WEBVTT 00:00:00.000 --> 00:00:02.560 If you think back just a few years ago, getting 00:00:02.560 --> 00:00:06.280 a professional cinematic video ad, that was a 00:00:06.280 --> 00:00:09.460 huge undertaking. Oh, yeah. Weeks, maybe months 00:00:09.460 --> 00:00:11.419 of production. You're talking about hiring the 00:00:11.419 --> 00:00:14.539 full crew, finding locations, and spending, what, 00:00:14.660 --> 00:00:16.280 tens of thousands of dollars before you even 00:00:16.280 --> 00:00:18.120 see a single frame. That was the cost of entry, 00:00:18.260 --> 00:00:20.929 right? A huge barrier. But the reality today, 00:00:21.149 --> 00:00:23.469 I mean, it's completely changed the game. It 00:00:23.469 --> 00:00:25.929 has. We have these incredible generative tools 00:00:25.929 --> 00:00:29.410 now, you know, Nano Banana Pro for images and 00:00:29.410 --> 00:00:33.329 VO 3 .1 for adding motion. And these tools allow 00:00:33.329 --> 00:00:36.429 you, the creator or the small team, to generate 00:00:36.429 --> 00:00:40.009 really high -end story -driven ads from, well, 00:00:40.049 --> 00:00:41.899 just a text prompt. And it happens in hours, 00:00:42.060 --> 00:00:44.460 not weeks. Exactly. It's a fundamental shift. 00:00:44.579 --> 00:00:46.740 It is. And the source material we're looking 00:00:46.740 --> 00:00:49.600 at today gives us the perfect shortcut. It lays 00:00:49.600 --> 00:00:53.219 out a precise framework for key steps to start 00:00:53.219 --> 00:00:56.259 filming and start generating. Welcome to the 00:00:56.259 --> 00:00:58.500 deep dive. Our mission today is to unpack that 00:00:58.500 --> 00:01:01.520 guide on cinematic AI video ad creation. We're 00:01:01.520 --> 00:01:03.859 going to distill it down to that essential, repeatable 00:01:03.859 --> 00:01:06.359 four -step process. And it really is a process, 00:01:06.400 --> 00:01:09.200 like a workflow. And the biggest insight right 00:01:09.200 --> 00:01:11.400 up front is that you have to respect all four 00:01:11.400 --> 00:01:14.319 steps. If you get excited and just jump straight 00:01:14.319 --> 00:01:17.519 to generating the video. Or you rush the planning. 00:01:17.719 --> 00:01:20.000 The output is just going to be a mess. It's going 00:01:20.000 --> 00:01:22.859 to be disjointed, unusable, a total waste of 00:01:22.859 --> 00:01:24.239 time. So we're going to walk through this in 00:01:24.239 --> 00:01:27.140 order. Step one is storyboarding. The creative 00:01:27.140 --> 00:01:30.400 foundation. Then step two, image creation, which 00:01:30.400 --> 00:01:33.640 is our visual anchor. Step three is video generation. 00:01:33.980 --> 00:01:37.060 Where the magic happens. And finally, step four, 00:01:37.180 --> 00:01:39.560 final production, stitching it all together. 00:01:39.659 --> 00:01:41.739 Let's dive into that first step, storyboarding. 00:01:41.939 --> 00:01:45.280 This is, it's not technical at all. Right. It's 00:01:45.280 --> 00:01:48.299 all about defining a super clear creative vision 00:01:48.299 --> 00:01:50.620 and then breaking it down into separate scenes 00:01:50.620 --> 00:01:53.159 before you even touch a generative tool. And 00:01:53.159 --> 00:01:55.379 this is a great place to use AI assistance. You 00:01:55.379 --> 00:01:59.760 know, Chad GPT, Claude, Gemini, not as the creator, 00:01:59.780 --> 00:02:02.079 but as an accelerator. Totally. You use them 00:02:02.079 --> 00:02:04.299 conversationally, you feed them your concept, 00:02:04.319 --> 00:02:06.260 and they help flesh it out, break it down into 00:02:06.260 --> 00:02:08.439 maybe seven or ten scenes. It helps you get past 00:02:08.439 --> 00:02:10.990 that blank page problem. Exactly. You can have 00:02:10.990 --> 00:02:14.270 a high level idea like I want a high energy running 00:02:14.270 --> 00:02:16.770 ad and it'll suggest specific shots, specific 00:02:16.770 --> 00:02:19.229 angles. But you're the curator. You're always 00:02:19.229 --> 00:02:21.569 the one in charge. Always. The source had a great 00:02:21.569 --> 00:02:24.889 example concept, a Nike running ad, but with 00:02:24.889 --> 00:02:28.750 this futuristic kind of Blade Runner City vibe. 00:02:28.969 --> 00:02:32.590 Yeah, that aesthetic is key. Sleek, tech forward. 00:02:33.389 --> 00:02:36.449 It fits the brand so perfectly. It does. And 00:02:36.449 --> 00:02:39.439 to get that look, you need... A specific sequence 00:02:39.439 --> 00:02:41.919 of shots. Right. So you'd outline, let's say, 00:02:41.939 --> 00:02:45.500 seven scenes. You start wide, distant shot of 00:02:45.500 --> 00:02:48.259 a neon skyline. Then you move to an empty, illuminated 00:02:48.259 --> 00:02:51.219 street. Then the really important one. A close 00:02:51.219 --> 00:02:53.560 -up of the shoes hitting wet pavement. Followed 00:02:53.560 --> 00:02:55.580 by the runner accelerating. Maybe a side profile. 00:02:55.879 --> 00:02:57.699 Yep. Then an over -the -shoulder shot. You see 00:02:57.699 --> 00:03:00.199 the city lights reflecting in the puddles. Then 00:03:00.199 --> 00:03:02.500 a clean product shot of the shoe itself. And 00:03:02.500 --> 00:03:04.080 you end with the runner sprinting right at the 00:03:04.080 --> 00:03:05.900 camera. And that breakdown dictates everything. 00:03:06.039 --> 00:03:08.240 And this is the important part. You're not just 00:03:08.240 --> 00:03:10.479 taking the AI suggestions and running with them. 00:03:10.659 --> 00:03:13.539 No. The human element of taste is critical. You 00:03:13.539 --> 00:03:15.479 have to modify the output. If the AI suggests 00:03:15.479 --> 00:03:17.819 something that doesn't fit your brand, you push 00:03:17.819 --> 00:03:20.860 back. You have to know when to say no. The AI 00:03:20.860 --> 00:03:24.080 can generate ideas, but it can't feel the brand 00:03:24.080 --> 00:03:26.520 alignment, you know, that emotional tone. That's 00:03:26.520 --> 00:03:28.800 all human judgment. It's the difference between 00:03:28.800 --> 00:03:31.159 generating something and generating the right 00:03:31.159 --> 00:03:33.500 thing. So if the AI can do all that. You know, 00:03:33.520 --> 00:03:36.819 the scene breakdowns, the suggestions. What's 00:03:36.819 --> 00:03:38.860 the one thing that a human has to bring to the 00:03:38.860 --> 00:03:42.219 table? It's taste. Human taste guides the final 00:03:42.219 --> 00:03:45.879 concept. AI only assists idea generation. That's 00:03:45.879 --> 00:03:47.819 it. So once that storyboard is locked, we move 00:03:47.819 --> 00:03:51.800 to step two, image creation. And this is where 00:03:51.800 --> 00:03:53.620 things get tricky, but also really interesting. 00:03:53.800 --> 00:03:55.620 Because the quality of your starting frames, 00:03:55.719 --> 00:03:57.840 it determines everything. It's the anchor for 00:03:57.840 --> 00:03:59.900 the entire project. And we're using a tool like 00:03:59.900 --> 00:04:02.180 NanoBanana Pro for this, right? Yeah. Through 00:04:02.180 --> 00:04:05.080 Gemini or AI Studio. Yep. You need that level 00:04:05.080 --> 00:04:07.300 of control over the visuals, things like aspect 00:04:07.300 --> 00:04:09.919 ratio, the lighting. Those details are what give 00:04:09.919 --> 00:04:12.800 you a stable, high quality video later on. It's 00:04:12.800 --> 00:04:15.020 the difference maker. And here's a great tip 00:04:15.020 --> 00:04:17.660 from the source. Don't just generate one image 00:04:17.660 --> 00:04:20.029 per scene description. No, you need options. 00:04:20.509 --> 00:04:22.870 Generate four or five variations for each scene. 00:04:23.069 --> 00:04:26.170 Why so many? Because you need to see how the 00:04:26.170 --> 00:04:29.110 AI interprets the prompt in different ways. It 00:04:29.110 --> 00:04:31.509 lets you choose the single best frame that really 00:04:31.509 --> 00:04:33.949 nails that Blade Runner aesthetic we talked about. 00:04:34.170 --> 00:04:35.970 Okay, but here's the part where I think a lot 00:04:35.970 --> 00:04:38.050 of people fail. Yeah. The part that can ruin 00:04:38.050 --> 00:04:40.759 the whole video. Yeah. maintaining visual consistency. 00:04:41.019 --> 00:04:43.300 This is the critical step. You need all seven 00:04:43.300 --> 00:04:45.199 of those scenes to look like they were shot by 00:04:45.199 --> 00:04:47.879 the same person in the same place at the same 00:04:47.879 --> 00:04:50.120 time. Absolutely. So for your very first scene, 00:04:50.220 --> 00:04:53.680 that distant skyline, you generate images until 00:04:53.680 --> 00:04:56.480 one is perfect. Just perfect. And that single 00:04:56.480 --> 00:04:59.639 image becomes, what, the style anchor. Exactly. 00:04:59.819 --> 00:05:02.279 Your style anchor. But wait, if I use the exact 00:05:02.279 --> 00:05:05.160 same text tromped for scene two, why do I still 00:05:05.160 --> 00:05:07.439 need to reference that first image? won't the 00:05:07.439 --> 00:05:10.439 prompt keep it consistent great question and 00:05:10.439 --> 00:05:14.680 the answer is surprisingly no Generative models, 00:05:15.040 --> 00:05:16.920 even really good ones, they have what's called 00:05:16.920 --> 00:05:20.459 prompt drift. You can type the exact same prompt 00:05:20.459 --> 00:05:22.779 in seven times and you'll get seven slightly 00:05:22.779 --> 00:05:25.220 different lighting setups, different color palettes. 00:05:25.279 --> 00:05:27.519 They just, they drift away from the original 00:05:27.519 --> 00:05:29.959 intent. So the text prompt alone isn't stable 00:05:29.959 --> 00:05:32.560 enough for a whole sequence. Precisely. The image 00:05:32.560 --> 00:05:35.639 is stable. So the rule is for every single scene 00:05:35.639 --> 00:05:38.300 after the first one, scene two, three, four, 00:05:38.420 --> 00:05:40.980 all of them. You have to add that first anchor 00:05:40.980 --> 00:05:43.199 image as a reference in your prompt. And you 00:05:43.199 --> 00:05:46.399 tell the AI explicitly, use the same style, lighting, 00:05:46.500 --> 00:05:48.899 and aesthetic. from image one you're forcing 00:05:48.899 --> 00:05:50.720 it you're locking it down and this isn't just 00:05:50.720 --> 00:05:53.379 for the mood it's for objects too like the shoe 00:05:53.379 --> 00:05:56.319 exactly if you have the shoe in scene one you 00:05:56.319 --> 00:05:58.620 reference that image for every other shoe shot 00:05:58.620 --> 00:06:00.600 otherwise it might change color or even model 00:06:00.600 --> 00:06:03.079 mid -run i'll admit i still wrestle with prompt 00:06:03.079 --> 00:06:05.839 drift myself it happens so fast using that anchor 00:06:05.839 --> 00:06:08.120 image yeah that really feels like the only way 00:06:08.120 --> 00:06:10.360 to get a professional cohesive look it's the 00:06:10.360 --> 00:06:13.600 secret So if we skip referencing that first scene 00:06:13.600 --> 00:06:16.300 image for every shot, what's the guaranteed result? 00:06:16.600 --> 00:06:18.800 The scenes become visually disjointed, ruining 00:06:18.800 --> 00:06:21.720 the professional look of the final video. Guaranteed. 00:06:21.779 --> 00:06:24.740 Okay. So now we have our beautiful, consistent, 00:06:24.819 --> 00:06:29.220 static images. Seven approved frames. We move 00:06:29.220 --> 00:06:34.860 to step three. Video creation. Time to add some 00:06:34.860 --> 00:06:37.079 motion. And we actually go back to the AI assistant 00:06:37.079 --> 00:06:40.250 first. You take your seven final images and you 00:06:40.250 --> 00:06:42.649 ask it to write motion -specific prompts for 00:06:42.649 --> 00:06:45.009 each one. So the AI looks at the static picture 00:06:45.009 --> 00:06:46.930 and suggests the camera movement. Yeah, like 00:06:46.930 --> 00:06:49.850 a quick dolly in or a slow pan left. It gives 00:06:49.850 --> 00:06:52.569 you ideas for the motion. But here, unlike with 00:06:52.569 --> 00:06:55.490 the images, the source says to only ask for one 00:06:55.490 --> 00:06:58.050 video prompt per scene. Right. It's all about 00:06:58.050 --> 00:07:01.350 cost and time. Generating video is just way more 00:07:01.350 --> 00:07:03.829 resource intensive than images. So it would be 00:07:03.829 --> 00:07:06.470 too expensive and slow to generate 50 video clips 00:07:06.470 --> 00:07:09.470 just to find seven good ones. Prohibitively so. 00:07:09.959 --> 00:07:11.620 You have to be much more targeted with your video 00:07:11.620 --> 00:07:14.259 prompts. And for the tool, we're using VO 3 .1 00:07:14.259 --> 00:07:16.759 Fast. Why the fast version? It's just so much 00:07:16.759 --> 00:07:19.040 cheaper and quicker. And honestly, for this kind 00:07:19.040 --> 00:07:20.879 of work, the quality difference isn't big enough 00:07:20.879 --> 00:07:23.139 to justify the extra cost and wait time of the 00:07:23.139 --> 00:07:25.939 slower model. You need to iterate. And the most 00:07:25.939 --> 00:07:28.019 important technical feature of whatever tool 00:07:28.019 --> 00:07:30.759 you use, it has to let you use your own image 00:07:30.759 --> 00:07:33.560 as the starting frame. It has to. Otherwise, 00:07:33.699 --> 00:07:35.860 all that consistency work you did in step two 00:07:35.860 --> 00:07:37.939 is just gone. It's the link that holds the whole 00:07:37.939 --> 00:07:40.319 pipeline together. Now, the reality of the output 00:07:40.319 --> 00:07:43.959 is still, well, it's not perfect. You're still 00:07:43.959 --> 00:07:45.800 going to have to regenerate some scenes. Oh, 00:07:45.819 --> 00:07:48.319 for sure. If you have 10 scenes, you're absolutely 00:07:48.319 --> 00:07:50.259 going to redo at least three or four of them, 00:07:50.300 --> 00:07:52.480 maybe two, three, four times each. That's just 00:07:52.480 --> 00:07:55.720 where the tech is right now. Whoa. But just imagine 00:07:55.720 --> 00:07:59.240 scaling this, scaling this structured system 00:07:59.240 --> 00:08:02.000 across an entire campaign. You could produce 00:08:02.000 --> 00:08:05.259 hundreds of high -end localized ads in just a 00:08:05.259 --> 00:08:08.129 few days. It's a profound shift in production 00:08:08.129 --> 00:08:10.870 capacity. It really is. But there's a key piece 00:08:10.870 --> 00:08:13.430 of expectation management here. The source notes 00:08:13.430 --> 00:08:15.370 that if the AI gives you an eight second video 00:08:15.370 --> 00:08:17.269 clip. You're not going to use all eight seconds. 00:08:17.410 --> 00:08:19.209 Right. You might only get one to three seconds 00:08:19.209 --> 00:08:23.029 of truly usable footage. It's because of model 00:08:23.029 --> 00:08:25.589 coherence over time. The longer the AI has to 00:08:25.589 --> 00:08:28.589 generate, the more likely it is to, well, to 00:08:28.589 --> 00:08:30.939 get. Things start melting. People grow extra 00:08:30.939 --> 00:08:34.659 fingers. Yeah, exactly. The first few seconds 00:08:34.659 --> 00:08:36.580 are stable because they're close to that perfect 00:08:36.580 --> 00:08:38.740 starting image. But as it gets further away, 00:08:38.919 --> 00:08:41.139 it kind of loses its grip on reality. So you 00:08:41.139 --> 00:08:43.259 have to treat it like raw footage. You know you're 00:08:43.259 --> 00:08:44.940 going to have to trim the beginning and the end. 00:08:45.039 --> 00:08:47.519 You have to. You judge the quality by asking, 00:08:47.600 --> 00:08:51.399 can I get one? to three perfect seconds out of 00:08:51.399 --> 00:08:54.519 this, not is the whole eight -second clip a masterpiece. 00:08:54.940 --> 00:08:57.519 That one mindset shift probably saves a lot of 00:08:57.519 --> 00:09:00.360 frustration. A ton. So the source says an eight 00:09:00.360 --> 00:09:02.379 -second generation usually gives you only one 00:09:02.379 --> 00:09:05.820 to three seconds of good footage. What does that 00:09:05.820 --> 00:09:08.620 tell us about judging the quality? Quality comes 00:09:08.620 --> 00:09:11.799 from aggressive trimming, not expecting eight 00:09:11.799 --> 00:09:14.360 perfect seconds of output. Okay, now for the 00:09:14.360 --> 00:09:17.690 home stretch. Step four. Production. We've got 00:09:17.690 --> 00:09:20.409 our short, beautiful, consistent clips. It's 00:09:20.409 --> 00:09:22.610 time to stitch them all together. Yep. Using 00:09:22.610 --> 00:09:25.629 an editor like CapCut, Premiere, or DaVinci Resolve. 00:09:25.769 --> 00:09:27.610 And the key concept here, the thing that really 00:09:27.610 --> 00:09:30.230 elevates the final product, is audio -driven 00:09:30.230 --> 00:09:32.570 editing. The music is not an afterthought. It 00:09:32.570 --> 00:09:34.470 dictates everything. The whole rhythm of the 00:09:34.470 --> 00:09:36.649 piece. So you pick your music first. Something 00:09:36.649 --> 00:09:39.679 with a clear beat. A clear beat, yeah. You import 00:09:39.679 --> 00:09:42.080 your clips, you import the audio, and then you 00:09:42.080 --> 00:09:44.100 trim your little one to three second clips so 00:09:44.100 --> 00:09:46.539 that the cut, the end of the clip, lands exactly 00:09:46.539 --> 00:09:49.240 on a beat. That's what gives it that quick, rhythmic, 00:09:49.240 --> 00:09:51.960 professional feel. The visual cuts are synced 00:09:51.960 --> 00:09:54.779 to the music. It's all about timing, not fancy 00:09:54.779 --> 00:09:57.659 effects, just timing. Once that's locked, you 00:09:57.659 --> 00:10:00.460 do some simple enhancements. Like text overlays, 00:10:00.460 --> 00:10:02.980 maybe some simple fades. Exactly. Keep it simple. 00:10:03.080 --> 00:10:05.559 But the last critical step is color grading. 00:10:06.110 --> 00:10:08.929 Why is that so important if we already use the 00:10:08.929 --> 00:10:12.169 consistency anchor? The anchor handles the style, 00:10:12.269 --> 00:10:15.110 but not the final calibration. You might still 00:10:15.110 --> 00:10:17.590 have tiny differences in exposure between clips. 00:10:17.789 --> 00:10:20.830 So you apply one final unified color correction 00:10:20.830 --> 00:10:23.230 across everything. It's the final glue that makes 00:10:23.230 --> 00:10:25.009 it feel like it was all shot on the same day 00:10:25.009 --> 00:10:27.850 with the same camera. Then you export in 4K. 00:10:27.990 --> 00:10:30.110 This whole workflow, it really brings up the 00:10:30.110 --> 00:10:32.870 question of automation. We're using AI for so 00:10:32.870 --> 00:10:35.309 much of it already. What parts can we automate 00:10:35.309 --> 00:10:37.950 and what parts must stay human? You can automate 00:10:37.950 --> 00:10:40.830 the repetitive stuff, you know, generating prompt 00:10:40.830 --> 00:10:43.110 variations, the pipeline that takes a description 00:10:43.110 --> 00:10:46.029 and spits out five images, file naming. But the 00:10:46.029 --> 00:10:48.169 source is really firm on what you should not 00:10:48.169 --> 00:10:50.090 automate. And that is the creative selection 00:10:50.090 --> 00:10:52.509 process. This is the golden rule. You cannot 00:10:52.509 --> 00:10:54.789 have an AI pick the best starting frame or the 00:10:54.789 --> 00:10:57.440 best video clip. You can't. Automation should 00:10:57.440 --> 00:11:00.860 handle repetition, not taste. And AI will miss 00:11:00.860 --> 00:11:03.620 the nuance. It will miss the feeling. It will 00:11:03.620 --> 00:11:06.120 pick something that's technically correct, but 00:11:06.120 --> 00:11:08.639 creatively dead. Because you have to visually 00:11:08.639 --> 00:11:11.759 inspect those five options to see which one really 00:11:11.759 --> 00:11:14.360 captures the vision. That curation, looking at 00:11:14.360 --> 00:11:16.419 five images and saying, this one feels right. 00:11:16.519 --> 00:11:19.200 That is the highest value human skill in this 00:11:19.200 --> 00:11:21.919 whole process now. So if we can technically automate 00:11:21.919 --> 00:11:25.139 the entire process end to end. Why should we 00:11:25.139 --> 00:11:28.139 intentionally leave that creative selection part 00:11:28.139 --> 00:11:31.220 to a human? Over -automation sacrifices taste. 00:11:31.460 --> 00:11:34.039 The final output needs human creative selection 00:11:34.039 --> 00:11:37.019 to be engaging. It all comes back to taste. Always. 00:11:37.179 --> 00:11:40.759 So to recap the big idea here for you, the listener, 00:11:41.000 --> 00:11:44.360 the era of AI marketing is here. It's happening. 00:11:44.480 --> 00:11:46.879 We're collapsing weeks into hours with tools 00:11:46.879 --> 00:11:50.379 like Nano Banana Pro and VO 3 .1. But the real 00:11:50.379 --> 00:11:53.220 power isn't just in the tools. It's in the structured 00:11:53.220 --> 00:11:56.090 workflow. using images to lock in consistency, 00:11:56.350 --> 00:11:59.389 and using audio to dictate the pace. And the 00:11:59.389 --> 00:12:01.250 critical mistake to avoid, as we've said, is 00:12:01.250 --> 00:12:03.230 trying to automate your own creative judgment. 00:12:03.289 --> 00:12:05.909 The AI can't do that. That's your job. Which 00:12:05.909 --> 00:12:08.429 really brings us to a final provocative thought. 00:12:08.690 --> 00:12:11.970 The winners in this new world. They're not going 00:12:11.970 --> 00:12:13.649 to be the best coders or even the people who 00:12:13.649 --> 00:12:16.029 write the longest, most technical prompts. They're 00:12:16.029 --> 00:12:17.590 going to be the ones with a clear creative vision 00:12:17.590 --> 00:12:21.169 and impeccable taste who use AI simply as a tool 00:12:21.169 --> 00:12:23.730 for execution. The tools are ready. The workflow 00:12:23.730 --> 00:12:26.090 is proven. The only question left for you is 00:12:26.090 --> 00:12:28.470 what original creative vision are you bringing 00:12:28.470 --> 00:12:29.009 to the table?