WEBVTT 00:00:00.000 --> 00:00:03.740 Imagine for a moment just wishing you could conjure 00:00:03.740 --> 00:00:07.620 a scene, you know, maybe a gritty detective walking 00:00:07.620 --> 00:00:10.460 through these neon streets, rain everywhere, 00:00:10.980 --> 00:00:12.720 or I don't know, something totally different, 00:00:12.759 --> 00:00:15.720 like a fluffy kitten landing a skateboard trick, 00:00:16.359 --> 00:00:18.519 that kind of thing. It felt like pure sci -fi, 00:00:19.000 --> 00:00:24.300 right? A dream. But now it's becoming remarkably 00:00:24.300 --> 00:00:28.469 real to sex silence. Welcome to the Deep Dive. 00:00:28.609 --> 00:00:31.250 We try to unpack complex stuff, make it digestible, 00:00:31.370 --> 00:00:34.149 maybe even a bit exciting. Today, we're plunging 00:00:34.149 --> 00:00:36.710 into Google Veo. It feels like a really new frontier 00:00:36.710 --> 00:00:39.590 in AI video. It's pretty fascinating, and we've 00:00:39.590 --> 00:00:41.030 gathered quite a bit to share with you. Absolutely. 00:00:41.070 --> 00:00:42.929 We're going to look at how this tool really shakes 00:00:42.929 --> 00:00:45.909 up video creation, especially with its native 00:00:45.909 --> 00:00:48.250 audio capabilities. That's a huge deal. And then, 00:00:48.310 --> 00:00:50.109 yeah, we'll break down this Golder formula for 00:00:50.109 --> 00:00:51.409 writing the perfect prompts. That's where the 00:00:51.409 --> 00:00:54.219 magic happens. We'll also explore Vio's two main 00:00:54.219 --> 00:00:56.200 ways to use it. The interfaces, we'll look at 00:00:56.200 --> 00:00:58.500 some advanced tips, touch on the current limits 00:00:58.500 --> 00:01:00.320 and the ethical side too, and maybe even get 00:01:00.320 --> 00:01:01.920 a little glimpse of where this tech is heading. 00:01:02.299 --> 00:01:04.439 Yeah, get ready to maybe turn some of your wild 00:01:04.439 --> 00:01:07.879 ideas into actual moving pictures. With sound, 00:01:08.400 --> 00:01:09.819 straight from text. It's kind of like having 00:01:09.819 --> 00:01:13.120 a mini film crew on call. Okay, let's dive in. 00:01:13.299 --> 00:01:16.060 So just to start, what is Google Vio fundamentally? 00:01:16.230 --> 00:01:19.370 It's an advanced AI model for generating video. 00:01:19.730 --> 00:01:21.989 Think of it like an AI that creates short video 00:01:21.989 --> 00:01:24.250 clips, usually up to eight seconds. Just from 00:01:24.250 --> 00:01:26.689 the words you type, you describe, it creates. 00:01:26.930 --> 00:01:28.469 And here's where it gets really interesting. 00:01:29.030 --> 00:01:32.650 Like, game changer interesting. Native audio 00:01:32.650 --> 00:01:35.590 generation. This is a massive step. Remember 00:01:35.590 --> 00:01:38.989 the older tools. Silent videos. Which meant this 00:01:38.989 --> 00:01:41.090 whole clunky process, right? Generate the video, 00:01:41.469 --> 00:01:43.870 then hunt for sounds, license music, spend ages 00:01:43.870 --> 00:01:46.989 syncing it all up. Yeah, it sounds painful. Vio 00:01:46.989 --> 00:01:49.549 just breaks through that. It seems to understand, 00:01:49.569 --> 00:01:52.310 like right from the start, how visuals and sound 00:01:52.310 --> 00:01:54.909 belong together. So if you describe, say, the 00:01:54.909 --> 00:01:56.849 pitter -patter of rain on a window. Exactly. 00:01:56.909 --> 00:01:58.569 It doesn't just make the rain image. It makes 00:01:58.569 --> 00:02:00.750 the sound of the rain hitting the glass all at 00:02:00.750 --> 00:02:03.510 once. Unified. Or like your example, a chef chopping 00:02:03.510 --> 00:02:06.659 vegetables fast. Right. You don't just see it. 00:02:06.659 --> 00:02:09.120 You hear that rhythmic clack clack clack of the 00:02:09.120 --> 00:02:11.840 knife Maybe a little sizzle from a pan nearby 00:02:11.840 --> 00:02:15.199 all generated together perfectly think and that 00:02:15.199 --> 00:02:18.159 eight second length Sounds short maybe but it's 00:02:18.159 --> 00:02:21.319 perfect for tick -tock for reels quick punchy 00:02:21.319 --> 00:02:23.719 stuff So for someone making that kind of fast 00:02:23.719 --> 00:02:26.919 content, how does this native audio really change 00:02:26.919 --> 00:02:30.300 their game? Like practically it just brings visuals 00:02:30.300 --> 00:02:32.639 and sound together instantly makes producing 00:02:32.639 --> 00:02:35.240 that quick stuff way faster, and honestly much 00:02:35.240 --> 00:02:37.680 richer. Okay, so we know what it does, but how 00:02:37.680 --> 00:02:40.360 do you actually use it? Google set up like two 00:02:40.360 --> 00:02:42.259 different doors into Vio, right? For different 00:02:42.259 --> 00:02:44.360 needs. That's right. First you got Google Gemini. 00:02:44.860 --> 00:02:47.240 Think of that as the easy entry point. Kind of 00:02:47.240 --> 00:02:49.319 like a chatbot, but for video. Great for playing 00:02:49.319 --> 00:02:51.680 around, rapid ideas if you're just starting out, 00:02:51.900 --> 00:02:54.199 or for making standalone things like memes or 00:02:54.199 --> 00:02:56.719 quick clips. Sparks inspiration fast. And then 00:02:56.719 --> 00:02:59.319 the other door is Google Flow. That sounds more 00:02:59.319 --> 00:03:01.979 serious. Yeah, Flow is much more like a pro -AI 00:03:01.979 --> 00:03:03.900 film studio. It's built for the bigger, more 00:03:03.900 --> 00:03:06.219 complex stuff. Projects where you need more control, 00:03:06.599 --> 00:03:08.819 maybe tell a bit of a story across multiple shots. 00:03:09.259 --> 00:03:11.419 What makes Flow stand out then? What are the 00:03:11.419 --> 00:03:13.379 key features there? Well, storyboarding is a 00:03:13.379 --> 00:03:15.500 big one. You can actually lay out a sequence 00:03:15.500 --> 00:03:18.520 of scenes, build a narrative, and crucially, 00:03:18.680 --> 00:03:20.639 it tackles character consistency. That's been 00:03:20.639 --> 00:03:23.159 a huge headache with AI video. You can upload 00:03:23.159 --> 00:03:25.599 an image of your character, and Flow tries to 00:03:25.599 --> 00:03:27.729 keep them looking the same. across different 00:03:27.729 --> 00:03:30.509 shots. Face, clothes, everything. No more weird 00:03:30.509 --> 00:03:33.990 morphing. Plus, you get manual camera controls, 00:03:34.349 --> 00:03:36.949 better scene building tools, and ways to manage 00:03:36.949 --> 00:03:39.310 all your project assets. It's much more robust. 00:03:39.430 --> 00:03:41.689 And there are subscriptions involved. Yep. There's 00:03:41.689 --> 00:03:45.050 a pro plan with VO3 fast for quicker results 00:03:45.050 --> 00:03:48.469 and an ultra plan with the full quality VO3 for 00:03:48.469 --> 00:03:51.310 the absolute best output. Both plans get you 00:03:51.310 --> 00:03:53.979 that native audio, which is key. So Flow sounds 00:03:53.979 --> 00:03:56.360 like it's tackling some major AI video hurdles, 00:03:56.439 --> 00:03:58.439 especially that character consistency you mentioned. 00:03:58.520 --> 00:04:00.819 Is that the biggest thing it solves? It's definitely 00:04:00.819 --> 00:04:03.539 huge, especially for telling stories. Flow lets 00:04:03.539 --> 00:04:05.699 you actually build a sequence, not just generate 00:04:05.699 --> 00:04:08.680 isolated clips. That's a big shift. OK, this 00:04:08.680 --> 00:04:10.740 next part feels really important. You mentioned 00:04:10.740 --> 00:04:13.400 a golden formula for prompts. This seems like 00:04:13.400 --> 00:04:15.740 where you really unlock video's power, right? 00:04:16.019 --> 00:04:18.120 Because vague prompts probably don't work well. 00:04:18.319 --> 00:04:20.220 Exactly. Most beginners just type something simple, 00:04:20.319 --> 00:04:23.680 like dog running. But the pros? Use a more structured 00:04:23.680 --> 00:04:25.899 approach. We're calling it the seven element 00:04:25.899 --> 00:04:28.560 formula. Think of it like a checklist for describing 00:04:28.560 --> 00:04:31.000 a mini movie scene to the AI. OK, break it down 00:04:31.000 --> 00:04:35.240 for us. Element one. One. Subject, who or what. 00:04:35.660 --> 00:04:38.920 Be super specific, not just a man. Try a grizzled 00:04:38.920 --> 00:04:41.759 detective weary in a worn beige trench coat. 00:04:42.300 --> 00:04:45.120 Details matter. Got it. Two. Action. What are 00:04:45.120 --> 00:04:48.420 they doing? Use strong verbs. Not, he walks, 00:04:48.879 --> 00:04:50.779 but he trudges wearily through the downpour. 00:04:50.899 --> 00:04:53.199 Makes sense. Three. Context. Where and when. 00:04:53.459 --> 00:04:55.439 Set the scene. Build the world like in a narrow 00:04:55.439 --> 00:04:57.740 Tokyo alley soaked in pulsating neon light on 00:04:57.740 --> 00:05:00.699 a rainy night. Okay. Number four. Motion. How's 00:05:00.699 --> 00:05:03.019 the camera moving? This is crucial for that cinematic 00:05:03.019 --> 00:05:06.300 feel. Is it a pan? A tilt? A dolly or a tracking 00:05:06.300 --> 00:05:08.879 shot moving with the subject? A zoom? A drone 00:05:08.879 --> 00:05:12.360 shot from above? Or maybe handheld for that raw 00:05:12.360 --> 00:05:16.839 feel. Specify it. Style. The visual look. Genre. 00:05:17.259 --> 00:05:19.339 Artistic influence. Could be in the style of 00:05:19.339 --> 00:05:21.759 Wes Anderson. Black and white film noir. Studio 00:05:21.759 --> 00:05:24.379 Ghibli animation. Get creative. Yeah. Framing. 00:05:25.000 --> 00:05:27.139 How is the shot composed? Establishing shot. 00:05:27.519 --> 00:05:31.920 Wide. Medium. Close up. Extreme close up. Each 00:05:31.920 --> 00:05:33.939 tells a different story. And the last one, seven. 00:05:34.259 --> 00:05:37.449 Audio. Crucial for VO, always add an audio section. 00:05:37.949 --> 00:05:41.050 Describe everything. Sound effects, SFX, background 00:05:41.050 --> 00:05:43.470 noise, music mood, dialogue snippets, ambient 00:05:43.470 --> 00:05:46.149 sounds. Make it immersive. And you had a pro 00:05:46.149 --> 00:05:48.480 tip too. Oh yeah, super important. Always had 00:05:48.480 --> 00:05:50.839 no subtitles at the end. Otherwise, the AI might 00:05:50.839 --> 00:05:52.819 just slap some random text on your video. It 00:05:52.819 --> 00:05:54.759 can be annoying to deal with. You know, honestly, 00:05:55.079 --> 00:05:57.220 even knowing all these steps, I still sometimes 00:05:57.220 --> 00:05:59.759 struggle with prompt drift myself. Where you 00:05:59.759 --> 00:06:02.199 refine it, but the AI kind of wanders off from 00:06:02.199 --> 00:06:04.439 what you first wanted, it really does take practice. 00:06:04.519 --> 00:06:06.519 Oh, absolutely. It's an iterative process. So 00:06:06.519 --> 00:06:08.879 just to underline it, how does using all seven 00:06:08.879 --> 00:06:11.740 elements really elevate a simple idea compared 00:06:11.740 --> 00:06:14.189 to just, you know, Detective and Rain. It gives 00:06:14.189 --> 00:06:17.769 the AI such a rich, detailed blueprint. It lets 00:06:17.769 --> 00:06:20.290 it generate something truly unique, specific, 00:06:20.589 --> 00:06:23.730 and cinematic, not just generic. Right. And you 00:06:23.730 --> 00:06:25.569 mentioned iteration. It sounds like you rarely 00:06:25.569 --> 00:06:27.730 get it perfect on the first go. Almost never. 00:06:28.170 --> 00:06:30.230 The best way to work with VO, or really any of 00:06:30.230 --> 00:06:33.269 these tools, is iteration. Refinement. It's more 00:06:33.269 --> 00:06:35.389 like sculpting than just hitting generate. So 00:06:35.389 --> 00:06:37.910 what's the recommended process there? Start simple. 00:06:38.689 --> 00:06:41.990 Seriously. Just subject plus action plus context. 00:06:42.199 --> 00:06:45.259 Generate that, see what the AI gives you generically, 00:06:45.879 --> 00:06:49.079 then layer it, add motion and framing, generate 00:06:49.079 --> 00:06:51.160 again, see how it starts feeling more like a 00:06:51.160 --> 00:06:54.300 movie shot? Finally, add the last layer, style 00:06:54.300 --> 00:06:56.639 and audio, then tweak and polish from there. 00:06:57.240 --> 00:06:58.839 Breaking it down like that gives you way more 00:06:58.839 --> 00:07:00.600 control and stops you from feeling overwhelmed. 00:07:00.800 --> 00:07:02.500 That makes a lot of sense. And you can use other 00:07:02.500 --> 00:07:04.120 tools to help, right? You don't have to invent 00:07:04.120 --> 00:07:06.339 the perfect prompt from scratch. Definitely not. 00:07:06.740 --> 00:07:08.899 Don't feel like you have to be a master screenwriter 00:07:08.899 --> 00:07:12.839 instantly. LMs like Chat GPT can be amazing creative 00:07:12.839 --> 00:07:16.000 partners. Give it a basic idea. Ask it to flesh 00:07:16.000 --> 00:07:18.160 it out cinematically. Ask it to think about light, 00:07:18.639 --> 00:07:21.800 sound, emotion. Then take that richer description 00:07:21.800 --> 00:07:25.319 and refine it for your VO prompt. Great brainstorming 00:07:25.319 --> 00:07:27.600 tool. And what about nailing a specific visual 00:07:27.600 --> 00:07:29.740 style? Sometimes words are hard for that. Totally. 00:07:30.120 --> 00:07:31.839 That's where image generators like Mid Journey 00:07:31.839 --> 00:07:34.420 come in handy. You can generate still images 00:07:34.420 --> 00:07:36.959 first to really pin down the exact look you want, 00:07:37.120 --> 00:07:39.300 the lighting, the colors, the vibe. Once you 00:07:39.300 --> 00:07:41.899 have a still image you love, then describe that 00:07:41.899 --> 00:07:44.899 style very precisely in your VO prompt. It's 00:07:44.899 --> 00:07:47.339 like visual prototyping. So for someone just 00:07:47.339 --> 00:07:49.980 diving into this iterative process with VO, what's 00:07:49.980 --> 00:07:52.500 the main piece of advice? Don't chase perfection 00:07:52.500 --> 00:07:55.000 right away. Build it up. Refine it in layers, 00:07:55.699 --> 00:07:58.220 step by step. OK, so it's powerful, but obviously 00:07:58.220 --> 00:08:01.110 not. perfect yet. What are some common mistakes 00:08:01.110 --> 00:08:03.449 or pitfalls people should watch out for? Well, 00:08:03.589 --> 00:08:05.889 we've hit on vague prompts already. That's number 00:08:05.889 --> 00:08:08.649 one. Forgetting the audio instructions is another 00:08:08.649 --> 00:08:11.470 big one, given VO's strength there. Forgetting 00:08:11.470 --> 00:08:13.930 no subtitles, trying to cram complex dialogue 00:08:13.930 --> 00:08:16.790 into an eight -second clip rarely works well. 00:08:17.250 --> 00:08:19.410 And just completely ignoring camera movement 00:08:19.410 --> 00:08:22.589 that makes videos feel very static and, well, 00:08:22.730 --> 00:08:24.610 AI -generated. And there are technical limits 00:08:24.610 --> 00:08:27.089 too right now. Yeah, definitely. That eight -second 00:08:27.089 --> 00:08:29.290 clip length is one. It's great for shorts, but 00:08:29.290 --> 00:08:32.149 not for long scenes yet. It outputs in HD, which 00:08:32.149 --> 00:08:35.340 is good. But processing times can vary. Sometimes 00:08:35.340 --> 00:08:37.899 it's quick, sometimes you wait a bit. And yeah, 00:08:37.899 --> 00:08:40.240 you need that Google AI Pro or Ultra subscription 00:08:40.240 --> 00:08:43.120 to access it. And this tech, like a lot of AI, 00:08:43.279 --> 00:08:45.820 brings up some pretty significant ethical questions. 00:08:46.159 --> 00:08:48.440 Huge ones. The ability to make realistic videos 00:08:48.440 --> 00:08:51.659 easily? Well, it opens the door to misuse. Deep 00:08:51.659 --> 00:08:53.860 fakes are a major concern, creating fake videos 00:08:53.860 --> 00:08:56.159 of people saying or doing things they never did. 00:08:56.639 --> 00:08:59.019 Misinformation potential is high. And copyright 00:08:59.019 --> 00:09:01.740 is this massive gray area that AI trains on vast 00:09:01.740 --> 00:09:04.740 amounts of data. Was that data copyrighted? And 00:09:04.740 --> 00:09:06.740 who owns the video you create? Is it fully yours? 00:09:07.200 --> 00:09:09.240 Or does Google have some claim because their 00:09:09.240 --> 00:09:11.360 AI made it? These are big legal questions right 00:09:11.360 --> 00:09:13.460 now. What about the impact on creative jobs, 00:09:13.799 --> 00:09:16.120 stock video creators, animators? It's a valid 00:09:16.120 --> 00:09:18.440 concern for sure. There will likely be disruption. 00:09:19.100 --> 00:09:22.419 But the thinking is this will probably evolve 00:09:22.419 --> 00:09:25.100 into more of an assistive tool, letting creators 00:09:25.100 --> 00:09:27.980 work faster, maybe handle the more tedious parts 00:09:27.980 --> 00:09:30.299 and focus their energy on the higher level storytelling, 00:09:30.379 --> 00:09:34.019 the core ideas. Vio isn't really the end of human 00:09:34.019 --> 00:09:35.779 creativity. It feels more like a new chapter, 00:09:35.899 --> 00:09:40.899 you know? Human -machine collaboration. Just 00:09:40.899 --> 00:09:43.139 imagine, though, scaling this up. Imagine generating 00:09:43.139 --> 00:09:45.919 entire feature films just from detailed text. 00:09:46.080 --> 00:09:49.139 the sheer potential for individual creators to 00:09:49.139 --> 00:09:51.620 bring massive stories to life, bypassing all 00:09:51.620 --> 00:09:53.179 the usual production barriers. That's kind of 00:09:53.179 --> 00:09:54.480 mind -boggling to think about. We're not there 00:09:54.480 --> 00:09:57.759 yet, but... Wow. When you look at all the power, 00:09:57.919 --> 00:10:00.120 the ethics, the potential, what's a core message 00:10:00.120 --> 00:10:02.820 about how AI, like Vio, is changing creativity? 00:10:02.980 --> 00:10:05.279 I think it's fundamentally a tool for collaboration. 00:10:05.580 --> 00:10:08.240 It's about augmenting and enhancing what humans 00:10:08.240 --> 00:10:11.120 can do creatively, not just replacing them. So 00:10:11.120 --> 00:10:14.440 stepping back to see the big picture here... 00:10:14.570 --> 00:10:17.309 Google Veo feels like it's genuinely transforming 00:10:17.309 --> 00:10:19.990 visual storytelling. It's more than just a neat 00:10:19.990 --> 00:10:22.710 piece of tech. It feels like a statement about 00:10:22.710 --> 00:10:24.409 where the creative industry might be heading, 00:10:24.730 --> 00:10:27.490 democratizing things. Absolutely. That ability 00:10:27.490 --> 00:10:29.429 to go straight from language from an idea in 00:10:29.429 --> 00:10:33.269 your head to a complete audio visual piece. that 00:10:33.269 --> 00:10:36.610 lowers the barrier to entry massively. It really 00:10:36.610 --> 00:10:39.070 does democratize video production in a way. And 00:10:39.070 --> 00:10:41.690 that golden formula, that seven step prompt structure, 00:10:42.169 --> 00:10:43.850 that's your mental toolkit for making it work. 00:10:43.970 --> 00:10:46.639 It's like the key. So the skill isn't necessarily 00:10:46.639 --> 00:10:48.820 about having the fanciest camera anymore. It's 00:10:48.820 --> 00:10:51.039 shifting towards thinking like a director, writing 00:10:51.039 --> 00:10:53.240 like a screenwriter, really envisioning the final 00:10:53.240 --> 00:10:55.779 scene. Exactly. Start simple. Play around with 00:10:55.779 --> 00:10:58.620 that seven element formula. And honestly, don't 00:10:58.620 --> 00:11:00.899 be afraid to mix things up. Combine unexpected 00:11:00.899 --> 00:11:03.600 elements. Sometimes the weird combinations lead 00:11:03.600 --> 00:11:05.919 to the coolest results. Yeah. Like you said, 00:11:06.000 --> 00:11:08.299 a film noir scene, but with cartoon characters. 00:11:08.580 --> 00:11:10.759 Or a nature documentary shot, like a sci -fi 00:11:10.759 --> 00:11:12.820 epic. That's where you can really push the AI 00:11:12.820 --> 00:11:15.620 into interesting territory. The AI video revolution 00:11:15.620 --> 00:11:18.399 is definitely here. It's happening now. And hopefully, 00:11:18.659 --> 00:11:20.480 with what we've talked about, you feel a bit 00:11:20.480 --> 00:11:23.659 more equipped to jump in and be part of it. Go 00:11:23.659 --> 00:11:25.879 make something cool. We really hope this deep 00:11:25.879 --> 00:11:28.139 dive gave you some useful insights, some valuable 00:11:28.139 --> 00:11:30.440 nuggets to think about. Until next time, keep 00:11:30.440 --> 00:11:32.039 exploring, out of your row music.