:41.330 10 second limit incredibly frustrating if you 00:14:41.330 --> 00:14:44.230 want to make a short film? Not necessarily. Actively 00:14:44.230 --> 00:14:46.610 scripting your story around 10 -second cuts creates 00:14:46.610 --> 00:14:49.470 much faster, more attractive pacing. Oh, I see. 00:14:49.509 --> 00:14:51.970 It also strongly prevents the severe rendering 00:14:51.970 --> 00:14:54.570 errors that frequently happen in long, continuous 00:14:54.570 --> 00:14:57.370 AI generations. Right. Forced brevity actually 00:14:57.370 --> 00:15:01.029 leads to a much better, tighter story. It fundamentally 00:15:01.029 --> 00:15:03.549 forces you to think exactly like a traditional 00:15:03.549 --> 00:15:06.470 film editor. You shoot small, perfect scenes, 00:15:06.889 --> 00:15:09.250 and you stitch them all together later. So what 00:15:09.250 --> 00:15:11.759 does this all mean? This brings us to the big 00:15:11.759 --> 00:15:15.080 idea of this entire deep dive. Gemini Omni is 00:15:15.080 --> 00:15:18.039 an incredibly powerful multi -input system. It 00:15:18.039 --> 00:15:21.000 fundamentally democratizes high -end cinematic 00:15:21.000 --> 00:15:23.340 creation for absolutely everyone with this heart 00:15:23.340 --> 00:15:26.480 phone. It completely removes those massive complex 00:15:26.480 --> 00:15:29.179 technical barriers that used to stop creative 00:15:29.179 --> 00:15:32.039 people. By using very simple chat interactions, 00:15:32.419 --> 00:15:34.779 literally anyone can create breathtaking art. 00:15:35.039 --> 00:15:37.559 You simply follow a strict five -part prompt 00:15:37.559 --> 00:15:39.799 structure to guide the engine. You test your 00:15:39.799 --> 00:15:42.419 wild ideas safely inside the free flow workspace. 00:15:42.919 --> 00:15:45.279 And you must remember to edit only one specific 00:15:45.279 --> 00:15:48.120 detail at a time. Anyone can now direct a high 00:15:48.120 --> 00:15:50.820 quality, deeply emotional video in mere minutes. 00:15:51.320 --> 00:15:53.299 We highly encourage you to open the app on your 00:15:53.299 --> 00:15:55.419 phone today. Yeah, just give it a try. Set up 00:15:55.419 --> 00:15:57.879 your digital face carefully while facing a bright 00:15:57.879 --> 00:16:00.620 natural window. Try generating your very first 00:16:00.620 --> 00:16:05.200 10 second cinematic clip. We started today by 00:16:05.200 --> 00:16:07.360 talking about how easily we can now generate 00:16:07.360 --> 00:16:10.220 these cinematic scenes. We are just typing basic 00:16:10.220 --> 00:16:13.620 words into a small chat box. If we can now upload 00:16:13.620 --> 00:16:16.820 a simple, fading childhood photo and instantly 00:16:16.820 --> 00:16:20.120 generate a hyper -realistic moving memory, with 00:16:20.120 --> 00:16:22.259 physical wind blowing in our hair and a warm 00:16:22.259 --> 00:16:24.860 smile forming on our face, what happens to the 00:16:24.860 --> 00:16:27.779 truth of our own past? When our camera rolls 00:16:27.779 --> 00:16:30.580 become digital canvases, how will we remember 00:16:30.580 --> 00:16:32.379 what actually happened versus what we simply 00:16:32.379 --> 00:16:34.860 prompted into existence? Something to think about. 00:16:35.299 --> 00:16:35.940 Until next time.