How to Deliver a Full AI Ad Campaign with One Script | Seedance 2.5 API Tutorial

So these are eight commercials for my cold brew brand Alto. So we have four markets, two formats each, and that's a 30 second master cut for YouTube and TV and a 15 second vertical cut for shorts and Instagram reels. And I didn't click a generate button once to do that. I just wrote every job into one config file. And then a script on its own submitted all eight, tracked them, and then downloaded them into the right folders. And when the client asked for changes, I just reran the two versions that needed them. So this is a pretty cool step forward for making tons of awesome audio and visual content with AI. And so today I'm going to show you that full loop, step by step, from start to finish. So how the brief becomes API requests, how the tasks get tracked, how the files come back to you. That's going to be the video. It's going to be a fun one. Let's go. So first, why the API at all? We've covered a lot of tools on this channel where you can just type into some website and you can get a video back. And that's what's important, I think. If you want one video generated by AI, a web tool is great. But that's not really what real client work looks like, and you might have a product for a client that's launching in 19 different markets. You want three videos per market. That's something like 50, 60, 70 different videos that we need segmented out with a high quality bar, but we need all of them to be also generated by AI. And so all of those deliverables need the same logo. They need to look the same. They need a constant. Tracking, quality control, et cetera, kind of shared context mechanism. And if you do that in a browser, that's a little bit of a different story. But if you do that in an API and you just write one file with a certain format, then it's all programmatic, it's all deterministic. And if revisions come in, it's pretty easy to segment out the changes you want. And so that essentially is the cycle that we're looking at in the value add here, where if you want to deterministically produce these kinds of videos in bulk, You can do it with an API, but then what does that workflow look like? So we're going to use one of the best, if not the best video model out there. That's SeaDance 2.5. If you follow the channel at all, you definitely see them. We use it everywhere, but it's a video model out of ByteDance. And BytePlus is ByteDance's enterprise AI platform. And they partnered with me to make this video happen. So through BytePlus, you get the official Seedance 2.5 API that means official pricing. And then you also get the newest versions first. Now for the workflow, we're going to build two features of SeaDance 2.5 are most important. The first is that Seedance 2.5 can generate continuous 30-second generations. And then second, you can take up to 50 reference assets to create those videos. That means we can have a Product, the logo, et cetera, that is locked and looks the same throughout videos up to 30 seconds in length. And to get that right, we can upload up to 50 images of our logo, of our product, of a scene, or all of those things at once that the model will use and internalize to ensure that quality bar is there. Now, as far as pricing goes here, if you wanna follow along with the demos that we're going to do. To run the API, you need a Seedance 2.5 resource plan. So BytePlus does not hand out free credits. So you'll buy some kind of prepaid token quota. And then that will be spent through the API. So there are three plans. Lite is $32 for 5 million tokens. Production is $64 for 10 million tokens. And then premium is $640 for 100 million tokens. So tokens on every plan are valid for a period of three months. That's important to keep in mind. And if you run out, then it switches to a pay-as-you-go model. So we're going to use the production plan in this video. So there's 64 bucks for 10 million tokens. So if you wanna follow along support the channel there's a resource plan link in the description so you'll buy the plan create your API key and then just follow along with me on the script side of things to run that and that said without further ado let's jump into my computer to get started with the build so let's go ahead and buy the production plan on this demo account to get started with our demos okay that Okay, that went through, now let's click on Model Ark. Now, what you should do if you're offering services off the back of this is just price and understand your margins through the lens of those token plans. So you can say, okay, if my client wants 25 videos this month, it takes me 75 generations to come to 25 total. Well, how much do 75 cost? Maybe that plan costs $120. $20 and so if I sell this at A thousand dollars, I have your 80-90% margins. So now from this page, I can go to my profile and I can click on API key. Now that brings me to this page where I can go ahead and click create API key. Now I'm going to name this Auto Launch Batch in line with the demos we want to create. I'll click create. And now I have my API key here. I'll go ahead and copy that key. And now we can go ahead with this API key and our 10 million tokens, and we can use ChatGPT Astra to create the script that will generate our videos in bulk. I can go under here to select the GPT-4o Astra media model, and I'm going to paste this pretty lengthy prompt. And what we're going to do is put this exact prompt in the description of this YouTube video so you can just copy it directly into your Workflow or adjust it for your business and what this says is you are the production lead for Alto a premium cold brew coffee brand We're launching in four markets and each market needs two deliverables a 30 second master and a 15 second vertical cut So what I want is a task config file then a python script that's going to read the API key and then download each finished video for me and then a flag so that I can run specific videos again if I want to without touching all of them. So that said after those load let's go ahead and let this prompt run. So that said after those load let's go ahead and let this prompt run. Okay we can see that's done. So we can see here we have our reference images, our Python script that we'll use to generate these videos in bulk through the BytePlus API. Let's go ahead and check in to this script. So here we can see the API right. So all this script is really doing is letting us go and use that API that we've set up and just automating that whole sequence with this And so this script on the right here executes the batch script or the prompt which we can see on the left here. So these are all of the eight videos that we're going to generate. And so the Python script is going to execute then with these prompts on this alto underscore batch dot JSON. And so if we want to go in and we don't like how one of the videos turns out, we can just go ahead, you know, alter the prompt right here or just re or rerun just one or two. of these videos instead of doing it all over again. I'm going to jump into our downloads and then I'm going to and then I'm going to paste the prompts to go ahead and run this. Okay so we've gone into downloads and so here we can see the first of those has been queued. Second of those has been queued as well and we'll let it run for a minute and then we can check back once the videos are done. Okay, we can see the script has finished running. See, all eight of it was generated. Now the usage is going to take a few minutes to update, and then we can go check in to see. Let's go ahead and just check out the results here and see how good they are. France is number one. Really awesome quality. Okay, yeah. So literally no errors. We actually have the French super high quality generation. We can see the full 30 second cut. Also really good quality here as well. And then same thing here, right? We have our Japanese version, other Japanese version, Spanish. These are incredibly high quality generations, completely honest to the references. And that was all completely generated from the script. Honestly, I'm happy with all eight of these. So that's a win in my book. And let's play four of them on the screen side by side now, then we'll move on to the next demo. And so let's say we do want to revise one of those videos. I can just talk to ChatGPT to upgrade the script if I want to. Or like I was saying, I can go into here. I can edit the prompt manually if I want to in this file. Or since these were created by Astra, I can just jump into Astra to ask for the change. So I just went ahead and did that and it's going to go ahead and update the script. So with the script updated let's go ahead and give it this prompt. So just rerun the Japan 30 second and 15 second cuts. Okay here we can see the video was downloaded. Let's check out this result. Looks awesome. So we have the Japanese localized text and we have our updated color set. So overall, there you go. You can create in bulk and you can revise and iterate easily and do it in a highly programmatic fashion. This is, I think, how people are going to start to incorporate to very serious enterprise app scale. AI video generation into what you're doing. If you're a solopreneur, if you're just working on some tasks through a dashboard here and there for video ads, that's great. But if you want to generate 200 videos a week for a bunch of clients in your custom coded software, this arguably I think at least is the way to do it and there's a lot more here that you can do inside of model arc inside of byte plus we're not going to get into all of them in this video that'll be for another time and I really just wanted to hone in on that core workflow of using this this programmatic fashion that we haven't covered on the channel before to make your videos with currently one of the if not the best video model in the world that being Dreamina The Seedance 2.5. Now my overall take on it is that the real work here is in the setup. If you've never used an API even with AI you're going to need some time to get used to that approach and yes once you set it up it's going to be a lot more efficient but just keep in mind you want to allocate some time initially for that setup because the setup determines the quality and the scale that you can get down the line so that is really important. Now, 30 second takes, of course, are going to use the most amount of tokens. So what I would recommend, like we went over, just test first with lower resolution runs or with shorter runs. And then once you're sure you have what you want, then use that as a reference to go apply the best model on the longest generation with the best. quality to get the final deliverable and on that note as well references take thought the quality of the references matter you shouldn't just go take 50 images in your finder you don't really know what they are just plug them in there see what happens you want 10 or 20 or 5 whatever the number is really good references that show say from different angles and different settings the product or the logo and then feed that into the agent that is how you're going to get the best quality results now once you've done all that legwork you your references you have your scripts you have your setup you've done it a few times then it gets really efficient then you just swap out some elements talk to your agent to get that done and you can create it scale right so what we're really talking to here i think about The downside or the challenges with this system is just that setup period. Now, there you go. That's everything I wanted to cover. Now, again, if you want to support the channel and give this a go for yourself and you haven't already, then you can use the link in the description to buy a SeaDance 2.5 resource plan on BytePlus, and that's at byteplus.com. Now, once you've done that, like we've gone over, you're going to create your API key in the console. Then put it in an .env file. Then you'll paste the prompt we used. That'll write the script. And then just go ahead and run it and iterate from there and create whatever it is you want to create. So that's the loop. Again, a pretty cool shift, I think, and something that'll really start to get used and implemented into real production workflow so you can create, you can distribute at scale. I think there's a lot of room, a lot of mispricing and arbitrage out there for all of you who are willing to hustle and make it happen with this explosion of tools that we're seeing. So that's the video. Appreciate you being here and I will see you in the next one. Cheers.