I Built the Ultimate UGC Content System with AI Agents (free template)

0:00 VO3.1 Nano Banana Sora 2.

0:02 There are all of these amazing models dropping.

0:04 So, I figured why not just build a system

0:06 where we can use all of them.

0:07 So, what we're going to be looking at today is

0:09 the ultimate UGC ads system where all

0:11 you have to do is fill in some raw

0:12 information on a Google sheet like a

0:14 product photo, the ICP, the features of

0:16 that product, and a setting of the

0:18 video.

0:18 And then all you have to do is

0:19 come in here and choose your model.

0:21 Whether that is V3.1, a combination of

0:23 Nano Banana and V3.1, which is super

0:25 cool.

0:25 I'll show you guys exactly how we

0:26 do that in a sec here, or using Sora 2.

0:29 This lets you seamlessly test a bunch of

0:30 different creatives and product features

0:32 and settings across a ton of these

0:34 different AI video generation models.

0:36 So, the question that we're going to be

0:37 trying to answer today is which one is

0:39 best for UGC ads.

0:40 So, taking a look at

0:41 this workflow, you can see that there's

0:42 basically three paths.

0:43 There's the VO3.1 path, the Nano Banana Plus V3.1 path,

0:47 and then the Sora 2 path.

0:49 So, we're going to jump into a live demo.

0:50 We're going to run all three of these paths,

0:51 and I'm going to explain what every

0:52 single node is doing so that you guys

0:54 can set this up for yourself.

0:55 And as always, I'm giving away the entire

0:57 system for free.

0:58 All you have to do is

0:58 join my free school community.

0:59 The link for that is down in the description.

1:01 So, before we go ahead and run the live

1:02 example, let's look at a few of our

1:04 outputs that we've already gotten with

1:06 this exact system.

1:07 So, the first product

1:08 we tried was creatine gummies.

1:09 Here is what the actual product photo looks

1:11 like.

1:11 So, you can see it's a creatine

1:12 gummy jar.

1:13 We then have the ICP, which

1:14 is young adults wanting to stay fit.

1:16 The product features for this are delicious

1:18 gummies, easy to remember to take daily,

1:20 makes workouts better, more energetic,

1:21 stuff like that.

1:22 In the video setting,

1:23 we have a young man who is parked in his

1:25 car about to go into the gym holding the

1:27 gummies.

1:27 So, the first one we'll look at

1:28 is Nano Banana Plus Google V3.1.

1:31 I love that these creatine gummies

1:33 actually give me more energy for my sets

1:35 and they're tasty, so I actually

1:36 remember to take them every day.

1:39 All right, here's the same one with Sora

1:40 2.

1:41 I love these creatine gummies.

1:43 They actually taste amazing and I never

1:45 forget to take them.

1:46 They make my workout stronger and I feel more

1:48 energized.

1:49 And then here's V3.1.

1:51 These taste amazing and I actually remember them

1:53 every day for a my workouts feel

1:55 stronger and I've got more energy.

1:57 You may have noticed a few things with

1:58 the reference image and the way they

1:59 were speaking, but let's continue on to

2:01 the second example which was hair shine

2:03 spray.

2:03 So, I'm going to go in the same

2:04 order.

2:04 Nano Banana Plus VO3.1 Sor 2 and

2:07 then V3.1.

2:08 I love how this gives my hair that

2:10 glossy finish without any greasiness.

2:13 It dries instantly and feels weightless.

2:16 I love how this gives instant glossy shine

2:17 without any greasiness.

2:18 It dries fast and feels weightless.

2:20 I love how this

2:21 adds instant gloss without feeling

2:24 greasy.

2:24 It dries so fast and leaves no

2:26 sticky buildup.

2:28 All right, so we've seen a few examples.

2:29 We'll come back at the end and compare

2:30 more outputs and see which one we

2:32 ultimately deem being the king of these

2:34 models.

2:34 But let's go ahead and do a live

2:35 example.

2:36 So the first two that we did

2:37 were AI generated images.

2:38 This first one was creatine gummies, as you can see,

2:41 and the second one was our hairspray,

2:43 which looked like this.

2:44 So, what we're going to do for the third example is a

2:46 real product image, and this is actually

2:48 from an Amazon listing.

2:49 So, it is a

2:50 portable neck fan like this.

2:51 We have the ICP of middle-aged adults who spend long

2:54 hours outdoors or landscapers,

2:56 construction workers.

2:57 We have product features like it's comfortable, it's

2:59 light, it delivers powerful air upward

3:01 and downward, and it regulates your body

3:03 temperature.

3:04 And we have the video

3:05 setting for a friendly middle-aged woman

3:07 tending her garden in the afternoon sun.

3:09 So, hopefully you guys can see the value

3:10 prop here.

3:11 it'd be really easy to just

3:12 throw in your product information right

3:14 here and then have this thing every day

3:15 create tons of UGC ad content for you.

3:18 Then what happens in the workflow is it

3:20 takes that and we have different AI

3:22 agents here that are trained to prompt

3:23 in different ways and that's how we're

3:25 optimizing you know the features and ICP

3:27 and the setting to actually go into this

3:30 UGC content.

3:31 So I'm going to go ahead

3:31 and hit execute workflow.

3:32 It's going to pull in that data from the sheet.

3:34 It's going to do one row first and the first

3:36 row that it's doing is nano Banana plus

3:38 V3.1 because as you can see right here

3:41 it's basically processing this row and

3:42 that's the model that we chose was

3:44 NanoBanana plus V3.1.

3:46 So I'm actually just going to start to explain what's

3:48 going on here as this is running.

3:51 So you can see here we're pulling in data from

3:52 this sheet, right?

3:53 The only thing special going on here is we're making

3:55 sure that the status column equals ready

3:58 because we don't want to pull in all of

3:59 these rows that have already been

4:00 finished.

4:01 And then we also turned on

4:02 this option that says return only the

4:03 first matching row because we don't want

4:05 to do, you know, all six of these at a

4:07 time.

4:07 We want to just do one by one.

4:09 You could obviously change that if you want,

4:10 but that's the way we're rocking right

4:11 now.

4:12 Anyways, we then go into this

4:13 switch node and what happens here is it

4:15 basically just checks what the model was

4:17 selected as.

4:18 So if it was V3.1, it goes

4:20 up.

4:20 If it was nano plus V3.1, it goes to

4:23 the middle.

4:23 And if it was SOAR 2, it

4:25 goes down.

4:25 As you can see, these three

4:26 paths.

4:27 So, this one was obviously V3.1

4:29 plus nanobanana, which is why it went

4:31 here.

4:32 And that's why we're doing this

4:33 first step, which is an image prompt.

4:35 So, let me explain why I'm doing this.

4:37 What we're starting with is a picture of

4:39 our product because we need to make sure

4:40 that the product image looks actually

4:43 good in our final copy.

4:44 Otherwise, we're not going to be able to sell any of

4:46 that.

4:46 So, in my mind, the most ideal way

4:48 to do this is to take that product image

4:50 that we're given.

4:50 So, if we just want to

4:51 get a quick refresher, taking this

4:53 product image right here and using AI to

4:56 turn this into an image where someone is

4:58 wearing it or holding it and then we can

5:01 take that optimized image and turn that

5:03 into a video.

5:04 And so, ideally, I would

5:05 do this also for Sora.

5:06 But when you send

5:07 a image to Sora, if it looks like a

5:10 realistic human person, even if it's an

5:12 AI generated human, it's going to reject

5:14 it.

5:15 Google VO3.1 however does not reject

5:17 it which is why we have this little

5:19 extra bonus method here.

5:21 Now the workaround here is if you do Sora 2 you

5:23 can use cameos.

5:24 So if you haven't seen

5:25 that before then I'll drop my video I

5:27 made with Nitn and Sora 2.

5:28 I'll tag it right up here and you can see you could

5:30 use cameos.

5:31 So you could create one of

5:32 yourself or you could use some other

5:33 person and have them being in your

5:35 content with your product something like

5:38 that.

5:38 So anyways we're using Nano Banana

5:40 to create an image of the product being

5:42 held or worn by a person.

5:43 And then we take that image and we turn it into a

5:45 video with VO3.1.

5:47 So anyways, you can

5:48 see that that actually just finished up.

5:49 So that's telling me I need to speed up

5:51 a little bit.

5:51 Let's click into this AI

5:53 agent to understand how it is making an

5:55 image prompt.

5:56 We're giving it two

5:56 things.

5:57 We're giving it the product,

5:58 which as you can see, if I open this up,

6:00 it's coming through as portable neck

6:02 fan, and we're giving it the image

6:03 setting, which is actually just the

6:04 video setting, but it says, "A friendly

6:06 middle-aged woman is tending her garden

6:08 under the sun.

6:09 She pauses, smiles at the

6:10 camera, and gestures toward the sleek

6:11 fan resting around her neck." So the AI

6:13 agent takes that information and then it

6:15 reads through its system prompt to

6:17 understand what do I need to do with

6:18 that information.

6:19 I'm not going to read

6:20 this entire system prompt, but you guys

6:22 will be able to once again download this

6:23 template for free and you can dive into

6:25 this and understand why I have it set up

6:27 this way.

6:27 One thing I did want to

6:28 preface though is I made this workflow

6:30 to be a template.

6:31 So these system prompts are not perfect or optimized and

6:33 it would really be on you to get in here

6:35 and customize them a little bit for your

6:36 use case, but it gives us a great place

6:38 to start.

6:39 So anyways, you are an expert

6:40 in hyperrealistic UGC userenerated

6:43 content photography and your role is to

6:46 generate detailed image prompts, not the

6:48 images themselves.

6:49 So you will be

6:50 provided with a product photo which

6:51 should not be changed or altered in any

6:53 way.

6:54 And you will also be given a

6:55 specific setting or scene description.

6:57 So it knows that its role is to create a

6:59 prompt.

6:59 So we come in here and we give

7:00 it some prompt guidelines.

7:02 We talk about human realism.

7:03 We talk about product

7:04 accuracy.

7:05 We talk about composition and

7:07 perspective.

7:08 We talk about lighting and

7:09 environment.

7:10 We talk about authentic

7:11 details, technical style.

7:12 And then finally, some critical instruction like

7:14 only outputting the image prompt, not an

7:17 actual image or you know, hey, here's

7:19 your image prompt.

7:20 You know, we just

7:20 want the prompt.

7:21 So, out of that, what

7:22 we get is our image prompt.

7:24 And you can see it's pretty detailed.

7:25 It has stuff like lighting.

7:26 It has stuff like camera

7:27 angle and composition and stuff like

7:29 that.

7:30 And we're able to take that

7:31 output, feed it into the next node,

7:34 which is our HTTP request to a service

7:37 called Key AI, which lets us access tons

7:39 of different AI image and video

7:41 generation models.

7:42 So this is key.

7:43 As you can see, we have tons of stuff like

7:44 VO3.1, Sora 2 Pro, 40 image, Flux

7:48 Context, Cling Turbo.

7:50 It's kind of like

7:50 the open router for image and video

7:53 generation models.

7:54 So, I'm not going to

7:54 deep dive into exactly how I set up this

7:56 API call, but definitely go and watch

7:58 that sore video if you haven't because I

8:00 actually go step by step and show you

8:01 guys how I did that.

8:02 I'll also tag right

8:03 up here an API video that I made, which

8:05 you should watch anyways because it

8:06 really explains APIs and agents and

8:08 stuff like that.

8:09 Anyways, essentially what we're passing over here is our JSON

8:12 body, which is the most important part.

8:13 The model that we want to use is nano

8:15 bananait.

8:16 We're sending over the input

8:17 prompt, which as you can see right here

8:19 is coming through.

8:20 This is the output of

8:21 the image prompt agent that we just

8:22 looked at.

8:23 Now, there is one thing I did

8:24 here that's kind of special is I

8:25 replaced new lines because you can see

8:27 if I get rid of this expression real

8:28 quick, what happens is we get these

8:30 little line breaks in here and we don't

8:32 want that because that will actually

8:33 break our request to key AI.

8:35 So, that's why I use that little expression.

8:37 I also talk about that in the Sora video.

8:39 And then we're giving it the image URL,

8:41 which is the one that came from our

8:42 Google sheet right here as you can see.

8:44 Finally, we're just saying we want this

8:46 to be vertical because a lot of times

8:47 the UGC content is kind of selfie style

8:49 and it's for like a Tik Tok or an

8:51 Instagram reel.

8:51 So that's what we do

8:52 there.

8:53 Once key gets this request, it

8:55 basically says to us, okay, cool.

8:57 I got all this information.

8:59 We're working on that right now.

9:00 And so the next step

9:01 that we move into here is a wait node.

9:04 You can see that I have this set up for

9:06 5 seconds.

9:06 So it goes ahead and it waits

9:08 for 5 seconds and then it checks in on

9:10 key and says, "Hey, do you have my order

9:12 done yet?" And we're able to get to that

9:14 by sending over the task ID of the

9:16 previous order.

9:17 So it's like when you go

9:17 to a food truck and you order your food

9:19 and it says, "Okay, your order number

9:20 43." This is basically you walking back

9:22 up to the truck and saying, "Hey, I'm

9:23 order 43.

9:24 Is it done?" And they'll

9:25 either say yes or no.

9:27 And that's why we

9:27 use this little if node right here,

9:29 which is basically our yes or no check.

9:31 And we're looking to see if the state

9:32 equals success.

9:34 Because if you look at

9:34 the first time we checked in, the state

9:36 equals waiting.

9:37 The second time that we

9:38 checked in, the state equaled waiting.

9:40 And finally, the third time we checked

9:42 in, the state equals success, which

9:43 means that our order is ready.

9:45 And so notice that we have false branch or true

9:47 branch, and it's true when it's done.

9:49 So what we do is if it's false, we have

9:51 this line that goes back to the wait.

9:53 So this is why you can see it waited three

9:55 times, which means this took about 15

9:56 seconds to generate.

9:57 And so the first

9:58 time it wasn't ready, it came back.

10:00 Second time it wasn't ready.

10:01 We checked in again.

10:02 And then the third time after

10:03 it waited again, it was done.

10:05 And so when that's done, what we end up doing

10:07 is we want to real quick analyze that

10:09 image to see what is actually in there.

10:11 So here's the actual image that it

10:13 created for us, which looks awesome.

10:14 It's a green portable neck fan.

10:15 She's in her garden, and it even matches the

10:18 writing, as you can see.

10:19 See, if we go

10:19 back to the source image, there's a

10:20 little bit of gold text right there.

10:22 There's these circles.

10:23 So, that looks really good.

10:24 And so, I basically grabbed

10:25 this open AI note and said, "Describe

10:26 what's in the image.

10:27 Describe the environment." Stuff like that.

10:29 And we get back, the image features a woman

10:31 standing outdoors in what appears to be

10:32 a garden.

10:33 The environment has raised

10:34 garden beds, blah blah blah.

10:35 The woman is wearing a light blue shirt.

10:37 She has her hair pulled back.

10:38 Around her neck, she has a green wearable device that

10:40 looks like a personal neck fan.

10:41 Blah blah blah.

10:42 So, the reason why I wanted

10:43 to analyze the image real quick is

10:45 because the next step is to use another

10:47 AI agent to create a video prompt.

10:49 And in order to create a video prompt that

10:51 is consistent with our image, not only

10:53 are we going to give it that image, but

10:54 we also want to give it a quick analysis

10:56 of what is actually in that image so

10:58 that its prompt is consistent.

11:00 And I have tried doing this without the

11:01 analyze image step and it still works.

11:03 But doing this, it just seems to be

11:04 higher quality.

11:05 So, anyways, we are

11:07 hitting another AI agent.

11:09 This time we're giving it a little bit more

11:10 information because keep in mind this

11:13 agent isn't just creating a video

11:15 prompt.

11:15 It's also creating the dialogue

11:17 that the person in that video is going

11:19 to say.

11:20 And so in order to do that, we

11:21 give it the product.

11:22 We give it the

11:23 product ICP.

11:24 We give it the product

11:25 features.

11:25 We give the video setting.

11:27 And here's where we give it the reference

11:28 image description.

11:29 So this is the

11:30 analysis of that image.

11:31 So it looks at

11:32 all that information and it says, "Okay,

11:33 what do I do with that?" And so now we

11:35 have our system prompt.

11:36 Once again, not going to read the whole thing, but you

11:38 guys can have access to it for free.

11:40 So, we said that your role as an expert UGC

11:42 video creator.

11:43 Your task is to generate

11:44 a prompt for an AI video model like

11:46 VO3.1.

11:47 Your goal is to create a

11:48 realistic selfie style video that

11:50 appears to be filmed by an influencer

11:51 using one hand to hold the phone and the

11:53 other to interact with the product.

11:55 The video needs to feel authentic, which is

11:57 why UGC ads are converting so well right

11:59 now because it's just real people

12:01 speaking real raw thoughts.

12:03 Anyways, we gave it some requirements like subject

12:05 and framing.

12:06 We talk about the visual

12:07 style.

12:08 We talk about tone and dialogue.

12:10 We give it some technical specs.

12:11 We give it some embedded elements in the prompt.

12:13 As you can see, we tell it that it's

12:15 going to get a reference image and it

12:16 needs to match that appearance and tone.

12:18 And then a real quick output prompt,

12:21 which is pretty concise.

12:22 And honestly, it looks like I might have accidentally

12:23 cut off the last sentence here, but

12:26 hopefully it still came out all right.

12:27 And so after that, we get this output.

12:29 You can see it starts off with a natural

12:30 selfie style 9x6 vertical video, 8

12:33 seconds long.

12:34 friendly middle-aged woman, gardener.

12:35 She's filming on her

12:36 phone.

12:37 She's wearing a light blue shirt.

12:38 And then down here is where you can see

12:39 what the dialogue says.

12:40 So, I love how

12:41 it's so light.

12:41 I almost forget it's on,

12:42 but it pushes a ton of air and the

12:44 battery lasts all afternoon.

12:45 So, that basically took the product features that

12:47 we had given it and it made a quick

12:49 little blurb for this influencer to say

12:52 in the video.

12:53 Now, we're going to take

12:54 this video prompt and we're going to

12:56 feed that into key once again and we're

12:59 going to send it to VO3.1.

13:01 So, here is our HTTP request where we're submitting

13:03 an order to VO3.1.

13:06 I'm going to open up this body, and you

13:07 can see that we have a prompt, which is

13:09 exactly what we just got from the lefth

13:10 hand side.

13:11 Now, the reason it looks all

13:12 messy like this is because I'm actually

13:13 using three replace functions.

13:15 I'm just going to replace new lines, which we

13:17 already talked about.

13:18 I'm going to replace double quotes right here.

13:20 It previously said, I love how it's so

13:21 light and pushes a ton of air, and this

13:23 was wrapped in double quotes, but we

13:25 took those away because that will also

13:26 break the JSON body.

13:28 And then I also had

13:28 to add another one.

13:30 Sometimes based on your chat model, it can be really weird

13:32 and output these double curly quotes

13:34 which don't actually get captured with

13:36 this previous replace function.

13:37 So I threw in this one just as an extra

13:39 guardrail which you guys will already

13:40 have all this set up.

13:41 So you should be

13:41 good to go.

13:42 But now we're basically

13:43 ensuring that our request will go

13:45 through.

13:45 You can see once again we're

13:46 giving it the image URL except for this

13:48 one is actually the image you know it's

13:50 this one that Nano Banana made for us.

13:52 And then for the model we're saying V3

13:54 fast.

13:54 We're using fast instead of

13:56 quality because it's cheaper and it's

13:57 faster and it's still really good.

13:58 And I know this says V3, but trust me, this is

14:00 using V3.1.

14:02 And then aspect ratio 9x6.

14:04 We wanted to make sure that it matches

14:05 the source image.

14:07 So now that we have

14:08 that, it basically does the exact same

14:09 thing.

14:10 It gives us back a order number

14:11 or some sort of ticket.

14:13 And we go ahead

14:13 and wait for 10 seconds right here.

14:15 We then go ahead and check back in on this

14:17 request, giving it our order number to

14:19 make to see if it's done or not.

14:21 And then you can see this happened eight

14:23 times.

14:23 And so we basically checked in

14:25 eight times.

14:25 So a total of 80 seconds.

14:27 So almost a minute and a half.

14:28 And then when we realize that the order is

14:30 actually done, we go ahead and we write

14:31 back to Google Sheets.

14:33 And let me show

14:33 you real quick how we set up this Google

14:35 sheet right back.

14:36 So we're using the

14:37 operation to update the row.

14:39 And we choose our sheet.

14:40 Of course, we shoot we

14:41 choose our document.

14:42 And then it says

14:42 that we have to match on a certain

14:44 column.

14:44 So what we decide to do is match

14:46 on the column number.

14:48 So you can see

14:49 right here, all of these rows have a

14:50 different unique number.

14:51 And when the workflow gets triggered, if we go all

14:54 the way back down to our initial get

14:56 rows, you can see that this row came in

14:58 and it was row number 10 or technically

15:00 row number 11, but the number was 10.

15:04 And so we're basically going to drag in

15:05 the number right here and say, okay, the

15:07 row that we want to update is the row

15:08 where the number column equals 10.

15:11 And so that's why it was able to write back

15:13 to this row right here, which you can

15:14 now see has been changed to status

15:16 finished.

15:16 And we have our finished file

15:18 right here.

15:19 because in Nitn we manually

15:21 set the status to be finished and then

15:23 we drag in the finished video URL that

15:25 we just got back from our key request.

15:29 And so that's basically the full process

15:31 and that's the most complicated one

15:32 because both the top one and the bottom

15:34 one are just doing reference image to

15:36 video rather than reference image to

15:38 image and then taking that image to

15:40 video.

15:41 So anyways, we just covered the

15:42 hardest one and then we'll look at the

15:43 other ones.

15:44 But real quick, let's just

15:45 go look at the actual output because of

15:47 course I'm very curious.

15:48 I love how it's

15:49 so light I almost forget it's on, but it

15:51 pushes tons of air and the battery lasts

15:55 all afternoon.

15:56 That's really impressive.

15:56 I was nervous to see because it's different from

15:59 someone holding a product.

16:00 She's actually wearing it.

16:01 But I mean, the

16:02 voice was really good.

16:03 The tonality was good.

16:04 I thought that this was an

16:05 impressive result.

16:06 But let's move on to

16:07 the next one, which is Sora 2.

16:09 So, what I'm going to do is go back into the

16:11 workflow and I'm going to execute it.

16:13 What this is going to do is pull in the

16:15 next And you can see it got pushed down

16:17 to Sora 2 because when it does this

16:20 check for the model, it knows that the

16:22 model was right here marked off as Sora

16:24 2.

16:25 So I'm honestly not going to spend as

16:26 much time in these next two flows

16:28 because you guys pretty much already

16:30 understand exactly what's going on.

16:32 We have this video prompt agent which once

16:34 again is looking at the product, the

16:36 product ICP, the product features, and

16:38 the video setting.

16:39 The only difference here is that it doesn't have a analysis

16:42 of the reference image because it'll

16:44 just be given that.

16:45 But the system prompt once again we basically say

16:48 you're an advanced UGC video creator.

16:50 You're optimizing for video prompts for

16:51 Sora 2.

16:52 Here is what you'll be given.

16:54 And we go over basically the same exact

16:56 headers.

16:56 Subject and framing, visual

16:58 style, uh tone and dialogue, technical

17:01 specs, prompt, construction, instructions, and an example output

17:05 prompt as you can see down there.

17:07 So what that does is it once again it

17:08 outputs us a video prompt.

17:10 And you can see in this one there actually are new

17:12 lines.

17:12 So, good thing we have that

17:13 guardrail baked in to get rid of those

17:15 new lines.

17:16 As you can see in this HTTP

17:17 request to key, we fill in our body by

17:21 saying, okay, the model we want to use

17:22 is store to image to video.

17:24 Here is the prompt.

17:25 And of course, we're using all

17:26 of those nasty replace functions once

17:28 again.

17:28 We've got the image URL, which

17:30 we're grabbing from the Google sheet,

17:31 which once again looks like this right

17:33 there.

17:33 And we're basically just sending

17:34 all of that over.

17:35 And so, it's going to

17:36 take that video prompt and it's going to

17:38 take that source image and it's going to

17:40 turn that into a video.

17:41 We're doing the exact same thing here where, you know,

17:43 we submitted the order, we have to wait

17:44 10 seconds and then check in and we're

17:46 going to go ahead and constantly be

17:48 checking until we know that our video is

17:50 done.

17:50 On average, I have been seeing

17:51 that V3 fast is finishing in anywhere

17:54 from a minute to 2 minutes.

17:55 And Sora 2 has been taking typically a little bit

17:57 more than that, maybe a minute and a

17:59 half to 3 minutes.

18:00 There are a few

18:00 things to consider.

18:01 Sometimes if you do

18:02 something like a cameo, it's going to

18:03 take longer.

18:03 If you've got a really long

18:04 video prompt, it'll take longer.

18:06 Also, what can influence it is how many people

18:07 in the world are trying to use keys

18:09 endpoints.

18:10 that can make it take longer,

18:11 too.

18:11 But typically, Google V3 fast is

18:13 faster, but it's the exact same flow

18:15 from there.

18:16 We're pulling it back in.

18:17 We're doing the same match to update the

18:19 row, and then we're just updating the

18:20 status of finished.

18:21 And we are putting

18:22 in the final video link into the Google

18:24 sheet.

18:24 There you go.

18:25 It looks like it

18:25 just finished up.

18:26 Let's go back into the

18:27 Google sheet.

18:28 It just got marked as

18:28 finished.

18:29 And we have our file.

18:30 So, let's take a look at the Sora 2 output.

18:33 Man, this thing is so light and the

18:35 airflow hits my face perfectly.

18:37 Keeps me cool while I work.

18:39 And the battery lasts

18:40 for hours, so I don't have to worry

18:41 about it dying out here.

18:43 Man, well, that was another really good

18:44 one.

18:44 A little bit confused where this

18:46 thing came from.

18:46 That was a bit of a

18:47 hallucination, but as you can see, this

18:49 was the reference image, and it looks

18:50 really good in this video.

18:52 Super authentic, and it looks like she's

18:53 obviously standing there taking a selfie

18:55 video.

18:55 All right, so the final one for

18:56 this example is V3.1.

18:58 So, I'm going to

18:59 go ahead and zoom out a little bit, hit

19:01 execute workflow, and it should shoot it

19:02 up this top branch now.

19:04 And I'm honestly not even going to break this down

19:06 because it's the exact same thing.

19:08 I copied over basically the exact same

19:09 system prompt.

19:10 I just switched out 10

19:12 seconds, which is how long the Sor

19:13 videos are, for 8 seconds for how long

19:15 the V3.1 videos are.

19:17 And then I switched

19:17 out sore 2 for V3.1.

19:19 But I wanted to

19:20 keep these prompts across all of these

19:22 flows as consistent as possible to kind

19:25 of limit the variability that we have in

19:27 order to truly see the power of these

19:28 models when we have as many things

19:30 consistent as we can.

19:31 So, I'm just going

19:32 to let this finish up and I will check

19:33 in with you guys when we get our

19:34 finished output from V3.1.

19:36 All right, so you can see that that one just finished

19:37 up.

19:38 Once again, took about 80 seconds.

19:40 Let's go ahead and make sure we got this

19:42 updated.

19:42 And let's take a look at the

19:44 V3.1 output.

19:46 I love how light this is turning.

19:47 It actually blows enough air to keep me

19:50 cool for hours while I'm working.

19:52 Okay, so it's not too bad.

19:53 I honestly think that right now my my order is this

19:56 exact order that we have here, which is

19:58 Nano plus V3.1, then Sora 2, then just

20:01 V3.1.

20:03 A lot of these VO ones have like

20:04 this super HDR weird orange glow looking

20:08 effect.

20:08 I'm not sure if you guys had

20:09 noticed that.

20:10 Here's another example of

20:11 the VO3.1.

20:12 It's not it's not terrible,

20:13 but just in comparison to some of the

20:15 other ones, it definitely looks a bit

20:16 more orange.

20:17 And then another thing I

20:18 noticed is this is the V3.1 example, and

20:21 I explicitly told it to not change

20:23 anything about the reference image

20:24 itself, but we can see the creatine

20:26 gummies is a jar and then in the video

20:28 it's a bag.

20:29 And so it does have the same

20:30 branding and same font as you can see.

20:32 He even actually, this is funny, he's

20:33 got the logo on his hoodie, which is

20:35 honestly a nice touch, but this is a

20:37 bag.

20:37 And in the source image, it was a

20:39 jar.

20:39 And the other creatine ones didn't

20:41 have a bag.

20:41 They had the the correct

20:42 jar, too.

20:43 And I know we didn't look at

20:44 the forearm strengthener example, but

20:46 this was another one where, for example,

20:47 Nano Plus V3.1.

20:49 Let me just show you

20:49 guys this one.

20:50 I love that the

20:51 adjustable resistance actually makes my

20:53 grip get stronger week to week, and it's

20:55 small enough to use right at my desk.

20:57 like super good, super natural, and the

20:59 product photo looked exactly like it did

21:01 that I gave it, which looked like this,

21:03 as you can see right here.

21:04 But then that same one with VO3.1 without Nano Banana.

21:08 Once again, it looks a bit it has some

21:10 weird shadows and it looks orange.

21:11 I love how the adjustable resistance

21:12 actually lets me progress without extra

21:14 gear.

21:15 But it the product photo also once again

21:17 does not exactly match the source image.

21:19 So that's kind of like a huge no no for

21:22 me.

21:22 And another thing that you guys will

21:24 notice when you send a source image and

21:26 you turn that into a video both with

21:28 V3.1 and Sora is the first frame is the

21:31 reference image.

21:32 And this is that first

21:33 creatine example we looked at with sore

21:35 2.

21:35 And you'll notice the very first

21:36 frame is once again the reference image.

21:38 And so when we do nanobanana plus Google

21:40 V3.1 it still does that but our

21:44 reference image is this.

21:46 So it just is

21:46 able to you know pick up right from here

21:48 and it looks way more natural.

21:50 So the point being you could kind of like

21:52 automate the content creation and you

21:54 could have it auto post as well with

21:56 this branch, but I probably wouldn't

21:58 auto post Google V3 like this or sore 2

22:01 like this because of that whole first

22:03 frame, first couple milliseconds thing.

22:05 Now some people argue that it's good

22:07 because then you have a thumbnail, but

22:08 then every single thumbnail on your feed

22:10 would look the exact same and that I

22:12 think would just come across really bad.

22:14 So that's why I think right now my

22:16 favorite is honestly Nanobanana plus

22:18 V3.1.

22:20 Now I think Sora would give it a run for

22:22 its money if it allowed you to upload a

22:24 realistic photo of a human.

22:25 Because if we go back to this first example with

22:27 the portable neck fan when Nano Banana

22:29 made that image, even though this is a

22:31 fake person and an AI generated image,

22:33 if you tried to feed that into Soore 2,

22:35 it would block you because of content

22:37 restrictions.

22:38 So that's why this combo

22:39 has my vote right now.

22:40 But another thing to consider, of course, is cost.

22:43 So comparing these options, I guess there

22:45 were technically three, but let's just

22:46 look at these two because V3 is in here

22:48 for both.

22:49 But option one is nanobanana

22:50 plus V3 fast.

22:52 When you're going through

22:53 KI, which is the one that we were on

22:55 right up here, a nano banana image is

22:57 going to cost you 2 cents.

22:58 So not bad.

22:59 And an 8second V3 fast video will cost

23:01 you 30.

23:03 So total cost per piece of

23:04 content with this system would be 32.

23:07 Now, for option two, if you're using

23:09 Sora 2 and you're going through KAI,

23:11 which is the cheapest I've seen it, so

23:12 definitely do that.

23:13 It will cost you 15

23:15 cents for 10-second video.

23:16 So, really not bad at all.

23:17 About half the cost of

23:19 VO3 fast.

23:20 So, option one is roughly two

23:21 times more expensive than Sora 2.

23:23 So, the question is, is it two times higher

23:26 quality and will it result in two times

23:28 more conversions?

23:28 Or maybe it's not

23:29 exactly a two times match because it's a

23:31 lot cheaper than how much money you'd

23:33 make per sale or whatever it is.

23:35 But there is a bit of a trade-off there

23:37 because you can essentially make double

23:38 the amount of short form UDC content

23:40 with SOR 2 for the same price as using

23:42 Nano Banana and V3 Fast.

23:44 So anyways, I just wanted to sort of give you guys all

23:46 the info, give you the template, show

23:48 you the system, and explain the

23:49 differences between these two models.

23:51 And of course, I'm really really bullish

23:52 on all of this because the fact of the

23:54 matter is you guys can get in here and

23:56 make these prompts better.

23:57 You can play around with different chat models if you

23:59 want.

23:59 We used GPT5 Mini for all of them

24:01 as you can see here.

24:03 And think about in

24:05 6 months from now, a year from now, how

24:07 much better these models will be when

24:08 Sora 4 comes out and when V4 comes out.

24:11 They're just going to get better and

24:12 better and better and cheaper and

24:14 cheaper and cheaper.

24:15 Anyways, I don't want this video to go too long, but I

24:17 did say that you guys could access this

24:18 entire template for free.

24:19 So, all you have to do is join my free school

24:21 community.

24:21 The link for that will be

24:22 down in the description.

24:23 There will also be a full setup guide right over here

24:25 when you download this template.

24:26 And when you join my free school community,

24:28 this is what it will look like.

24:29 You'll just have to click on YouTube resources

24:30 or you can search for the title of this

24:32 video.

24:33 And when you click on the post

24:34 associated with the video, you will have

24:36 right here the JSON file to download and

24:38 you import that into niten and any other

24:40 guides or PDFs that you need.

24:42 I will also write here similar to this post, I

24:44 will include the link to copy this

24:46 Google sheet template so that you guys

24:48 can plug everything in and have a very

24:50 minimal amount of custom configuration

24:52 and just start, you know, producing

24:54 these types of results.

24:56 And if you want

24:56 to see me actually build this system

24:57 live and just kind of talk about what

24:59 I'm doing, why I'm doing it, and my

25:01 thought process, then definitely check

25:03 out my plus community.

25:04 The link for this

25:04 will also be down in the description.

25:05 We've got a great community of over 200

25:08 members who are building with naden

25:09 every day, asking questions, sharing

25:11 what they're learning, helping each

25:12 other out, and a lot of these people are

25:13 building businesses with NAND right now.

25:16 We've also got a classroom section with

25:17 three full courses.

25:18 We've got agent zero, which is the foundations for

25:20 beginners.

25:21 We have 10 hours to 10

25:22 seconds where you learn how to identify,

25:23 design, and build time-saving

25:25 automations.

25:26 We have one person AI

25:27 agency which is for our premium members

25:29 laying the foundation to build a

25:31 scalable AI automation business.

25:32 And then here's the course I was just

25:34 talking about with projects where we

25:35 actually dive into step-by-step setups

25:38 of practical workflows that you can

25:39 actually use.

25:40 Probably one of the best

25:41 ways to actually learn NN in and out.

25:43 We also have one live call per week.

25:45 They're super fun.

25:46 Everyone gets on there and we ask questions and we have

25:48 some cool conversations about the space,

25:49 the industry, all this kind of stuff.

25:51 So, I'd love to see you guys in those

25:52 live calls in the community.

25:53 But that's going to do it for today.

25:54 So, if you enjoyed this one or you learned

25:56 something new, please give it a like.

25:57 It definitely helps me out a ton.

25:58 And as always, I appreciate you guys making it

26:00 to the end of the video.

26:01 I'll see you on

26:01 the next one.

26:02 Thanks everyone.