// transcript — 453 segments
0:00 Why Sora 2 Matters
0:01 So, Sor 2 has been taking the internet by storm. So, what I'm going to do today
0:04 is show you guys how you can use Nitend to get 10 times the output, higher
0:09 quality outputs, no watermarks, and use Sor 2 for six times cheaper than through
0:13 OpenAI directly. So, Sor 2 has some really fun use cases, and I love that
0:16 it's getting people who aren't super interested into AI interested in AI. But
0:20 it's more than just having these funny, cool videos. Businesses and
0:23 organizations are actually using Sora 2 to power their creatives, their media,
0:27 their content, their marketing, all this kind of stuff. Not only does it create
0:30 video, but it automatically creates the audio, too. Just look at this Starbucks
0:43 >> Starbucks. Discover your flavor. >> So, if you understand how to use this
0:45 technology, not only can you save yourself or a business a ton of time,
0:48 but you could make a lot of money. So anyways, today I'm going to be showing
0:51 you guys step by step how to connect to SOR 2 in NADN over API, but then I'm
0:55 also going to be going over these examples where you can do text to video,
0:59 image to video. You can have cameos, so people like Mark Cuban or Sam Alman in
1:03 your Sor 2 videos. You can create storyboards so you can lay out different
1:05 scenes and have consistent characters throughout. And then I'm also going to
1:09 talk about prompting. So feel free to follow along. I'm going to give you guys
1:12 this entire template for free so that you guys don't have to actually go build
1:15 this yourself. You can just use what I have here already. And you can access
1:17 that by joining my free school community. The link for that will be in
1:19 the description, but I don't want to waste any time. Let's get straight into
1:20 Connecting to Sora 2 in n8n
1:22 the video. All right, so before we dive into these examples, I want to start
1:25 from scratch and show you guys how I connect to Sora 2 in Naden. So the first
1:30 step is to go to a platform called Key AI, which is what we were just looking
1:35 at over here. It's spelled Kie.ai. In the past, you guys have seen me use
1:39 foul.ai, which is a very, very similar platform. Essentially, it's just like a
1:42 marketplace for all of these image and video generation models, as you can see
1:45 here. But here's the key difference on price. FAL and OpenAI are charging you
1:50 10 cents per second of Sora 2 video generation whereas keys only charging
1:55 you 1.5 cents per second. So it's six times cheaper to make these videos. And
1:58 of course you're still getting the same quality of output if not better. If we
2:02 make a 10-second video, that's only going to cost us 15. Whereas on Fowl or
2:06 OpenAI, that 10-second video would cost us a dollar. Anyways, the first thing
2:09 you're going to do when you get to key.ai is go to your billing information
2:12 and just make sure you have some credits. You'll probably have some for
2:15 free when you get in there, but otherwise just grab five bucks worth.
2:18 It's going to last you a long time. And then on the lefth hand side, you can see
2:21 there's a section called API key. And we'll have to use this in just a minute
2:24 or so, but just remember for now, that's where it is. But anyways, I'm going to
2:28 click on models market. We have all of these different providers and options we
2:33 can use like Google Vo 3.1, Sora 2 Pro, and for now, I'm just going to go ahead
2:36 and click on Sora 2. And so this opens up a playground environment where you
2:40 could write in here start to prompt Sora 2 and get your outputs right here just
2:43 to play around with how the prompting works and how the outputs look. But what
2:47 we're interested in is using this over API. So I'm on Sora 2 textto video and
2:51 I'm going to click on API right here. And now all we have to do is scroll to
2:54 the bottom and I'm going to show you guys how we can set up a curl which
2:58 makes it really really easy. You can see down here there is a request example and
3:01 you make sure you're on curl and then we're going to copy this right here. I'm
3:05 going to go back into Nitn. I'm going to add an HTTP request to our workflow. And
3:10 I don't have to configure this method, this URL, any of this stuff. I'm just
3:13 going to go ahead and hit import curl. Paste in that curl command that we just
3:17 copied. And then when I import it, it basically fills out pretty much
3:20 everything that we need. Now, we just have to make a few tweaks. So, the first
3:24 step would be to add your API key. So, right here it says authorization and
3:28 then it says bearer space API key. And so, this is where you would go back into
3:33 key. You go to your API key and you'd copy this value right here. And then
3:37 you're just going to paste that right there like that. And then you'd be able
3:41 to access key. And this would basically access your billing information that you
3:44 put in there. But because we're going to be making multiple requests to key, I
3:48 don't want to copy and paste this every single time I want to make a request. I
3:51 just want to save this. So what I'm going to do is I'm going to scroll up a
3:54 little bit. And right here you can see authentication. I'm going to go ahead
3:58 and open this up and click on generic. For generic type, I'm going to go ahead
4:01 and choose header because you can see right here, this is a header parameter.
4:04 And then you can see I have all of these different ones that I've already saved.
4:07 You can also see that I already have one for key, but I'm just going to go ahead
4:10 and make a new one with you guys right here to show you how it works. So, you
4:13 click on create new credential. And remember, it's the same thing as we just
4:16 saw down there. So, for the name, we're going to type in authorization. And for
4:19 the value, you're going to type in capital B bearer space and then paste in
4:25 your API key. And then when you go ahead and save that, now you're connected to
4:29 key and you can name this credential like keyai and save that. And so now I
4:34 have this saved every single time that I need to use key, which as you can see in
4:39 this workflow, each one of these little workflows is one that we're going to be
4:43 using key. So now I just have it saved and all I'd have to do is choose it
4:46 right there rather than go back and copy my API key every time. So just a little
4:49 fun trick. And then once you do that, you can just turn off the headers cuz we
4:52 don't need to send our API key twice. But from here, what we have is the JSON
4:56 body. And I'm just going to change this to an expression and open this up full
4:59 screen so we can all look at it. So what we're looking at here is basically all
5:03 of the little filters that we're sending over to Sora 2 and saying this is the
5:07 type of video we want. We can see we have model is Sora 2 textto video. We
5:11 have a callback URL, which I'm just going to go ahead and delete because
5:13 that's an optional field. The only reason I know that's optional is because
5:17 in this API documentation is what it's called. It basically shows you what you
5:21 need to send over. So you can see the model that's required callback URL
5:27 optional input prompt required aspect ratio optional. So you can actually go
5:32 ahead and look at this API documentation and understand how you can change the
5:37 behavior of the Sor2 API. So I'm not going to dive too deep into that right
5:40 now. If you want to deep dive, I made a full course on that. You can go ahead
5:42 and watch that video. I'll tag it right up here. Anyways, going back into Nitn,
5:47 we can see that we have a prompt. We have um an aspect ratio. We have a
5:51 number of frames which we can choose between 10 or 15. And then we have
5:55 remove watermark true. So right here, what I'm going to do is just change this
5:59 prompt. And this is turning our text into video. So I could just say a video
6:05 of a young man throwing a coffee mug against the wall. So we have our JSON body ready. What I'm
6:11 going to do now is hit execute step. So this just executed. It gave us a 200
6:16 code which is good. It says success. And then we see our task ID is this long
6:20 string. So if we go back to the documentation now, we can see what
6:23 happens is when we make that request, Sora 2 or key basically says, okay, we
6:28 got this request. We're working on it. And so what we need to do next is
6:31 actually grab that back. So there's another endpoint right here that says
6:35 query task. I'm going to click on that. And all I'm going to do is copy this
6:38 curl statement once again. So the request example, I'm going to copy that.
6:41 We're going to come back into NAN and we're going to add another HTTP request
6:46 and once again we're just going to import that curl. And so this is going
6:50 to set us up with what we need. The first thing you'll notice is that we
6:53 have a task ID parameter. So basically it's saying what request are you looking
6:57 to get back. So I'm going to delete this example task ID and all I have to do is
7:01 drag in the task ID that we just got from the previous node. So I'll put that
7:04 right there. Now that's a dynamic expression. And then the last thing you
7:07 can see is once again it's sending over our API key. So, I'm just going to turn
7:11 off the headers because we know we already set this up up here as a generic
7:15 as a header. And then we should have our key API somewhere in here. There we go.
7:19 Key AI. And so, now we're set up once again. And so, I'm going to go ahead and
7:23 execute this step. And you can see when I run that, it comes back and it says,
7:26 okay, the state is generating. And while this is generating, I'll just show you
7:29 guys in key how you can actually check on your requests. So, if I go to my
7:34 logs, you can see right now that this says it's running. And you can see my
7:39 past runs have taken 195 seconds, 227 seconds. So it may just take like 3 to
7:43 four minutes for us. But this lets us look at all of our inputs. So we can see
7:46 our prompts. We can see all of the things that we've requested from um
7:50 these different models. So anyways, I'll check back in with you guys once this
7:53 one is done. Okay, so I ran it again and you can now see that it says state
7:57 equals success. And then down here, what we get is a result. So we got two
8:00 different URLs. We got one with a watermark and one without a watermark.
8:05 So, if I go ahead and copy this URL and I just basically say go to this URL, it
8:09 downloads a file. And when I open up that download, this is what I get.
8:22 Okay, that's pretty ridiculous. And first of all, what you'll notice is it
8:25 was like it was in slow motion. The sound was a little weird. And the reason
8:29 why this all happened is because we hardly prompt this thing at all. If you
8:33 remember in the request that we made, I literally just said a video of a young
8:36 man throwing a coffee mug against a wall. And so, you really can't expect to
8:39 get a good output if you don't prompt it very well. And so, I'll talk a little
8:42 bit later about prompting and how you can really get some cool outputs from
8:46 Sora 2. But anyways, the point of what I just did there right up here was just to
8:49 show you guys that we're making two requests. The first one we set up to
8:52 say, "Hey, Sora, here's the type of video I want." The second one we set up
8:56 to say, "Okay, Sora, is that video done yet? Like, can I can I see it?"
8:59 Basically, so we're basically going to follow that pattern for all of these
9:02 different workflows and it will start to make sense. But like I said, you'll be
9:05 able to download all of this. So you'll be able to play around with it and see.
9:07 How Polling Works
9:09 The first example that we have here is turning text into video, which is kind
9:12 of what we just did up here. But what I wanted to introduce to you guys is this
9:16 concept of polling. So let me real quick start this request and then I'll explain
9:20 what polling is. So, if you guys got into this template, all you'd have to do
9:23 is go to this video prompt node. And right here, you could basically just
9:26 input your video prompts of what you want to get back. So, right here, we
9:29 have the default example about a professor giving a lecture and
9:32 explaining that Sor 2 is live. So, what I'm going to do is execute this
9:35 workflow. We're going to see this run. We can see what happened here is it made
9:39 the request. So, Sor 2 is currently working on our video. Then, we have a
9:42 wait node because we know this takes anywhere from, you know, 2 to 3 to 4
9:47 minutes. But what you just saw happen is we checked in to see if it was done
9:51 after the wait node and it's not done yet and it just happened again. So this
9:55 is going to be an infinite check where it's going to every 10 seconds go ask
9:59 sor 2 if it's done until we know for a fact that it is done. And this is called
10:03 polling because we're constantly making checks. And this is better than just
10:06 estimating and saying okay well roughly this takes 3 minutes so I'm going to
10:09 just set my weight for 4 minutes to be safe. Well that could be inefficient.
10:13 And also what happens if for some reason it takes five minutes and then your flow
10:16 moves on and there's just a big error. So anyways, the reason why this works is
10:21 because in this if node what I'm doing is if I just make this section a little
10:25 bit bigger over here, the state equals generating. And what I'm doing is I'm
10:30 saying if the state equals success, then we're good. And then we go up the true
10:33 branch. But you can see we've had six items come through and they've all gone
10:37 through the false branch because the state equals generating. There was
10:40 another seventh one. So, it's a super simple conditional check. We're just
10:44 looking to see if it's done or not. Now, one thing you would want to consider is
10:47 potentially using a switch node because there's other states that could happen.
10:51 You could get success. You could get generating. You could potentially get
10:54 failed. And if you get failed, it wouldn't really know what to do here.
10:57 So, if you wanted to make this more production ready, you would probably
11:00 also work in a conditional check to see if the state equals failed. And that
11:04 would send you some sort of notification or something like that. All right. So,
11:06 that just finished up. You can see that it took 18 tries, so about 180 seconds.
11:11 And then what we get at the end is our final video URL. I did a quick
11:17 expression within this JSON variable to isolate just the link that we want. So,
11:20 I'm going to go ahead and open this up and we'll take a look.
11:23 >> And here's the exciting part. Sora 2 is now available on Kai AI, making it
11:26 easier than ever to create stunning videos. >> You can experiment, iterate, and bring
11:31 your wildest ideas to life right from your lap. >> Okay, I mean, that's pretty good. You
11:34 can see there's dialogue. You can see there's energy. And once again, we
11:37 didn't even really put any best practices with prompting into play yet.
11:41 Image to Video
11:41 So, we'll show that near the end of the video. But let's move on to this next
11:44 example, which is turning an image into video, which is really, really cool. And
11:48 I think this is what unlocks tons of potential. So, what we have here is
11:51 similar. When you guys get in here, you'll have an image URL. So, right
11:54 here, I have this image URL, which if I open this up real quick, you can see
11:59 it's a AI generated image of a fake curl cream product for your hair. And then if
12:03 we go to the video prompt, you can see I'm basically saying a realistic UGC
12:07 style video of a young woman with curly hair sitting in her car recording a
12:10 selfie style video explaining what she loves about the product. So UGC ads is a
12:15 huge use case here because that's like what converts really well nowadays
12:18 online on Tik Tok shop, stuff like that is just real authentic people holding
12:23 something with a selfie style, you know, video and just saying, "This product is
12:26 awesome. This is why I love it." And so imagine you have a product and you can
12:30 just pump out five organic videos like that every single day without hiring
12:34 actors or anything like that. So what I'm going to do is just go ahead and run
12:37 this workflow and then we'll dive into once again how it's working. So you can
12:40 see we have the polling flow set up very similar to the way we had it up here.
12:44 The only difference really is in this HTTP request to key is we have the video
12:50 model which is not text to video. Now it's imageto video. We have the prompt
12:54 once again, but now we have a section called image URLs. We're sending over
12:57 that public image which the model will use as reference. So in our final video,
13:01 it should be holding the product that looks exactly the same as our source
13:04 image. As you can see here, I said nothing about the product from the
13:07 source image should change. It should appear exactly as given. We're also
13:10 telling Sora that the woman in the video should say, "I absolutely love this curl
13:14 cream. It keeps my hair bouncy, curly, and lightweight all day long. You guys
13:17 have to try it." So you can have full control over what the AI person in the
13:21 video is saying. You could also control their accent and their tone and their
13:25 style, that kind of stuff, too. Just keep in mind that for this image URL, it
13:29 has to be a publicly accessible file. So, it can't be local. And then for
13:32 aspect ratio, we said portrait, and that's pretty much the only difference
13:34 because we want it to be like a Tik Tok or an Instagram reel style. So, this is
13:38 running. This is doing its polling check. And I will check in with you guys
13:42 once it is finished up. All right, we just got that one back. Let's go ahead
13:45 and take a look at this video. >> I absolutely love this curl cream. It
13:49 keeps my hair bouncy, curly, and lightweight all day long. You guys have
13:53 to try it. >> Okay, that's insane. If you guys see what happened here at the end, she puts
14:00 it near the camera and this looks pretty much identical to the source image that
14:04 we had. It has the same font. It has all of the words looking really good and it
14:07 has that little logo. One thing you'll notice though about Sora 2 right now is
14:11 that the first like millisecond will be your source image. Right now, there's
14:14 just not great support for this, but I imagine that will be fixed very, very
14:17 soon. But you can see there's the original image and then there's the
14:20 product appearing in the video. So it's like pretty much the exact same. One
14:23 other thing that you may notice is if you want to do image to video with Sora
14:28 2, you can't have it be a person in the image. It will basically tell you we
14:32 can't do this because it's a realistic looking person. even if that person is
14:36 AI generated, unlike V3 where you could give it an AI generated person holding
14:39 your product in case you wanted that to be like your brand ambassador on all of
14:43 your videos and then V3 could take that image of the person holding the product
14:46 and turn that into a video and that would help you sort of like leaprog over
14:49 the issue with the first millisecond being the source image, but that's very
14:53 easy to crop out. But anyways, I think that those UGC content use cases are
14:57 super cool because you could basically just keep generating those and it will
15:00 be a new person every time and it will look and feel very real and it will look
15:04 good for your brand. And once again, that would only get better if you
15:07 Video Cameos
15:07 actually prompted it with best practices. Okay, so this next thing is
15:11 really cool. This is using cameos, which basically means you can use famous
15:15 people's faces and likeness in your videos. So on the Soru app, it's kind
15:19 of, you know, like a social feed and people have profiles. So, if I went here
15:23 and I searched Sam A, this is going to pull up Sam Alman's profile. And you can
15:26 see here in his profile, if I go to cameos, we can see all of the times that
15:30 people have made videos with Sam in them. And all we have to do to use Sam's
15:33 cameo because he made it publicly accessible to people is we just copy his
15:37 username, which is Sam A. So, in our Cameo example here, I have Sam A
15:42 recording a selfie style video in a car explaining how Gravity works in two
15:45 short sentences. So, let's go ahead and run this one. And it's going to do the
15:48 same polling flow and everything like that. Keep in mind you can also use
15:54 cameos in your um image to video generations as well. Right now I'm just
15:57 doing a text to video to keep it simple. But one thing you may notice with cameos
16:01 is Sora 2 may be a little bit more restrictive. So I tried doing some
16:05 crazier prompts or I tried throwing in some other cameos like Shaq and Mark
16:09 Human and it was kind of rejecting my prompts. So it just may be a little bit
16:12 sensitive especially right now especially with so many people using
16:16 cameos and using Sora 2. So just keep that in mind. Anyways, I will check in
16:19 with you guys once this is done polling and once our video has been generated.
16:23 All right, looks like we got that done. I'm going to go ahead and download this
16:27 video and we'll see what Sam has to say for us. So, here we go.
16:31 >> Everything with mass pulls on everything else. That pull makes things fall toward
16:36 Earth. [laughter] >> Okay, that was kind of ridiculous, but
16:41 looked like him, sounded like him. we maybe just would want to prompt that a
16:44 little bit better so he doesn't sound like he wants to die because I literally
16:48 just said a selfie style video in a car explaining how gravity works in two
16:51 short sentences. And what's cool is you can go ahead and create an account on
16:54 Sora and you know do the process of setting up your own cameo and then you
16:58 could use that in your automations. So you can have you know an avatar of
17:01 yourself on your social media and you can just generate all these videos. So
17:05 Storyboards
17:05 very cool stuff. All right, so now let's look at another really cool feature
17:08 which is the storyboards. And to explain how this works, I'm real quick going to
17:12 actually switch back to KAI and we're going to click on right here Sora 2 Pro
17:16 storyboard. And for those of you wondering, we see Sora 2 text to video,
17:20 Sora 2 Pro text to video. Honestly, I haven't seen a huge difference. And for
17:25 the cost, how cheap Sora 2 regular is on KAI, I would just stick with this for
17:29 now. But if maybe you do start to scale up the system and you want all of your
17:32 content going out on Tik Tok or Instagram and you want to use Pro, then
17:35 go for Pro. But anyways, for the storyboard, how this works is you're
17:40 able to set different scenes. And like I said, you can have consistent characters
17:43 within those scenes. And you're able to allocate a different amount of time per
17:48 scene as long as it adds up to the total duration, which you can choose between
17:52 10, 15, or 25. But as you can see, it has to allocate correctly across the
17:56 three scenes. So anyways, what I would do is I would go ahead and copy the
17:59 request example once again like you guys saw in the demo. And what I did here was
18:02 I set this up so you guys could basically put in an image URL. And then
18:07 you could have three different scenes. So this image, as you can see, is an AI
18:11 generated image of a curious, adventurous frog. And then our three
18:15 scenes are basically that frog finding treasure and jumping around in the
18:18 forest. So we're going to go ahead and run this. I will say I have sometimes
18:24 had a 500 internal server error from key when I've been trying to do these
18:26 storyboards, but we're going to go ahead and give it a try. So I'll execute the
18:29 workflow. It's going to run our prompts through this request. It's going to go
18:32 ahead and do the polling feature. And I will also say when I've done storyboards
18:36 in the past, it's taken anywhere from like 500 to 700 seconds rather than just
18:43 like, you know, your typical 180 to 250. So, this one may take a little bit
18:46 longer, but I'll just check in with you guys and see if it worked or not. All
18:48 Sora 2 Prompting
18:49 right. So, while this one's running, I thought it would be a good opportunity
18:52 to just talk about prompting because this has taken 740 seconds and I've just
18:56 been sitting here staring at it and I'm getting bored. So, let's talk about
18:59 prompting a little bit. So, there are people right now that are making a lot
19:02 of money by going into like marketing departments and teaching them how to
19:06 prompt Nanobanana, which is one of Google's image generation models and
19:11 things like Sora 2 or Google V3. Because if you understand how to prompt these
19:14 things and you can make high quality like UGC ads or VFX ads or scenes for
19:20 movies and TV, whatever it is, there's a lot of money in that space because it
19:24 costs so much money to get drones out there or have all the mics and you have
19:27 to make sure the weather's correct when you're doing all of this in real life.
19:30 And so, if you can just generate these scenes with good prompting, like I said,
19:35 it's a really, really cool opportunity. So, I'm not a prompting expert when it
19:39 comes to these creative generation models, but what I do know is there are
19:44 certain things that you want to have in your video prompts. So, here is a very
19:48 very basic highle prompt that I have for this AI agent that's going to take our
19:52 raw input, optimize it, and then shoot it off to Sora 2. So, I'm sort of going
19:55 to skim over this, but you guys will be able to access this for free when you
19:58 download the free template. And so what I have here is you are an expert AI
20:03 video prompt engineer trained to design optimized prompts for Sora 2. Your role
20:07 is to take a raw input concept and transform it into a highly detailed
20:11 video prompt. So this agent needs to always describe the main subject like
20:14 appearance, clothing, age, gender, expression, and motion. The setting, so
20:18 what's the location, what's in the background, what's the lighting like?
20:21 What's the time of day? All this kind of stuff. The camera style, so the angle,
20:25 the lens, the type of video, if there's any camera movement, all of that stuff
20:28 really matters. The idea is that the prompt should sound like a professional
20:32 cinematographer describing a shot to an actual like visual effects team. So
20:35 anyways, this agent will take a raw input and transform it. So what I'm
20:39 actually going to do is I'm going to stop this generation because we will be
20:43 able to check on that in our key dashboard right here. You can see this
20:46 one is still running. And what I'm going to do is drag this down here and run
20:51 this one. And so if you remember what we did earlier with our texttov video
20:55 prompt which was up here we had the one of the professor saying to its students
21:00 that sor 2 is available on key. We have that exact same system prompt. So this
21:03 is the exact same raw input that we gave earlier. And now what we're having the
21:07 agent do is take that raw input and make it better and optimized for sore 2. So
21:12 you can see now we get this huge input. We get a lively cinematic classroom shot
21:17 as a sequence of natural documentary style coverage that highlights a
21:20 charismatic professor and the engaged students. We've got a wide shot which is
21:26 24 mm tripod dolly. We later have medium twoshot which is 35mm gentle handheld.
21:30 We've got a different type of reaction cutins with 50 mm. We've got the um
21:35 quote right here. We have lip sync. We have all of this kind of stuff. It even
21:39 gives us the overall tone, directorial notes, all of this stuff. So now what's
21:44 happening is that got sent off to Sora 2 as you can see and it tells us that once
21:48 again it's generating this message and so I'll check in with you guys when this
21:51 one has been completed and we'll compare it to that video that we had earlier.
21:54 All right, so this one finished up. It was a bigger prompt so it took a little
21:57 longer. It took about 300 seconds. So first let's watch the original one that
22:01 had no prompting, just the raw input. >> And here's the exciting part. Sora 2 is
22:05 now available on Kai AI making it easier than ever to create stunning videos.
22:08 >> You can experiment, iterate, and bring your wildest ideas to life. Okay, nice.
22:13 So, that's still not bad, right? But look at this one. Because we had the
22:17 best practices of prompting worked in, >> Sora 2 is now available on KA AI, making
22:23 it easier than ever to create stunning All right, so I hope you could tell that
22:29 the second one felt a lot more cinematic and there was different shots and there
22:32 was like different scenes going on in there. So, I just thought that that's
22:35 pretty cool. And once again, the idea is all you have to do is give it the raw
22:38 input and then the AI agent that you prompt to specialize in making it into a
22:42 video prompt is going to take care of that for you. And so, yes, it's nice
22:46 that we have kind of like a highlevel here are some good practices for
22:50 prompting. But what you would do is when you have your specific use case, like
22:53 let's say it's your UGC ads, on top of giving it just these basic rules, you
22:57 would come in here and really tailor this towards specifically UGC content
23:02 video prompting. And as you refine that prompt and make it better and better,
23:05 your outputs are just going to get better and better as well. So, for
23:08 example, and just to sort of hint at a future video that may be coming, you
23:11 could have a Google sheet like this where you have a product photo of your
23:16 own product, you have just the ICP, you have the features of the product, and
23:19 you have a video setting, and that's all you have to give it. And then your AI
23:23 agents in NN could take all of that and make the optimized script, make the
23:27 optimized video prompt, and then you're just getting all of these UGC content
23:30 videos pumped out automatically because you could have 10 coming out a day, 10
23:34 coming out an hour, however many you want. Okay, so the storyboard video that
23:40 we generated together live here, it took 35 minutes, but it did finish up. So,
23:43 let's go ahead and give it a quick >> Okay. So, that's pretty funny. You can
24:03 see what it did though is it was able to use our image URL that we provided right
24:08 here. So, this is the character that we wanted it to be and it was consistent
24:11 throughout the three scenes. it would now just be a matter of making these
24:13 prompts a little bit better because all of these prompts had, you know, not
24:17 really the elements that we discussed as far as lighting, background, camera
24:21 movement, camera style, all that kind of stuff. But hopefully now you can just
24:24 get a sense of how those storyboards work and how you can control the scenes
24:28 and timing to create some consistent character videos. The final thing I want
24:29 Avoiding Tricky Errors
24:31 to talk about here is doing some data cleanup. So something that you guys
24:35 didn't see that happened in this step up here was our AI agent output this prompt
24:41 with new lines. So we had new lines and then also down here you can see we had
24:46 double quotation marks and both new lines and double quotes will break the
24:51 JSON body request. So as a best practice you can use these expressions which I'll
24:55 show you guys in a sec in your request to Sora 2. So in this body request you
25:01 can see I don't just have the output. Let me actually just delete this real
25:04 quick. So this is the output of the agent and on the right hand side is the
25:08 result. So you can see there's new lines here which would break this. And then
25:12 also we have our wherever they are somewhere down here. Right here we have
25:16 the double quotes. Sor 2 is now available on KI. And so when we put in
25:20 this little expression, it's basically replacing those new lines. As you can
25:23 see, they got chopped off. And I'm not going to be able to find it again now,
25:26 but the double quotes would have been removed as well. And that's how you make
25:29 sure no matter what your AI agent outputs, because sometimes they'll throw
25:32 in new lines and double quotes even if you prompt them not to, you can
25:34 basically get rid of that no matter what. And so what I wanted to show real
25:38 quick is if I run this setup down here, so you guys can kind of just isolate
25:42 those variables and look at it. We have right here we have new lines. So this is
25:45 line one, this is line two, this is line three, and then we have another text
25:48 over here with double quotes that says pizza pizza. And if we run this second
25:52 one, we have the replace function up here to get rid of new lines. And then
25:55 we have the replace function here to get rid of the double quotes. And now you
25:57 can see on the right hand side they're both coming out completely clean. So
26:01 that's how those replace function works. And then you can basically just copy
26:04 these and save them. So whenever you need to use that replace you've got it
26:06 Sora 2 Logs
26:07 right here. The last thing I wanted to talk about is errors. When you're maybe
26:12 getting a failure in your sore video generations. So here are some failures
26:15 that I got. You can see all of these have error code 500 which means there's
26:19 something internally going on and that's why it errored. And by internally I mean
26:24 on the key or sor 2 on the server side of things. So, this could mean that
26:27 they're getting way too many requests and there's just an issue. It could mean
26:31 that AWS blacked out and they're down or something like that. But it could also
26:35 mean that your content is being restricted because maybe you put
26:38 something in there that is automatically getting flagged and they're just
26:41 basically rejecting you. So, here's an example of a Sor 2 Pro storyboard that I
26:45 did and you can see it got rejected. Internal error. I'm not exactly sure
26:49 why. But then here's another Soru storyboard that I did and it was
26:53 successful and this one took um almost 7 minutes. So, this was the highle video
26:57 of different ways that you can use sore 2. If you guys have some specific use
27:00 cases that you'd want to see me build an automation for or certain agents for,
27:03 then definitely let me know down below. I'd love to bring some more Sor 2 and
27:07 some VO 3.1 content for you guys cuz I think this creative AI space is super
27:10 Want to Master AI Automations?
27:10 cool. And once again, you guys can download this entire template and get
27:13 all of this stuff so you can play around for completely free. All you have to do
27:16 is join my free school community. The link for that is down in the
27:18 description. When you join this free school community, all you have to do to
27:22 find that is go to YouTube resources or search for the title of this video. And
27:27 in that post associated with the video, you'll find the JSON right here. And you
27:30 just have to download that JSON, import it into your NADN, and then this exact
27:35 workflow will pop up. There will also be a setup guide over here that shows you
27:37 like what you need to connect and all that kind of stuff. And if you're
27:40 looking to dive deeper with AI automations and connect with over 200
27:43 members who are learning every day and building businesses with AI every day,
27:46 then definitely check out my plus community. The link for that is also
27:49 down in the description. Like I said, it's a super supportive community of
27:52 members who are building businesses and sharing what they're learning. And we
27:55 have three full courses. We have Agent Zero, which is the foundations of AI
27:59 automation for beginners. We've got 10 hours to 10 seconds where you learn how
28:02 to identify, design, and build time-saving automations. And then for
28:06 our annual members, or if you've been with us for 6 months, we've got
28:10 oneperson AI automation agency where we start to talk about how you can actually
28:12 get in front of business owners and start to sell these solutions. On top of
28:16 that, we also have one live Q&A per week where I get to just basically talk to
28:19 you guys and have some pretty fun discussions. So, I'd love to see you
28:22 guys in those calls. I would love to see you in these communities. But that's
28:25 going to do it for today. So, if you enjoyed the video or you learned
28:27 something new, please give it a like. It definitely helps me out a ton. And as
28:30 always, I appreciate you guys making it to the end of the video. I'll see you on