WEBVTT 00:00:00.000 --> 00:00:02.020 Okay, think about the usual hassle of hiring 00:00:02.020 --> 00:00:04.059 creators. Yeah. You know, all the contracts, 00:00:04.280 --> 00:00:06.540 the back and forth, weeks of waiting. Right, 00:00:06.639 --> 00:00:09.919 revisions. And you're spending maybe hundreds, 00:00:10.140 --> 00:00:14.560 even thousands, for just one 30 -second ad. An 00:00:14.560 --> 00:00:16.760 ad that might totally bomb, by the way. Exactly, 00:00:16.920 --> 00:00:19.620 a huge upfront gamble before you even see if 00:00:19.620 --> 00:00:22.339 the idea connects. Now, compare that whole headache 00:00:22.339 --> 00:00:27.149 to... Getting an almost unlimited supply of professional 00:00:27.149 --> 00:00:31.030 looking kind of viral style UGC ads. Generated 00:00:31.030 --> 00:00:33.770 automatically by AI. Yeah. Ready to go for maybe 00:00:33.770 --> 00:00:36.450 15 cents each. That difference is just, it's 00:00:36.450 --> 00:00:38.630 huge. It really flits the script on how marketing 00:00:38.630 --> 00:00:40.950 creative gets made. So today we're going to dig 00:00:40.950 --> 00:00:43.030 into the blueprint for building exactly that 00:00:43.030 --> 00:00:45.070 system. Right. We're unpacking this automated 00:00:45.070 --> 00:00:49.170 AI thing. Let's call it a UGC ad generator. And 00:00:49.170 --> 00:00:52.140 it uses no code tools like NAN. Google Sheets. 00:00:52.200 --> 00:00:54.899 All hooked up to some pretty advanced AI models. 00:00:55.000 --> 00:00:57.060 So the plan is, first we'll look at what you 00:00:57.060 --> 00:00:59.539 put in and what you get out. Super simple inputs, 00:00:59.740 --> 00:01:01.899 surprisingly powerful outputs. Then we'll get 00:01:01.899 --> 00:01:03.659 into the interesting part, comparing the three 00:01:03.659 --> 00:01:06.640 different AI workflows, like three teams competing 00:01:06.640 --> 00:01:09.079 to be the best engine for this. And finally, 00:01:09.079 --> 00:01:10.700 we'll talk about how you go from just building 00:01:10.700 --> 00:01:13.620 one ad to scaling this whole thing up into like 00:01:13.620 --> 00:01:17.379 a 247 content factory. Sounds good. Let's jump 00:01:17.379 --> 00:01:22.680 in. So the old way, getting good UGC. It's tough. 00:01:22.859 --> 00:01:25.900 Right. And slow. Oh, yeah. Costs can be anywhere 00:01:25.900 --> 00:01:30.620 from like 50 bucks to maybe $500 for single video. 00:01:30.859 --> 00:01:34.140 And just coordinating everything. Emails, shipping 00:01:34.140 --> 00:01:36.379 products, waiting for approvals that can easily 00:01:36.379 --> 00:01:38.920 eat up days, sometimes weeks. Plus, there's always 00:01:38.920 --> 00:01:41.180 that risk hanging over you. You pour in the time, 00:01:41.219 --> 00:01:43.939 the money, and crickets, the ad just doesn't 00:01:43.939 --> 00:01:45.840 perform. It's a real logistical nightmare for 00:01:45.840 --> 00:01:47.780 something that often has a pretty short shelf 00:01:47.780 --> 00:01:50.810 life anyway. Okay, but the beauty of this AI 00:01:50.810 --> 00:01:54.209 approach is how little input it actually needs. 00:01:54.450 --> 00:01:56.569 It's kind of amazing. Yeah, you basically kick 00:01:56.569 --> 00:01:58.390 off the whole process just by filling out one 00:01:58.390 --> 00:02:00.629 single row in a spreadsheet, like Google Sheets. 00:02:00.650 --> 00:02:03.310 Symbol data in, finish content out. We found 00:02:03.310 --> 00:02:05.849 there are basically five key pieces of info you 00:02:05.849 --> 00:02:09.650 need. Right. Number one, the URL for the product 00:02:09.650 --> 00:02:11.650 photo. Then who's your target audience? Yeah. 00:02:11.770 --> 00:02:14.550 Your ICP. Yep. What product features do you want 00:02:14.550 --> 00:02:16.740 to talk about? Uh -huh. And where should the 00:02:16.740 --> 00:02:19.159 video look like it's taking place? The setting, 00:02:19.199 --> 00:02:21.800 a kitchen, a car. And the last one, which is 00:02:21.800 --> 00:02:24.120 important for testing, is choosing which AI model 00:02:24.120 --> 00:02:27.020 setup you want to use for that specific ad. And 00:02:27.020 --> 00:02:28.979 what comes out the other end? It's pretty slick. 00:02:29.159 --> 00:02:32.860 You get a fully automated, ready to use, maybe... 00:02:33.400 --> 00:02:36.620 eight to ten second ugc style video ad what's 00:02:36.620 --> 00:02:39.539 professional yeah surprisingly so it includes 00:02:39.539 --> 00:02:42.300 a realistic looking human presenter ai generated 00:02:42.300 --> 00:02:45.639 dialogue that sounds pretty natural and critically 00:02:45.639 --> 00:02:48.800 it's ready to post immediately already formatted 00:02:48.800 --> 00:02:52.000 for tick tock reels you know vertical video you 00:02:52.000 --> 00:02:53.860 just get a video link drop it straight into your 00:02:53.860 --> 00:02:56.289 ad campaign The real power here, though, it seems, 00:02:56.449 --> 00:02:58.569 is the scale and the speed of testing. Exactly. 00:02:58.629 --> 00:03:01.129 That's the game changer. You can test like 50 00:03:01.129 --> 00:03:03.090 different creative angles at the same time, focus 00:03:03.090 --> 00:03:05.150 on different features, different settings. And 00:03:05.150 --> 00:03:07.449 see what works almost instantly instead of waiting 00:03:07.449 --> 00:03:10.110 weeks per test. Right. It slashes that time to 00:03:10.110 --> 00:03:12.430 content. That's the edge. So boiling it down, 00:03:12.550 --> 00:03:14.830 what's the main advantage of this scale compared 00:03:14.830 --> 00:03:17.409 to just hiring one person? It's automatically 00:03:17.409 --> 00:03:19.990 scale testing tons of different creative ideas 00:03:19.990 --> 00:03:22.409 all at once. Okay, got it. Let's get into those 00:03:22.409 --> 00:03:24.599 three workflows then. This is where you see the 00:03:24.599 --> 00:03:26.780 different ways to build this engine. Yeah, and 00:03:26.780 --> 00:03:28.580 we should probably define a couple of terms first. 00:03:28.620 --> 00:03:32.020 We keep saying NN. Right, so NNN, think of it 00:03:32.020 --> 00:03:35.099 like the supervisor on a factory floor. It's 00:03:35.099 --> 00:03:37.120 a no -code tool that connects all the different 00:03:37.120 --> 00:03:39.780 steps and APIs together in the right sequence. 00:03:40.020 --> 00:03:42.319 It tells everything what to do and when. And 00:03:42.319 --> 00:03:46.060 the other piece is FAL AI. Ah, yes, FAL AI. That's 00:03:46.060 --> 00:03:48.180 basically a service that bundles up a bunch of 00:03:48.180 --> 00:03:50.530 different cutting -edge AI models. So instead 00:03:50.530 --> 00:03:53.090 of juggling multiple accounts and APIs, Fal gives 00:03:53.090 --> 00:03:56.789 us one place to access models like Nanobanana 00:03:56.789 --> 00:03:59.789 and Vio, which we used here, makes things simpler. 00:03:59.990 --> 00:04:02.210 Okay, so Workflow 1, this is the one you recommend, 00:04:02.330 --> 00:04:04.229 the pro version. Yeah, this is the one we landed 00:04:04.229 --> 00:04:06.509 on as the most reliable. It's a two -step process, 00:04:06.830 --> 00:04:10.810 Nanobanana plus Vio 3 .1. Step 1 uses Nanobanana. 00:04:10.870 --> 00:04:13.729 What's that? It's an AI image model. Its job 00:04:13.729 --> 00:04:16.069 is to take your product photo and your prompt 00:04:16.069 --> 00:04:20.029 and create a new really realistic image of a 00:04:20.029 --> 00:04:22.290 person actually holding or using your product 00:04:22.290 --> 00:04:24.569 correctly. Okay, so it generates the person with 00:04:24.569 --> 00:04:28.029 the product. Then step two. Step two uses VO 00:04:28.029 --> 00:04:31.810 3 .1, which is an AI video model. It takes that 00:04:31.810 --> 00:04:34.350 image Nano Banana just made and animates it, 00:04:34.410 --> 00:04:37.410 adds the talking, the subtle movements. And the 00:04:37.410 --> 00:04:40.769 big advantage here is? Accuracy, mainly. The 00:04:40.769 --> 00:04:43.350 product usually looks right, held correctly. 00:04:43.750 --> 00:04:46.470 And really importantly, this two -step thing 00:04:46.470 --> 00:04:49.290 avoids that horrible static thumbnail problem. 00:04:49.610 --> 00:04:51.490 Because the first frame isn't just the product 00:04:51.490 --> 00:04:53.810 photo. It's the AI -generated person already 00:04:53.810 --> 00:04:56.019 moving. Exactly. The video starts with action, 00:04:56.199 --> 00:04:58.399 which is way better for grabbing attention on 00:04:58.399 --> 00:05:00.860 social feeds. What's the downside? Well, it's 00:05:00.860 --> 00:05:03.259 two steps, so it takes a little longer. And it 00:05:03.259 --> 00:05:05.740 costs a bit more, came out to around 32 cents 00:05:05.740 --> 00:05:08.420 per video in our test. Okay, workflow two then, 00:05:08.519 --> 00:05:10.720 the speed demon. Yeah, the one everyone wants 00:05:10.720 --> 00:05:13.779 to work, Sora 2 only, just straight. Image to 00:05:13.779 --> 00:05:16.560 video. And it's fast and cheap. Super fast and 00:05:16.560 --> 00:05:18.759 the cheapest. We clocked it at about 15 cents 00:05:18.759 --> 00:05:20.860 for a 10 -second video. But there's always a 00:05:20.860 --> 00:05:23.519 catch. Big catch here. Sora 2, at least right 00:05:23.519 --> 00:05:26.279 now, has pretty tight content rules. It often 00:05:26.279 --> 00:05:29.839 flags and blocks realistic AI -generated human 00:05:29.839 --> 00:05:33.160 faces. Oof. That's a non -starter for believable 00:05:33.160 --> 00:05:35.560 UGC ads. Pretty much. Plus, you still get that 00:05:35.560 --> 00:05:37.959 static product photo as the first frame, that 00:05:37.959 --> 00:05:40.399 bad thumbnail issue again. All right. And workflow 00:05:40.399 --> 00:05:44.319 three, the middle ground. Kinda. This one uses 00:05:44.319 --> 00:05:48.399 VO 3 .1 only, so direct image to video like Sora 00:05:48.399 --> 00:05:51.980 2. It's reasonably fast, about 30 cents for an 00:05:51.980 --> 00:05:54.120 8 second clip. And does it have the face restriction 00:05:54.120 --> 00:05:56.560 problem? Nope, no face restrictions, which is 00:05:56.560 --> 00:05:58.980 good. So what's wrong with this one? Oh, this 00:05:58.980 --> 00:06:01.699 one had a major flaw. A deal breaker, honestly. 00:06:02.120 --> 00:06:05.550 Product alteration. Meaning? It kept changing 00:06:05.550 --> 00:06:07.730 the product. We were using this example of a 00:06:07.730 --> 00:06:12.009 glass jar of gummy supplements. VO 3 .1 kept 00:06:12.009 --> 00:06:15.089 turning the jar into a flexible bag in the video. 00:06:15.310 --> 00:06:17.709 Seriously, it just swapped the packaging. Yep. 00:06:18.040 --> 00:06:19.779 Consistently. We wasted a bunch of runs trying 00:06:19.779 --> 00:06:21.860 to fix it. We even nicknamed it the gummy thief 00:06:21.860 --> 00:06:24.040 internally because it kept stealing the jar. 00:06:24.220 --> 00:06:26.819 Wow. Why would it do that? Just misinterpret 00:06:26.819 --> 00:06:29.480 the image. Our best guess is it over indexes 00:06:29.480 --> 00:06:32.839 on context. Like if the prompt talks about grabbing 00:06:32.839 --> 00:06:34.759 something quickly on the way out, the AI thinks 00:06:34.759 --> 00:06:37.639 quick grab must be a flexible bag and ignores 00:06:37.639 --> 00:06:40.100 the fact that the input image was clearly a rigid 00:06:40.100 --> 00:06:43.240 jar. That's. Not good for brand consistency. 00:06:43.519 --> 00:06:45.139 Not good at all. It completely undermines the 00:06:45.139 --> 00:06:46.860 point if the product isn't shown accurately. 00:06:47.220 --> 00:06:50.459 Okay, so given that risk with VO 3 .1 alone, 00:06:50.699 --> 00:06:53.899 why is that more complex two -step process and 00:06:53.899 --> 00:06:56.279 workflow one necessary? To make sure the product 00:06:56.279 --> 00:06:59.500 looks right. And crucially, to get that dynamic 00:06:59.500 --> 00:07:02.079 first frame with action. Right. Reliability wins 00:07:02.079 --> 00:07:04.699 out. Okay. Okay, so workflow one it is. Let's 00:07:04.699 --> 00:07:06.560 peek behind the curtain now at how this actually 00:07:06.560 --> 00:07:09.339 works in ANN. Sure. So prerequisites, you need 00:07:09.339 --> 00:07:11.800 an NA done setup, a Google Sheet ready, your 00:07:11.800 --> 00:07:15.139 file AI account, and an OpenAI API key for the 00:07:15.139 --> 00:07:17.740 brains. And the workflow starts how? It kicks 00:07:17.740 --> 00:07:19.540 off with an ANN trigger. It's basically just 00:07:19.540 --> 00:07:21.500 watching that Google Sheet, looking for any new 00:07:21.500 --> 00:07:24.220 row you mark as ready. Finds a ready row, then 00:07:24.220 --> 00:07:26.850 what? hits a switch node that's just like a traffic 00:07:26.850 --> 00:07:30.089 controller it looks at which ai model you chose 00:07:30.089 --> 00:07:32.790 in the spreadsheet for that row and sends the 00:07:32.790 --> 00:07:35.709 job down the right path for our winning workflow 00:07:35.709 --> 00:07:38.470 it sends it to the nano banana path first okay 00:07:38.470 --> 00:07:41.329 so node one in that path is the image prompt 00:07:41.329 --> 00:07:44.899 agent what's that doing This uses an AI model 00:07:44.899 --> 00:07:47.939 like GPT -4 .0 as an agent. We give it a really 00:07:47.939 --> 00:07:50.560 detailed system prompt. Think of the system prompts 00:07:50.560 --> 00:07:53.240 like the AI's job description and rulebook. You're 00:07:53.240 --> 00:07:55.420 telling it exactly how to behave. Exactly. We 00:07:55.420 --> 00:07:57.579 tell it, your job is to write a prompt for an 00:07:57.579 --> 00:08:00.920 image generation AI. Make the image hyper -realistic. 00:08:01.019 --> 00:08:04.000 Think lifelike skin, tiny imperfections, maybe 00:08:04.000 --> 00:08:06.920 a selfie angle. And critically, make sure the 00:08:06.920 --> 00:08:09.300 product in the image looks exactly like the one 00:08:09.300 --> 00:08:11.699 in the photo URL we gave you. So it crafts the 00:08:11.699 --> 00:08:14.050 instructions for now. nanobanana, then nanobanana 00:08:14.050 --> 00:08:17.290 starts making the image. But that takes time. 00:08:17.449 --> 00:08:19.589 Right. AI generation is an instance. So that 00:08:19.589 --> 00:08:21.449 brings us to the polling loop. This is super 00:08:21.449 --> 00:08:23.529 important. Because you can't just wait indefinitely. 00:08:23.750 --> 00:08:26.829 Nope. The workflow uses a wait node to pause 00:08:26.829 --> 00:08:29.629 for a bit. Then an alpha node to check the status 00:08:29.629 --> 00:08:32.610 from fal .ai. Is the image done yet? If not, 00:08:32.730 --> 00:08:35.429 it loops back, waits again, checks again. Keeps 00:08:35.429 --> 00:08:37.850 knocking on the door until fal .ai says completed. 00:08:38.620 --> 00:08:41.759 Precisely. That loop stops the whole system from 00:08:41.759 --> 00:08:44.080 timing out or breaking while it waits. Okay, 00:08:44.159 --> 00:08:45.960 image is done. Now, this next part is really 00:08:45.960 --> 00:08:49.080 interesting. Node 7, analyze generated image. 00:08:49.580 --> 00:08:52.480 You use OpenAI Vision here. Yeah, this is maybe 00:08:52.480 --> 00:08:55.179 the cleverest bit. You take the image that NanoBanana 00:08:55.179 --> 00:08:58.000 just created, and you feed it back into another 00:08:58.000 --> 00:09:01.519 AI, GPT -4O, with vision capabilities. Hold on, 00:09:01.559 --> 00:09:04.000 you use AI number two to look at what AI number 00:09:04.000 --> 00:09:07.399 one just made? Why? Seems redundant. It's like 00:09:07.399 --> 00:09:10.019 quality control. It solves the problem of AI 00:09:10.019 --> 00:09:13.259 hallucination. Sometimes the first AI might slightly 00:09:13.259 --> 00:09:15.519 mess up or maybe the image isn't quite what you 00:09:15.519 --> 00:09:18.220 prompted. Ah, so the vision AI describes what's 00:09:18.220 --> 00:09:20.379 actually in the image. Exactly. It looks at the 00:09:20.379 --> 00:09:22.620 picture and says, OK, I see a woman with brown 00:09:22.620 --> 00:09:25.279 hair sitting in a blue car holding a white jar. 00:09:25.500 --> 00:09:28.320 It confirms the visual reality. And that description 00:09:28.320 --> 00:09:31.620 is then used for the next step. Yes. That description 00:09:31.620 --> 00:09:35.500 becomes a key input for Node -8, the video prompt 00:09:35.500 --> 00:09:39.080 agent. Now, the AI writing the video script knows 00:09:39.080 --> 00:09:41.659 for sure it needs to write dialogue for a woman 00:09:41.659 --> 00:09:45.340 in a blue car holding a white jar, not a red 00:09:45.340 --> 00:09:48.840 truck or a green bag. That's smart. It anchors 00:09:48.840 --> 00:09:50.740 the video script to the actual image that was 00:09:50.740 --> 00:09:53.820 generated, ensuring consistency. Totally. Prevents 00:09:53.820 --> 00:09:55.480 weird disconnects between the visuals and the 00:09:55.480 --> 00:09:57.990 dialogue. I can imagine getting these prompts 00:09:57.990 --> 00:10:00.110 right, especially chained together like this, 00:10:00.169 --> 00:10:02.950 must be tricky. You mentioned prompt drift. I 00:10:02.950 --> 00:10:05.450 still wrestle with prompt drift myself when managing 00:10:05.450 --> 00:10:08.490 API calls, getting the JSON clean and consistent. 00:10:08.690 --> 00:10:10.450 Oh yeah, it's a constant thing. Prompt drift. 00:10:11.039 --> 00:10:13.139 It's like playing telephone with the AI. You 00:10:13.139 --> 00:10:15.379 give it instructions, but by the third or fourth 00:10:15.379 --> 00:10:18.299 step in a chain, the AI might kind of start interpreting 00:10:18.299 --> 00:10:20.559 things a bit loosely. Forgets the original strict 00:10:20.559 --> 00:10:22.759 rules. Yeah, you asked for ultra -realistic, 00:10:22.779 --> 00:10:25.159 but maybe it starts leaning a bit more stylized 00:10:25.159 --> 00:10:27.159 down the line if you're not careful with how 00:10:27.159 --> 00:10:29.820 you pass context. It requires careful prompt 00:10:29.820 --> 00:10:32.299 engineering and sometimes explicit reminders 00:10:32.299 --> 00:10:35.080 in later prompts. Makes sense. Okay, so Node 00:10:35.080 --> 00:10:37.659 8, the video prompt agent, uses the audience 00:10:37.659 --> 00:10:40.100 info, product features, and that verified image 00:10:40.100 --> 00:10:42.960 description. To generate the final video prompt 00:10:42.960 --> 00:10:46.139 for VO 3 .1. This includes writing the eight 00:10:46.139 --> 00:10:48.200 seconds of dialogue the person should say, making 00:10:48.200 --> 00:10:50.700 it sound spontaneous and natural, matching the 00:10:50.700 --> 00:10:53.299 scene. Got it. And the last few steps. Nodes 00:10:53.299 --> 00:10:56.039 9 through 12 are basically send the final prompt 00:10:56.039 --> 00:10:59.100 to VO 3 .1 to generate the video, run another 00:10:59.100 --> 00:11:00.960 polling loop to wait for that to finish. The 00:11:00.960 --> 00:11:03.059 waiting. Yep, more waiting. And then the final 00:11:03.059 --> 00:11:05.960 step, update the Google Sheet, mark the status 00:11:05.960 --> 00:11:08.460 as finished, and paste in the URL of the final 00:11:08.460 --> 00:11:11.960 video. Boom, ad generated. So to recap that complex 00:11:11.960 --> 00:11:14.460 part. What's the absolutely essential function 00:11:14.460 --> 00:11:17.000 of analyzing the generated image mid -workflow? 00:11:17.139 --> 00:11:19.340 It guarantees the video script matches the visual 00:11:19.340 --> 00:11:21.799 reality of the generated image. Consistency. 00:11:21.799 --> 00:11:24.379 Mid -role sponsor, read placeholder. Okay, let's 00:11:24.379 --> 00:11:26.919 talk results. The brass tacks. Cost. You said 00:11:26.919 --> 00:11:29.500 the winning workflow, Nano Banana plus VO3 .1, 00:11:29.620 --> 00:11:32.620 landed around $0 .18 an ad. Yeah, about $1 .18. 00:11:32.620 --> 00:11:35.480 And remember, Sort 2 was cheaper at $0 .10. VO3 00:11:35.480 --> 00:11:38.659 .1 only was $0 .15. But the comparison isn't 00:11:38.659 --> 00:11:40.980 really between $0 .10, $0 .15, and $0 .18, is 00:11:40.980 --> 00:11:44.649 it? It's between 18 cents and, what was it, $50 00:11:44.649 --> 00:11:47.490 to $500. Exactly. That's the money ball moment, 00:11:47.590 --> 00:11:49.629 right? We're talking orders of magnitude cheaper 00:11:49.629 --> 00:11:53.070 than traditional methods. Whoa. Okay, just thinking 00:11:53.070 --> 00:11:56.509 about that, testing 50 different creative ideas 00:11:56.509 --> 00:11:59.950 for less than $10. compared to maybe thousands 00:11:59.950 --> 00:12:03.149 for just one human creator test. That's the democratization 00:12:03.149 --> 00:12:06.549 aspect. Small teams, even solo founders, can 00:12:06.549 --> 00:12:09.169 suddenly test creative at a scale that was previously 00:12:09.169 --> 00:12:12.629 only possible for huge agencies. That's a massive 00:12:12.629 --> 00:12:14.350 advantage for anyone who jumps on this early. 00:12:14.529 --> 00:12:18.009 But cost isn't everything. What about the quality? 00:12:18.230 --> 00:12:20.950 Do the 18 -cent ads actually look good? That's 00:12:20.950 --> 00:12:22.669 the crucial question. Because, you know, saving 00:12:22.669 --> 00:12:24.889 8 cents per ad sounds great. But if the cheaper 00:12:24.889 --> 00:12:27.149 ads don't convert because they look bad or have 00:12:27.149 --> 00:12:29.200 issues... Then it's false economy. Especially 00:12:29.200 --> 00:12:30.559 if you're running thousands of these. Right. 00:12:30.679 --> 00:12:33.220 And this is where reliability becomes the deciding 00:12:33.220 --> 00:12:36.419 factor. Workflow One, the Nano Banana Plus VO 00:12:36.419 --> 00:12:40.039 3 .1 combo, was the clear winner on quality and 00:12:40.039 --> 00:12:43.139 reliability. Why specifically? Best natural look, 00:12:43.279 --> 00:12:46.539 consistent product accuracy, no gummy thief incidents, 00:12:46.860 --> 00:12:49.820 and that vital action -first frame. You pay a 00:12:49.820 --> 00:12:52.480 few cents more, but you get an ad that's much 00:12:52.480 --> 00:12:54.559 more likely to actually work on social platforms. 00:12:54.879 --> 00:12:58.029 So even though Sora 2... Workflow 2 is cheapest. 00:12:58.289 --> 00:13:00.809 Yeah, the face blocking and the static first 00:13:00.809 --> 00:13:03.549 frame really hurt its potential for genuine -looking 00:13:03.549 --> 00:13:06.870 UGC. It's maybe useful for some things, but not 00:13:06.870 --> 00:13:10.970 ideal. And VO 3 .1 only, Workflow 3. Dead on 00:13:10.970 --> 00:13:13.090 arrival because of the product alteration risk. 00:13:13.549 --> 00:13:15.990 Turning a jar into a bag? You just can't have 00:13:15.990 --> 00:13:18.049 that. It makes the cost savings totally irrelevant. 00:13:18.470 --> 00:13:20.549 So the real benefit of spending that extra, what, 00:13:20.710 --> 00:13:23.070 3 to 8 cents on the winning workflow boils down 00:13:23.070 --> 00:13:25.779 to? Quality and reliability simply outweigh tiny 00:13:25.779 --> 00:13:28.559 cost savings, especially avoiding critical errors 00:13:28.559 --> 00:13:31.000 like product changes. Okay, so you've built your 00:13:31.000 --> 00:13:33.720 generator. It's making great ads one by one using 00:13:33.720 --> 00:13:36.340 Workflow One. How do you scale this up? Go from 00:13:36.340 --> 00:13:38.720 a little workshop to a full -blown factory. It 00:13:38.720 --> 00:13:40.740 actually starts pretty simply, right? With batch 00:13:40.740 --> 00:13:42.500 processing. Yeah, you just tweak that initial 00:13:42.500 --> 00:13:45.259 Google Sheet trigger node in NEN. By default, 00:13:45.320 --> 00:13:47.480 it's set to only grab the first row it finds 00:13:47.480 --> 00:13:50.940 marked ready. You just untick that box. Basically, 00:13:51.000 --> 00:13:53.879 yeah. Remove that limit. Now, NEN will grab all 00:13:53.879 --> 00:13:56.720 the rows marked ready. So you could line up, 00:13:56.779 --> 00:13:59.440 say, 20 different ad ideas in your sheet, different 00:13:59.440 --> 00:14:02.139 angles, features, audiences. Hit ready on all 00:14:02.139 --> 00:14:04.139 of them, and the workflow will just chew through 00:14:04.139 --> 00:14:05.919 them one after another, maybe overnight while 00:14:05.919 --> 00:14:07.620 you sleep. That's the factory mode unlocked. 00:14:07.980 --> 00:14:10.549 But just making more isn't enough. You want to 00:14:10.549 --> 00:14:13.070 make better ads too. Right. Optimization. This 00:14:13.070 --> 00:14:15.309 goes back to those system prompts we talked about. 00:14:15.429 --> 00:14:17.970 You can create different versions tailored to 00:14:17.970 --> 00:14:20.490 specific needs. Like if you have a luxury product, 00:14:20.750 --> 00:14:24.129 you tweak the prompt to ask for a premium aesthetic, 00:14:24.429 --> 00:14:27.950 maybe soft, elegant lighting. Or for a fitness 00:14:27.950 --> 00:14:30.090 gadget, you'd write prompts demanding energetic 00:14:30.090 --> 00:14:33.309 movement, dynamic angles, maybe even visible 00:14:33.309 --> 00:14:36.549 sweat for realism. You can bake the brand tone 00:14:36.549 --> 00:14:39.070 right into the generation instructions. And you 00:14:39.070 --> 00:14:40.830 should also... test different messages, not just 00:14:40.830 --> 00:14:43.690 visuals, right? Absolutely. Use four rows for 00:14:43.690 --> 00:14:46.429 the same product image and setting, but in row 00:14:46.429 --> 00:14:49.529 one, focus the script on convenience. Row two, 00:14:49.629 --> 00:14:53.590 results. Row three, value. Row four, maybe social 00:14:53.590 --> 00:14:55.690 proof. Then you run them all and see which message 00:14:55.690 --> 00:14:58.110 actually connects with people. Exactly. Let the 00:14:58.110 --> 00:15:01.009 real world data tell you what resonates. Which 00:15:01.009 --> 00:15:04.549 brings us to the really advanced move. Closing 00:15:04.549 --> 00:15:06.789 the loop. This is where it gets really powerful 00:15:06.789 --> 00:15:09.669 integrating performance data back in. Yeah. You 00:15:09.669 --> 00:15:12.970 add a webhook node to your workflow. This node 00:15:12.970 --> 00:15:15.590 listens for data coming back from your ad platforms 00:15:15.590 --> 00:15:18.850 like Facebook ads or TikTok ads manager. Pulling 00:15:18.850 --> 00:15:21.230 in actual results, views, clicks, conversions. 00:15:21.769 --> 00:15:24.350 You configure the ad platform to send that data 00:15:24.350 --> 00:15:28.389 to the webhook. Then you have NAIMN write that 00:15:28.389 --> 00:15:31.070 performance data back into new columns in your 00:15:31.070 --> 00:15:33.690 original Google Sheet right next to the ad it 00:15:33.690 --> 00:15:35.929 belongs to. Okay, so now your spreadsheet shows 00:15:35.929 --> 00:15:38.960 not just the ad, but how well it did. And here's 00:15:38.960 --> 00:15:42.159 the final piece. You add another AI agent. It's 00:15:42.159 --> 00:15:44.480 job. Read the sheet, analyze the performance 00:15:44.480 --> 00:15:46.299 data, and figure out what's working best. And 00:15:46.299 --> 00:15:49.299 then it automatically creates new ready rows 00:15:49.299 --> 00:15:51.840 based on the winners. If the convenience angle 00:15:51.840 --> 00:15:54.340 ads got way better click -through rates, this 00:15:54.340 --> 00:15:56.659 analysis agent automatically queues up 10 more 00:15:56.659 --> 00:15:59.600 variations on the convenience theme. Wow. So 00:15:59.600 --> 00:16:01.360 the system starts teaching itself and improving 00:16:01.360 --> 00:16:03.720 automatically based on real results. It becomes 00:16:03.720 --> 00:16:06.960 a self -optimizing content engine. a true factory 00:16:06.960 --> 00:16:09.360 that not only produces but also iterates and 00:16:09.360 --> 00:16:12.159 improves based on live market feedback. Okay, 00:16:12.240 --> 00:16:14.460 so what's the ultimate goal, the big win from 00:16:14.460 --> 00:16:16.700 integrating all that factory mode stuff? Creating 00:16:16.700 --> 00:16:19.320 a self -improving loop that automatically tests, 00:16:19.659 --> 00:16:22.720 learns, and iterates using real performance data. 00:16:23.019 --> 00:16:25.600 Let's just zoom out one last time and grasp the 00:16:25.600 --> 00:16:30.779 scale here. Ten traditional UGC ads. You're looking 00:16:30.779 --> 00:16:33.580 at, what, $500 minimum, maybe up to $5 ,000? 00:16:34.120 --> 00:16:36.580 and weeks of work, coordination, back and forth. 00:16:36.779 --> 00:16:39.980 Right, versus 10 AI -generated ads using this 00:16:39.980 --> 00:16:42.799 winning workflow, costing maybe $1 .80 total. 00:16:43.039 --> 00:16:46.059 Yeah, maybe $1 .80, $2 max, and generated in 00:16:46.059 --> 00:16:48.620 minutes, ready to deploy almost instantly. It's 00:16:48.620 --> 00:16:50.600 not just cheaper. It's a completely different 00:16:50.600 --> 00:16:52.820 economic model for creating marketing assets. 00:16:53.279 --> 00:16:55.919 The advantage clearly goes to whoever adopts 00:16:55.919 --> 00:16:57.700 this kind of automation and learns to iterate 00:16:57.700 --> 00:17:00.220 quickly, like we said, testing 50 ideas for the 00:17:00.220 --> 00:17:02.419 cost of maybe one old -school ad. And what about 00:17:02.419 --> 00:17:04.250 future -proofing? Are we going to have to redo 00:17:04.250 --> 00:17:07.750 this whole thing when SOAR 4 or VO5 comes out? 00:17:07.930 --> 00:17:10.009 That's another beautiful part of using tools 00:17:10.009 --> 00:17:13.769 like NEN and FAL AI. The core logic, the workflow 00:17:13.769 --> 00:17:16.890 structure stays the same. So when a better, faster, 00:17:16.990 --> 00:17:19.990 cheaper AI model drops? You literally just go 00:17:19.990 --> 00:17:22.789 into your NEN workflow, find the node that calls 00:17:22.789 --> 00:17:25.849 the AI model, and update the model name in the 00:17:25.849 --> 00:17:28.329 settings. Maybe tweak the prompt slightly if 00:17:28.329 --> 00:17:30.869 needed. And your entire factory instantly upgrades 00:17:30.869 --> 00:17:33.630 to the next generation of AI content. Exactly. 00:17:33.750 --> 00:17:35.950 The system itself is designed to be adaptable. 00:17:36.049 --> 00:17:38.430 So for you listening, we really encourage you 00:17:38.430 --> 00:17:41.329 to start exploring these ideas. Autonomous workflows, 00:17:41.789 --> 00:17:44.670 clever prompt engineering. The barriers to creating 00:17:44.670 --> 00:17:47.950 high quality, scalable content are rapidly disappearing. 00:17:48.210 --> 00:17:50.609 Really, the main constraint now is just the quality 00:17:50.609 --> 00:17:52.829 of the ideas you feed into that initial spreadsheet. 00:17:53.190 --> 00:17:54.269 So go build your engine.