WEBVTT 00:00:00.000 --> 00:00:02.940 You know, the idea of making professional cinematic 00:00:02.940 --> 00:00:05.160 videos usually means, well, cameras, lights, 00:00:05.280 --> 00:00:08.179 maybe a studio. You definitely have to be OK 00:00:08.179 --> 00:00:11.580 with being on screen. But what if you could just 00:00:11.580 --> 00:00:14.460 skip all that? We're talking about generating 00:00:14.460 --> 00:00:16.920 these really realistic personalized videos of 00:00:16.920 --> 00:00:19.679 yourself, like a talking avatar, without ever 00:00:19.679 --> 00:00:22.359 actually recording yourself. And this level of 00:00:22.359 --> 00:00:24.440 creation, it isn't some far off future thing. 00:00:24.620 --> 00:00:27.539 It's accessible, like. right now and mostly for 00:00:27.539 --> 00:00:29.699 free to start. Oh, it's absolutely true. And 00:00:29.699 --> 00:00:31.940 today we're really going to dive deep into the 00:00:31.940 --> 00:00:34.340 specific workflow from the source material we 00:00:34.340 --> 00:00:37.399 looked at. It uses this surprisingly powerful 00:00:37.399 --> 00:00:40.460 three tool AI stack. So our mission today is 00:00:40.460 --> 00:00:42.920 basically to unpack that whole five phase process. 00:00:42.939 --> 00:00:45.200 Yeah. And really focus on the practical stuff 00:00:45.200 --> 00:00:47.479 you need to master. We got to go beyond just 00:00:47.479 --> 00:00:50.039 naming the software. We're looking for that secret 00:00:50.039 --> 00:00:52.000 sauce for getting consistent prompts, the actual 00:00:52.000 --> 00:00:54.320 logistics of creating that avatar and how you 00:00:54.320 --> 00:00:56.679 genuinely direct. the final video output. We're 00:00:56.679 --> 00:00:58.500 aiming to give you a serious shortcut to making 00:00:58.500 --> 00:01:00.700 this kind of high quality content, and efficiently 00:01:00.700 --> 00:01:03.140 too. OK, sounds good. Let's unpack this setup 00:01:03.140 --> 00:01:05.219 then. What are the tools that make this whole 00:01:05.219 --> 00:01:07.719 thing possible? All right, so the foundation 00:01:07.719 --> 00:01:10.659 here is these three specific tools. When you 00:01:10.659 --> 00:01:12.180 chain them together, they seem to work really 00:01:12.180 --> 00:01:15.560 well. First up, Chat GPT. And the source material 00:01:15.560 --> 00:01:18.870 is pretty clear. you don't actually need the 00:01:18.870 --> 00:01:21.890 paid plus plan, the $20 a month one, just to 00:01:21.890 --> 00:01:23.709 get started. Yeah, that's super important for 00:01:23.709 --> 00:01:25.730 accessibility, right? The free version works 00:01:25.730 --> 00:01:27.569 fine, mainly because we're going to be using 00:01:27.569 --> 00:01:32.010 these specialized custom GPTs. Think of them 00:01:32.010 --> 00:01:34.430 like little helpers trained for really specific 00:01:34.430 --> 00:01:37.689 jobs, like writing these cinematic image prompts. 00:01:37.730 --> 00:01:39.769 You really only need that plus subscription later 00:01:39.769 --> 00:01:42.090 if you find you want faster speeds, or maybe 00:01:42.090 --> 00:01:44.730 access to the absolute newest models, but not 00:01:44.730 --> 00:01:47.379 essential at the start. OK. Tool number two is 00:01:47.379 --> 00:01:49.959 Nano Banana. That's for the image creation part. 00:01:50.099 --> 00:01:52.219 Yep. That's kind of the engine room. It's got 00:01:52.219 --> 00:01:54.239 a great free plan. You can make almost unlimited 00:01:54.239 --> 00:01:56.640 pictures, which is amazing. And its main strength 00:01:56.640 --> 00:01:59.060 is how it automatically handles that really tricky 00:01:59.060 --> 00:02:01.719 technical bit, keeping the face consistent across 00:02:01.719 --> 00:02:04.260 images. Yeah. Facial matching. Right. That consistency 00:02:04.260 --> 00:02:06.650 seems crucial. And then tool three. Then you 00:02:06.650 --> 00:02:09.770 move up to the big gun, Google VEO3. That's for 00:02:09.770 --> 00:02:11.969 the actual video generation. And what's really 00:02:11.969 --> 00:02:15.050 interesting here is the kind of financial opportunity 00:02:15.050 --> 00:02:17.930 built into how things are set up right now. There's 00:02:17.930 --> 00:02:20.810 a one -month free trial, which is great for testing. 00:02:21.270 --> 00:02:24.509 But get this, if you happen to be a student or 00:02:24.509 --> 00:02:27.610 have a .edu email address for any reason, you 00:02:27.610 --> 00:02:31.129 can currently get this massive 18 -month free 00:02:31.129 --> 00:02:34.819 deal. I mean, that's a total game changer for 00:02:34.819 --> 00:02:37.259 long term experimentation, right? That deal alone 00:02:37.259 --> 00:02:39.219 makes this project seem really worthwhile exploring. 00:02:39.639 --> 00:02:41.879 But is juggling three different tools really 00:02:41.879 --> 00:02:44.020 worth the hassle compared to maybe finding it 00:02:44.020 --> 00:02:46.400 all in one generator? Yeah, I think so because 00:02:46.400 --> 00:02:48.979 what you gain is control. And that control starts 00:02:48.979 --> 00:02:52.879 with a really key setup tip for VEO3. You absolutely 00:02:52.879 --> 00:02:55.020 must use the flow interface. If you just try 00:02:55.020 --> 00:02:57.400 and do it in the normal Gemini chat window, you 00:02:57.400 --> 00:02:59.840 lose all that director level control over video 00:02:59.840 --> 00:03:02.740 shape, quality settings. output formats, stuff 00:03:02.740 --> 00:03:04.860 you really need for professional results. Okay, 00:03:04.960 --> 00:03:07.139 so using Flow is like stepping out of the basic 00:03:07.139 --> 00:03:10.060 chat and into the production suite, basically. 00:03:10.300 --> 00:03:12.120 Exactly. And if we're talking about making the 00:03:12.120 --> 00:03:14.340 most of that free time, especially if you're 00:03:14.340 --> 00:03:16.919 on a trial, managing those credits has got to 00:03:16.919 --> 00:03:19.379 be super important. Oh, absolutely. Don't just 00:03:19.379 --> 00:03:21.699 burn through those precious free credits. A huge 00:03:21.699 --> 00:03:26.240 tip is organization and Testing smart. Always 00:03:26.240 --> 00:03:29.199 start by asking the E03 for just one video output 00:03:29.199 --> 00:03:31.400 first. Check the result, see if it's going the 00:03:31.400 --> 00:03:33.860 right direction, and then iterate. Don't ask 00:03:33.860 --> 00:03:36.800 for four variations right off the bat. And iterate 00:03:36.800 --> 00:03:39.539 using the fast mode first. Right. That saves 00:03:39.539 --> 00:03:42.330 a lot of resources. Totally. Quality mode looks 00:03:42.330 --> 00:03:44.650 amazing, but it chews through four times the 00:03:44.650 --> 00:03:47.469 credits. Use fast mode. It's about five times 00:03:47.469 --> 00:03:49.509 cheaper. It gives you a test video in under a 00:03:49.509 --> 00:03:52.050 minute, usually. Use that to quickly check your 00:03:52.050 --> 00:03:54.849 ideas, your prompts. Only switch over to quality 00:03:54.849 --> 00:03:56.669 mode when you're ready for the final polished 00:03:56.669 --> 00:03:59.330 output. So what's the biggest efficiency gain 00:03:59.330 --> 00:04:02.129 there, really, from testing in fast first? It 00:04:02.129 --> 00:04:04.710 just saves credits, lets you test way more ideas 00:04:04.710 --> 00:04:07.629 before the final render, maximize that experimentation 00:04:07.629 --> 00:04:09.629 phase. Right, maximum testing within the budget. 00:04:09.979 --> 00:04:12.099 Okay, let's talk about those prompts then because 00:04:12.099 --> 00:04:14.539 they seem like the real core of getting the visuals 00:04:14.539 --> 00:04:17.740 right Yeah, this brings us to what really separates, 00:04:17.740 --> 00:04:20.319 you know amateur results from professional looking 00:04:20.319 --> 00:04:23.060 AI video It's the quality of the prompt. We're 00:04:23.060 --> 00:04:25.980 not just having a basic chat with chat GPT here 00:04:25.980 --> 00:04:29.600 We're using a specialized custom GPT one that's 00:04:29.600 --> 00:04:32.300 designed specifically for image creation prompts 00:04:32.620 --> 00:04:35.079 You can usually find these in the GPT store. 00:04:35.180 --> 00:04:37.639 Just search for something like nano banana prompt 00:04:37.639 --> 00:04:42.060 or cinematic image prompt. So these custom GPTs, 00:04:42.300 --> 00:04:45.310 they've... kind of absorbed the lessons from 00:04:45.310 --> 00:04:48.069 thousands of successful visual requests. They 00:04:48.069 --> 00:04:49.930 act almost like a digital storyboard artist for 00:04:49.930 --> 00:04:52.629 you. Precisely. They're trained on tons of successful 00:04:52.629 --> 00:04:54.850 examples, so they just get cinematic description. 00:04:55.250 --> 00:04:57.170 They understand how to plan a picture that's 00:04:57.170 --> 00:04:58.850 going to look good once you add motion later. 00:04:59.230 --> 00:05:01.089 You've got to go way beyond simple stuff like, 00:05:01.089 --> 00:05:03.329 you know, person in a forest. You need to demand 00:05:03.329 --> 00:05:06.790 story, detail, mood. OK, so give us an example 00:05:06.790 --> 00:05:09.189 then. How should a listener frame that stronger, 00:05:09.209 --> 00:05:11.639 more cinematic request? Well, instead of just 00:05:11.639 --> 00:05:14.459 the basic description, you'd prompt the custom 00:05:14.459 --> 00:05:16.959 GPT with the scene's intent. Something like, 00:05:17.620 --> 00:05:20.480 create a cinematic style prompt for a young financial 00:05:20.480 --> 00:05:22.720 expert. She's presenting an idea in a modern 00:05:22.720 --> 00:05:26.139 office setting. The lighting needs to feel professional, 00:05:26.360 --> 00:05:29.019 making her look trustworthy. And then the GPT 00:05:29.019 --> 00:05:31.560 will spit back a whole picture plan. It'll include 00:05:31.560 --> 00:05:34.800 specific late details, maybe soft light flooding 00:05:34.800 --> 00:05:38.790 in from a large window. Camera angle ideas, like 00:05:38.790 --> 00:05:41.290 eye level shot, medium close -up, even color 00:05:41.290 --> 00:05:44.009 notes, like cool blue and gray tones dominate 00:05:44.009 --> 00:05:47.110 the palette. Wow, okay. That level of specificity, 00:05:47.129 --> 00:05:49.149 you can see how that would drastically improve 00:05:49.149 --> 00:05:51.589 the visual quality, the fidelity, and just the 00:05:51.589 --> 00:05:53.370 overall mood of the scene. Yeah. But when you 00:05:53.370 --> 00:05:55.709 use one of these super descriptive prompts, isn't 00:05:55.709 --> 00:05:58.589 there a risk that the image generator, Nano Banana 00:05:58.589 --> 00:06:00.829 in this case, just gets... well, too creative, 00:06:01.329 --> 00:06:02.949 and kind of ignores the facial reference photo 00:06:02.949 --> 00:06:05.290 you give it later. Mm -hmm. Ah, yeah. That's 00:06:05.290 --> 00:06:07.709 the constant battle with AI art, isn't it? That 00:06:07.709 --> 00:06:10.209 prompt drift. You have to keep refining. It's 00:06:10.209 --> 00:06:12.889 never quite perfect first time. Which is why 00:06:12.889 --> 00:06:15.350 I also really love this powerful trick for iteration. 00:06:16.029 --> 00:06:18.569 Once you get a prompt that works reasonably well... 00:06:18.600 --> 00:06:21.860 ask that same custom GPT to generate, say, five 00:06:21.860 --> 00:06:23.819 different versions of that prompt. Just ask it 00:06:23.819 --> 00:06:25.959 to change only one thing each time, maybe the 00:06:25.959 --> 00:06:28.439 location or the time of day or the clothing she's 00:06:28.439 --> 00:06:30.860 wearing. Doing that lets you rapidly build up 00:06:30.860 --> 00:06:33.399 a library of related, effective prompts. Saves 00:06:33.399 --> 00:06:36.720 hours of manual tweaking later. You know, honestly, 00:06:36.839 --> 00:06:39.019 I still wrestle with prompt drift myself sometimes. 00:06:39.160 --> 00:06:42.579 It's tricky. So using the GPT to help debug image 00:06:42.579 --> 00:06:45.339 errors. Yeah, that's crucial, even for me. That's 00:06:45.339 --> 00:06:47.639 actually helpful to hear that even experts hit 00:06:47.639 --> 00:06:50.139 that wall sometimes. So, okay, let's say Nano 00:06:50.139 --> 00:06:52.579 Banana keeps making weird visual mistakes like 00:06:52.579 --> 00:06:54.639 the eyes look strange consistently or there's 00:06:54.639 --> 00:06:56.660 some repeating pattern in the background. How 00:06:56.660 --> 00:06:58.959 exactly does the AI help you debug that? Well, 00:06:58.959 --> 00:07:01.600 you basically describe that specific visual mistake 00:07:01.600 --> 00:07:04.199 back to the custom GPT. You tell it, hey, the 00:07:04.199 --> 00:07:06.519 eyes look weird in the output or there's this 00:07:06.519 --> 00:07:09.360 distracting pattern appearing. And the GPT will 00:07:09.360 --> 00:07:11.779 suggest changes to your prompt to try and fix 00:07:11.779 --> 00:07:15.319 it. It might say... Try adding photorealistic 00:07:15.319 --> 00:07:18.459 eyes to the main prompt, or add uncluttered background 00:07:18.459 --> 00:07:21.199 to the negative prompt, or maybe suggest tweaking 00:07:21.199 --> 00:07:23.160 the lighting description to focus more light 00:07:23.160 --> 00:07:25.560 clearly on the face. It helps you kind of zero 00:07:25.560 --> 00:07:27.459 in on what part of the prompt might be causing 00:07:27.459 --> 00:07:29.980 the error. I see. So the goal isn't just one 00:07:29.980 --> 00:07:32.540 perfect prompt, but actually a whole suite of 00:07:32.540 --> 00:07:34.699 prompts that are carefully engineered to keep 00:07:34.699 --> 00:07:37.199 that facial consistency, which I guess is the 00:07:37.199 --> 00:07:39.180 perfect lead in to actually creating the avatar 00:07:39.180 --> 00:07:41.800 itself. Right. So with those well -cracked prompts 00:07:41.800 --> 00:07:44.779 ready, we move over to Nano Banana. And here, 00:07:45.100 --> 00:07:47.279 consistency becomes like the absolute number 00:07:47.279 --> 00:07:49.279 one priority, doesn't it? If this image is going 00:07:49.279 --> 00:07:52.459 to become your talking avatar, that initial reference 00:07:52.459 --> 00:07:54.860 photo is, well, it sounds like it's the most 00:07:54.860 --> 00:07:57.160 important piece of the whole puzzle. Oh, it dictates 00:07:57.160 --> 00:07:59.370 everything that comes after. Absolutely. That 00:07:59.370 --> 00:08:01.170 photo, it needs to be high resolution. You need 00:08:01.170 --> 00:08:03.529 to be looking straight at the camera. The lighting 00:08:03.529 --> 00:08:06.449 has got to be good, really even, no harsh shadows, 00:08:06.629 --> 00:08:09.470 nothing dramatic, and critically, nothing blocking 00:08:09.470 --> 00:08:13.439 the face. So no hats, no scarves, no big sunglasses. 00:08:13.879 --> 00:08:16.180 You know, Nano Banana basically studies this 00:08:16.180 --> 00:08:19.399 one photo intensely to maintain that core facial 00:08:19.399 --> 00:08:21.620 structure and look across every single image 00:08:21.620 --> 00:08:23.759 you generate afterwards. OK, so once that core 00:08:23.759 --> 00:08:25.600 identity is kind of locked in from the reference 00:08:25.600 --> 00:08:28.740 photo, the goal shifts to building out an entire 00:08:28.740 --> 00:08:32.659 avatar library. Hmm. Wait, building a full avatar 00:08:32.659 --> 00:08:34.600 library? That sounds like potentially a lot of 00:08:34.600 --> 00:08:36.419 upfront work. Is that time investment really 00:08:36.419 --> 00:08:38.519 worth it compared to just generating images one 00:08:38.519 --> 00:08:41.419 by one as you need them? It is so worth it, especially 00:08:41.419 --> 00:08:43.820 when you get to the editing phase later. Trust 00:08:43.820 --> 00:08:46.820 me on this. If you only have one single image 00:08:46.820 --> 00:08:49.320 of your avatar, the final video is going to look 00:08:49.320 --> 00:08:51.720 really static and frankly kind of boring. Like 00:08:51.720 --> 00:08:55.440 a slightly fancier webcam video, you know? The 00:08:55.440 --> 00:08:58.059 goal is diversity, but built on that foundation 00:08:58.059 --> 00:09:01.179 of consistency. Moment of wonder. Whoa. I mean, 00:09:01.259 --> 00:09:03.679 just imagine scaling this ability, creating a 00:09:03.679 --> 00:09:05.860 totally consistent, personalized avatar that 00:09:05.860 --> 00:09:08.779 you can place in dozens, hundreds of different 00:09:08.779 --> 00:09:11.759 scenes. You build up this collection, your avatar 00:09:11.759 --> 00:09:13.200 looking straight, looking left, looking right, 00:09:13.299 --> 00:09:15.279 maybe arms crossed, pointing, different subtle 00:09:15.279 --> 00:09:18.740 expressions. This visual variety is the absolute 00:09:18.740 --> 00:09:21.820 key to creating a final video that's engaging 00:09:21.820 --> 00:09:24.340 and doesn't feel repetitive or, well, robotic. 00:09:24.480 --> 00:09:25.960 That makes a lot of sense, actually. We're shifting 00:09:25.960 --> 00:09:28.620 from just making a still photo to essentially 00:09:28.620 --> 00:09:30.899 planning shots for a film. And speaking of images 00:09:30.899 --> 00:09:32.879 that are ready for video, the source had five 00:09:32.879 --> 00:09:35.039 advanced tips. Starting with lighting, you mentioned 00:09:35.039 --> 00:09:37.639 avoiding dramatic lighting. Yeah, VEO3, the video 00:09:37.639 --> 00:09:40.019 tool. It just loves consistency and clarity. 00:09:40.360 --> 00:09:43.679 So you want prompts that specify soft, even light, 00:09:43.840 --> 00:09:46.899 or natural daylight. Nothing too moody or high 00:09:46.899 --> 00:09:49.460 contrast. This really helps ensure that when 00:09:49.460 --> 00:09:52.200 VEO3 generates the motion, it looks natural. 00:09:52.419 --> 00:09:55.240 It prevents weird flickering or shadows suddenly 00:09:55.240 --> 00:09:57.120 jumping around when the avatar starts to move 00:09:57.120 --> 00:09:59.919 or speak. even light is crucial for the video 00:09:59.919 --> 00:10:02.799 output. Got it. Even light prevents motion artifacts. 00:10:03.379 --> 00:10:05.360 And what about composition? You mentioned leaving 00:10:05.360 --> 00:10:07.960 room for movement. Correct. Don't crop the image 00:10:07.960 --> 00:10:10.840 too tightly around the face in Nano Banana. VEO3 00:10:10.840 --> 00:10:13.000 needs a bit of space, some headroom, and shoulder 00:10:13.000 --> 00:10:15.080 room to make the avatar's movements look natural. 00:10:15.539 --> 00:10:17.980 So stick to prompts like medium shot or chest 00:10:17.980 --> 00:10:20.399 up portrait. Give the AI some canvas to work 00:10:20.399 --> 00:10:23.299 with. Okay. And for people making, say, vertical 00:10:23.299 --> 00:10:25.659 content for social media. Right. While the standard 00:10:25.659 --> 00:10:28.299 is 16 .9 horizontal video, you can absolutely 00:10:28.299 --> 00:10:31.100 generate vertical images, too. Just add portrait 00:10:31.100 --> 00:10:34.259 orientation or specify 9 .16 aspect ratio in 00:10:34.259 --> 00:10:36.240 your Nano Banana prompt. Perfect for reels or 00:10:36.240 --> 00:10:38.639 TikToks. Good tip. And the last one was about 00:10:38.639 --> 00:10:41.539 creating a sequence. Yeah. If you're aiming for 00:10:41.539 --> 00:10:44.340 a really polished professional edit, don't just 00:10:44.340 --> 00:10:47.200 make one main shot. Create a little set of three 00:10:47.200 --> 00:10:49.659 images using slight variations of your prompt, 00:10:49.980 --> 00:10:52.059 like a main shot looking straight ahead, then 00:10:52.059 --> 00:10:53.700 maybe a slightly different angle looking off 00:10:53.669 --> 00:10:56.250 to the side, and perhaps a close up for emphasis. 00:10:56.809 --> 00:10:59.110 These act like building blocks in your video 00:10:59.110 --> 00:11:01.330 editor later, giving you options for cutting 00:11:01.330 --> 00:11:03.450 between shots, just like in real filmmaking. 00:11:03.710 --> 00:11:06.070 It makes the final output much more dynamic. 00:11:06.210 --> 00:11:08.789 OK, so image consistency is the bedrock. The 00:11:08.789 --> 00:11:11.789 library provides variety. And these tips help 00:11:11.789 --> 00:11:15.129 make the images truly video ready. Now I guess 00:11:15.129 --> 00:11:16.929 it's time to actually direct the performance. 00:11:17.129 --> 00:11:19.490 Exactly. Now for the really exciting part, where 00:11:19.490 --> 00:11:21.309 we kind of switch hats from being a painter or 00:11:21.309 --> 00:11:24.870 photographer to being a... director. First thing 00:11:24.870 --> 00:11:26.809 though, we have to understand and work with the 00:11:26.809 --> 00:11:29.409 fundamental constraint of Google VE03 right now. 00:11:29.750 --> 00:11:31.509 Clips are limited. They can only be up to eight 00:11:31.509 --> 00:11:34.309 seconds long. Right, eight seconds, which means 00:11:34.309 --> 00:11:36.350 you absolutely have to plan your script differently. 00:11:36.450 --> 00:11:38.669 You need to break it down into these short, almost 00:11:38.669 --> 00:11:40.750 punchy eight -second segments, each one needing 00:11:40.750 --> 00:11:43.909 to contain basically one complete idea, or roughly, 00:11:43.929 --> 00:11:46.529 what, 15 to 20 spoken words. Sounds like the 00:11:46.529 --> 00:11:48.230 planning stage is almost more critical than the 00:11:48.230 --> 00:11:50.509 rendering itself. Oh, it totally is. Meticulous 00:11:50.509 --> 00:11:53.090 planning saves huge amounts of time and credits 00:11:53.090 --> 00:11:56.990 later. And the VEO3 prompt structure? It's different 00:11:56.990 --> 00:11:59.029 from the image prompt. It's focused on directing 00:11:59.029 --> 00:12:01.549 motion, emotion, and sound, not just describing 00:12:01.549 --> 00:12:04.450 how things look. The basic structure that seems 00:12:04.450 --> 00:12:06.970 to work pretty well is something like this. Speaking. 00:12:07.309 --> 00:12:10.730 You put the emotion here in a busa, you specify 00:12:10.730 --> 00:12:13.549 accent nationality if needed accent, then you 00:12:13.549 --> 00:12:16.429 paste the exact words they're saying. Ah, okay, 00:12:16.470 --> 00:12:18.570 so you need specific emotional direction. not 00:12:18.570 --> 00:12:21.490 just speaking, but using active verbs like explaining 00:12:21.490 --> 00:12:24.370 calmly, or announcing excitedly, or maybe even 00:12:24.370 --> 00:12:27.190 whispering secretly. And those words drive the 00:12:27.190 --> 00:12:29.230 specific facial movements and expressions that 00:12:29.230 --> 00:12:32.450 VEO3 generates. That's the idea, exactly. And 00:12:32.450 --> 00:12:34.649 you need to keep that accent and general tone 00:12:34.649 --> 00:12:37.070 consistent across all your clips, otherwise it'll 00:12:37.070 --> 00:12:38.610 sound really jarring when you stitch them together. 00:12:38.799 --> 00:12:41.159 Makes sense. So for longer content, you have 00:12:41.159 --> 00:12:43.460 to master chaining these eight second clips together. 00:12:43.620 --> 00:12:46.139 That sounds, well, it sounds like microscripting 00:12:46.139 --> 00:12:48.659 almost. It kind of is, yeah. So what's the challenge 00:12:48.659 --> 00:12:51.700 there? How do you stop a sequence of these short 00:12:51.700 --> 00:12:56.259 clips from just feeling like a disjointed slideshow 00:12:56.259 --> 00:12:58.419 with talking heads? Right, that's the art of 00:12:58.419 --> 00:13:00.480 it. You do it by planning the emotional arc of 00:13:00.480 --> 00:13:03.159 your overall message and, crucially, by using 00:13:03.159 --> 00:13:05.919 those different avatar poses we generated earlier 00:13:05.919 --> 00:13:07.919 in Nano Banana. Maybe you start the sequence 00:13:07.919 --> 00:13:09.860 with the avatar looking thoughtful in a medium 00:13:09.860 --> 00:13:12.240 shot. Then, as the point gets more exciting, 00:13:12.519 --> 00:13:14.879 you cut to that side angle shot we created, and 00:13:14.879 --> 00:13:17.539 maybe you end on the confident, chest -up, straight 00:13:17.539 --> 00:13:20.059 -to -camera shot for the conclusion. That visual 00:13:20.059 --> 00:13:22.279 variety helps mask the cuts between the eight 00:13:22.279 --> 00:13:24.919 -second clips and makes it feel more like a continuous 00:13:24.919 --> 00:13:27.139 directed piece. Okay, that makes sense. using 00:13:27.139 --> 00:13:29.960 different shots to smooth transitions. But we 00:13:29.960 --> 00:13:31.919 should probably touch on troubleshooting, because 00:13:31.919 --> 00:13:34.039 let's be real, these tools aren't flawless yet, 00:13:34.080 --> 00:13:36.700 right? Not at all. Still early days in some ways. 00:13:36.879 --> 00:13:38.779 What about that common problem people mention, 00:13:39.200 --> 00:13:42.210 the mouth movements? not quite syncing up perfectly 00:13:42.210 --> 00:13:44.029 with the audio, the lip sync being a bit off. 00:13:44.289 --> 00:13:46.450 Yeah, that often happens when there's a mismatch 00:13:46.450 --> 00:13:49.129 between the emotion you put in the VEO3 text 00:13:49.129 --> 00:13:51.750 prompt and the expression on the original still 00:13:51.750 --> 00:13:54.049 image you fed it. So if you give it a picture 00:13:54.049 --> 00:13:56.210 where the avatar is frowning or looking really 00:13:56.210 --> 00:13:58.610 serious, but then you ask it to speak enthusiastically 00:13:58.610 --> 00:14:01.409 or happily, the lip movements can look really 00:14:01.409 --> 00:14:03.529 unnatural. You got to try and match them. Use 00:14:03.529 --> 00:14:05.730 a smiling picture if the text prompt is happy. 00:14:06.370 --> 00:14:08.830 Okay, match image expression to text emotion. 00:14:09.100 --> 00:14:12.419 What if the whole video just looks kind of shaky 00:14:12.419 --> 00:14:14.679 or jittery? That can sometimes happen if the 00:14:14.679 --> 00:14:16.940 original Nano Banana image background was too 00:14:16.940 --> 00:14:20.580 complex or detailed. Try simplifying it. Generate 00:14:20.580 --> 00:14:22.980 an image with a cleaner, maybe slightly blurred 00:14:22.980 --> 00:14:25.419 background. Simpler backgrounds often lead to 00:14:25.419 --> 00:14:27.539 smoother, less artifact -filled video motion 00:14:27.539 --> 00:14:30.519 from VEO3. And what if things in the background, 00:14:31.120 --> 00:14:32.960 like, I don't know, plants on a shelf or a necklace 00:14:32.960 --> 00:14:35.779 the avatar is wearing, start moving weirdly on 00:14:35.779 --> 00:14:37.860 their own in the final video? Right, the rogue 00:14:37.860 --> 00:14:41.080 moving objects. Yeah, for that, try using VE03 00:14:41.080 --> 00:14:43.379 prompts that really focus the AI's attention 00:14:43.379 --> 00:14:46.220 on the face and minimize its attempts to animate 00:14:46.220 --> 00:14:49.460 the background. Things like adding speaking directly 00:14:49.460 --> 00:14:52.519 to the camera or specifying close -up portrait, 00:14:52.799 --> 00:14:55.480 shallow depth of field can sometimes help lock 00:14:55.480 --> 00:14:57.360 down the background elements. So fundamentally, 00:14:57.440 --> 00:15:00.120 what's the most critical difference then, between 00:15:00.120 --> 00:15:02.740 prompting for a still image and prompting for 00:15:02.740 --> 00:15:05.299 a moving video. Video prompts focus on directing 00:15:05.299 --> 00:15:07.919 emotion, motion, and sound, not just describing 00:15:07.919 --> 00:15:10.799 visuals. It's about performance. Hashtag, tag, 00:15:10.960 --> 00:15:13.460 tag, big idea recap and real world application. 00:15:13.820 --> 00:15:16.600 Okay, wow. That's a lot to take in, but it feels 00:15:16.600 --> 00:15:18.559 like a complete process. Let's quickly recap 00:15:18.559 --> 00:15:20.580 those five phases again, just to nail it down. 00:15:21.039 --> 00:15:23.340 The whole journey from idea to finished video. 00:15:23.799 --> 00:15:26.600 Sounds good. Phase one. Get ready. That's setting 00:15:26.600 --> 00:15:29.120 up your tools remembering to use the VEO3 flow 00:15:29.120 --> 00:15:31.840 interface. Yeah. Preparing that really good high 00:15:31.840 --> 00:15:34.179 quality reference photo and getting your project 00:15:34.179 --> 00:15:36.179 folder's organelles from the start saves headaches 00:15:36.179 --> 00:15:40.179 later. Phase two, create prompts. Use those specialized 00:15:40.179 --> 00:15:43.840 custom GPTs to write detailed cinematic and importantly 00:15:43.840 --> 00:15:46.860 emotion rich descriptions for the images you 00:15:46.860 --> 00:15:50.840 want. Phase three. Create images. Then you hop 00:15:50.840 --> 00:15:53.379 over to Nano Banana, upload your reference photo, 00:15:53.620 --> 00:15:55.039 paste in those prompts you just created, and 00:15:55.039 --> 00:15:57.000 start building out that diverse avatar library, 00:15:57.159 --> 00:15:59.639 different poses, angles, maybe expressions. Phase 00:15:59.639 --> 00:16:02.379 four, create videos. Upload those finished images 00:16:02.379 --> 00:16:05.159 into VEO3 Flow. Use the fast mode extensively 00:16:05.159 --> 00:16:06.820 for testing your eight -second script chunks 00:16:06.820 --> 00:16:09.039 and prompts. Get things right there before using 00:16:09.039 --> 00:16:10.840 the more expensive quality mode for your final 00:16:10.840 --> 00:16:13.820 renders. Exactly. And then finally, phase five. 00:16:14.009 --> 00:16:17.049 Edit and finish. Take all those eight second 00:16:17.049 --> 00:16:19.830 clips, bring them into a video editor, free ones 00:16:19.830 --> 00:16:23.070 like CapCut or DaVinci Resolve, work great. Stitch 00:16:23.070 --> 00:16:25.330 them together in sequence, add maybe some background 00:16:25.330 --> 00:16:27.990 music, titles, and boom, you've got a cohesive, 00:16:28.309 --> 00:16:30.549 professional -looking video. And the source material 00:16:30.549 --> 00:16:33.269 suggests this whole thing, once you're practiced, 00:16:33.710 --> 00:16:36.429 is potentially doable in, like, an afternoon. 00:16:36.950 --> 00:16:39.809 That's pretty incredible. The real world uses... 00:16:39.600 --> 00:16:41.259 They feel pretty transformative, don't they? 00:16:41.460 --> 00:16:43.519 For content creators, obviously this is huge. 00:16:43.700 --> 00:16:45.940 It completely removes camera shyness, the need 00:16:45.940 --> 00:16:49.059 for expensive gear or a studio space, plus that 00:16:49.059 --> 00:16:52.120 idea of creating multilingual versions just by 00:16:52.120 --> 00:16:54.919 changing the text prompt in VEO3. That's potentially 00:16:54.919 --> 00:16:57.100 massive for reaching global audiences with the 00:16:57.100 --> 00:16:59.720 same core video. Oh yeah. And for businesses, 00:16:59.980 --> 00:17:02.279 think about it. Quick product demos, consistent 00:17:02.279 --> 00:17:04.200 professional employee training videos delivered 00:17:04.200 --> 00:17:06.720 by a familiar avatar, or generating large -scale 00:17:06.720 --> 00:17:09.099 marketing content variations. You could test 00:17:09.099 --> 00:17:11.140 hundreds of ad angles in an afternoon because 00:17:11.140 --> 00:17:13.140 there's no physical production cost or delay. 00:17:13.400 --> 00:17:16.000 It's almost risk -free A -B testing for video 00:17:16.000 --> 00:17:18.859 creative. Totally. And even for personal projects, 00:17:19.099 --> 00:17:21.279 it just opens up so much flexibility, right? 00:17:21.279 --> 00:17:23.740 Yeah. Unique personalized birthday messages, 00:17:23.980 --> 00:17:26.420 keeping up a regular social media presence without 00:17:26.420 --> 00:17:28.880 the constant pressure of filming. yourself. It 00:17:28.880 --> 00:17:31.019 really feels like a risk -free sandbox to experiment 00:17:31.019 --> 00:17:33.279 with your presentation style, your personal brand, 00:17:33.660 --> 00:17:35.960 or testing ideas for your business. Absolutely, 00:17:36.079 --> 00:17:38.339 and the key thing that makes all that possible 00:17:38.339 --> 00:17:41.200 is nailing the consistency of the avatar image 00:17:41.200 --> 00:17:43.759 first. That unlocks all the creative freedom 00:17:43.759 --> 00:17:46.099 you have in the video generation stage later. 00:17:46.339 --> 00:17:49.200 Hashtag tag tag outro. So the big takeaway here 00:17:49.200 --> 00:17:51.890 feels like This is a skill, it's learnable, and 00:17:51.890 --> 00:17:54.589 it's available to you like today. The key really 00:17:54.589 --> 00:17:56.690 seems to be just starting small, maybe with those 00:17:56.690 --> 00:17:59.369 free tools and trials. Focus on building up that 00:17:59.369 --> 00:18:02.289 prompt library and really mastering that consistency 00:18:02.289 --> 00:18:04.210 process, especially with a reference photo and 00:18:04.210 --> 00:18:07.670 the lighting in your images. Yeah, exactly. The 00:18:07.670 --> 00:18:09.809 technology itself, it's changing almost daily, 00:18:09.869 --> 00:18:12.829 it feels like. But the source material really 00:18:12.829 --> 00:18:14.730 reminded us of something enduring, didn't it? 00:18:14.890 --> 00:18:17.910 The basic rules of good storytelling, clear communication, 00:18:18.410 --> 00:18:20.710 and actually creating valuable content for people. 00:18:21.309 --> 00:18:22.849 Those things always stay the same. Doesn't matter 00:18:22.849 --> 00:18:24.730 if there's a camera involved or not. That's a 00:18:24.730 --> 00:18:27.210 great point. So maybe the final thought for everyone 00:18:27.210 --> 00:18:29.809 listening is this. If the camera is no longer 00:18:29.809 --> 00:18:32.450 the obstacle, what's the story you're finally 00:18:32.450 --> 00:18:32.930 going to tell?