# How to Build an AI Video Producer with Claude & HeyGen | Automate Your Content Creation

Channel: The Boring Marketer
Video: https://www.youtube.com/watch?v=AeWk31GLWeQ
Duration: 25 min
Language: English
Words: 4329
Transcript page: https://viewrankai.com/tools/youtube-transcript/AeWk31GLWeQ

---

[0:01] Hey Vue, how are you? We're back with uh with more MCP use cases today. Yep, I'm good. James, how are you? I'm doing well. Uh a lot going on. A lot of new members to our online community. I'm heading out to San Francisco this week to hang out with uh with Greg and do some fun stuff. But uh yeah, wanted to kick off the week here with uh another great MCP use case. Yep. For for myself, my weekend was so good because, you know, with the MCPS that we we did so far and I thought we could do something more and let's just move out of text and

[0:36] do something with video and I did not really think that it would come out well, but here we are. Okay, awesome. So the first one that we did, we basically created our own analytics assistant where we built an inate workflow that connected with Google Analytics and piped all that into claude so that we could chat with our data, build interactive dashboards and essentially create our own marketing analyst. The second one, we created our own social media uh content assistant essentially that pulled in a bunch of top performing posts from various channels, allowed us to find out why they were working, identify pain points, and then automatically spin up full-blown content

[1:16] calendars with completed posts ready to go. So, uh I'm I'm really curious if you can top that uh with what you're going to show us today. For me this is I thought this is very interesting because you know how it is right like when you have to like shoot a video uh like edit a video do a lot of things around video it takes a lot of time especially the research script and things like that I thought why not we make use of clot and like you know get the clot to do that so that's the premise behind it. All right. So, we're going to create our own uh social media

[1:53] video producer uh with MCP here today, I guess. Yep, you're right. All right. Well, let's see what you have. Let me just share my screen. Okay. There are way too many N8 open in my screen. So, okay. Me for that. Okay. You're able to see my screen, right? Yep. Awesome. Just going to zoom in. So, usually when you see the MCP triggers that we've created for the last two MCPS, it was like something very simple and small, but this time we have so many things added over here. Um, just to give context, this is nothing but uh this you know there's YouTube and in YouTube we are going to search for a

[2:43] specific keyword like look for multiple videos around it. pick some top videos and based on the top videos we're going to transcribe the top videos. Let's say we're going to do it one by one. Let's say there are about 10 videos, 10 top videos and we're going to take one or two videos from there, transcribe them and find an like analyze it what's there, what is missing, what could be better in these videos and based on that particular topic we are going to like build a script. uh and that script using that script we're going to create that video and all of that is going to happen

[3:15] only in claude we don't have to come down to any other platform or tool to basically like generate a video or write a script or anything of that sort okay so the first step is we're going to identify a keyword that is important to our audience or our niche so I think in the example you're about to show us we use like marketing automation or something along those lines is that right okay So basically we're taking the marketing automation keyword, we're going to YouTube, we're finding top videos, we're transcribing them, and then we're doing kind of a gap analysis there. Uh you know, what is missing from

[3:51] these videos that our audience might be interested in. So from there, we're creating a script and then you're using the Hey Genen API to create an AI UGC video of me talking through this script automatically. And we're just able to pump out 60-second short form clips uh basically on autopilot. I'm really curious to see how this works. I've tried Hen a few times. Uh you can still tell it's kind of AI, so there's probably some things we can do. um you know integrating other tools in order to improve the final output but uh you know there's a lot of nodes here why don't you just give us like a quick highlight

[4:33] of of what these are and what each of them are doing absolutely uh so this one is basically the research part that we mentioned where we are using the YouTube API to basically get the videos for uh you know the keyword marketing automation. So this is just the base URL along with the uh API key that you need to give. So you will basically say get and you would basically get the it's going to scrape the YouTube videos based on the keyword. Okay, this is needed because what happens with YouTube when we scrape at the first go is it wouldn't give us certain details like view count

[5:14] uh URL in the first time. So we are going to use this code tool which is going to help us fetch the video ids. Uh each YouTube has a video each YouTube uh video has a YouTube video id and that video id using that we will get the URL of the YouTube we will get the view count like count share count and all of the numerical data that we're looking

[5:35] for with this particular tool node. Okay. And based on that what we would be doing uh this would still remain red because u we've not run it. Uh whenever only we run like when the cloud is working this would probably appear green or something. Usually in N8 when this is in red it means it's undefined because see there's no input connected. So that's why we get this. So again here we are going to use the video ids and identify the URL and the numerical

[6:06] related data with the HTTP node. Okay. So once that is done we are going to basically so we would get uh the URLs we would get the video ids and things like that the view count and all of it. Now we are going to just get the URL separately because when we want to transcribe a video and things like that we would need URLs separately. So we are going to get the URL alone along with the view count and things like that. So we are using this code code node to basically separate all of that instead

[6:40] of an unstructured single data. Okay. Now this is the transcript node. What this will do is basically like I've used an appy uh YouTube transcript scraper which is going to help us get the transcript. Uh the biggest uh mistake or the uh you know issues that we will face initially here is that when we use when we don't use the code node we would still be able to transcribe it but the problem is this would be like one single item all 50 or 100 searches that you do will still remain as one item not as 50 different items. So that's why we use the code node and even over here since

[7:22] it is single different thing we have put it in such a way that it is going to transcribe one by one it cannot take all of it together so in MCP we don't have an option to loop it so that's why in claude we are basically like using it to like take a specific video to transcribe so this particular node is going to help us transcribe the video that we are picking up okay so there's a lot of steps initially to sort of like get the video and get the video information and the video data and then structure it in

[7:56] a way that we can really uh utilize it. Is is that a good summary? Exactly. So I think these are all already there uh you know these things were already there in the video uh you know the YouTube workflow that we did before uh that we've shared in the resource. So it's pretty much the same. I've just added over here. The steps there are also like pretty much the same because when you first scrape the YouTube like I said you won't get all of these data. In order for us to get the data the structured data we are actually adding multiple uh tool nodes over here. Okay cool. So

[8:31] what's next? Next is this. This is the hijen part. So even in Hijin we have multiple things right like for example there's photo videos like where you just include your photos and the photo uh looks like it's like you know uh it becomes a video you'll be able to like build a video on top of it and there's avar videos avar video is nothing but where you have recorded one or two videos and stored it in hij and it will take that as an example for it to build multiple videos on top of it. So this is basically calling those afars. Uh so we need to do basic setup. So across all

[9:11] the so there's a huge interesting API documentation that hijen has provided and I've set this up all with those API documentation and it's very very simple. Okay. It took me like 10 minutes. Okay. So this this avatar call here uh is going to access the avatar video that I already created within Hen. Correct. And then based on the new script, it's going to use that video to create a new video output uh with me talking. Exactly. Yes, you're right. Correct. So after video, we'll just have the video. We will also take the voices because you would have recorded multiple voices and things like that and we wanted to take voices. So we

[9:54] would call out voices also separately. Okay, good to know. Yeah. So there is something called uh I've added an error tableable node because there's another way that we can do it over here. We can add a mult uh MCP client tool and uh things like that. I did not want to complicate it. What I tried was uh let's say that the claude is creating a script right the MCP is the trigger here. It is not that we can communicate from claude to the uh like let's say for example the trigger is what is enabling the entire communication but if we want to basically make sure that from the claude

[10:37] itself it has to create a video it is not possible. So what we need to do is let's say create a script add that script to air tableable and get the script from air tableable and generate videos on top of it. So that's why I've added an air tableable node over here. Got it. So the next one is like I said it's voices. Again it's pretty much once if we set up the aftar all the other setup is pretty much the same. It's very

[11:05] simple. Okay. And this is the get node of or uh air tableable because we're going to get the data after get the script after we created added the script to air table. So this is the get node. Yep. So you need somewhere to store the script so that Claude can access it and and pull it back in. Exactly. I think it's also better, right? Like for example, if you want to make changes to the script, I would rather not uh uh put everything in claude and then say that hey, if there is some edit that I wanted to do in the script uh where I want to

[11:40] change the introduction, change the name, change certain facts and things like that, I could probably like do that in air tableable and it can get from there. Okay, got it. Cool. Makes sense. And the next one is generate the video. Okay. And you have your uh Okay. You you have an API key and things of that nature in here. Okay. It's pretty much the same which is here it's going to get the afar and voice and the here it's

[12:07] going to generate the video. Okay. So uh this one is basically the status get because when you generate a video it is very important you add this particular step to understand if the video is processing is it complete or is it is the video able to get created or not. So this is going to give us a status check of video is getting processed it's getting created it's successful it failed or things like that. So this will be very helpful for us. Okay cool. So what happens when you uh when you run all this? Yeah. So, I'm going to the cloud. So, this is where I've set this up. Um, so if you see,

[12:47] I've connected my NATM MCP trigger and you can can you get me like the 10 YouTube videos for the marketing automation from there. So, it is going through the first HTTP request node which is going to scrape the videos for that particular keyword and it's giving me all of it, the channel, the URL, when it was published and all of it. If you

[13:06] want to check, you can check from here. Let's say for example this is the video.

[13:18] So you can get the videos. Okay. And from here I've said that will you be able to transcribe this video and if there is no transcription available for that video it'll basically like let me know that there is no transcription. So if I want to transcribe another video, so this is the one that I've

[13:39] transcribed. Mhm. So what is marketing? Is that a separate is that a separate chat? So it's pulling the videos uh in a list. You're able to watch those videos or or whatnot. And then say you want you find one that you know fits the use case or the the script that you want to generate. You pull that one. You you chat it back and you say, "Hey, go ahead

[14:00] and transcribe this for me." Exactly. because we've already added the transcribe node in the MCP trigger. So, it should be able to transcribe. Okay, this is transcribe the video and I've asked it to like, hey, uh, come up with a 60-cond viral script idea based on the above video transcript and see if you can find some gaps. Okay, the lazy

[14:22] marketer secret. I like that. I can be lazy and boring at the same time. So it came up with what was the gap in the original video. There was like you know the emotional appeal, social proof, common mistakes, urgency, all of these things. What's a gap? What you can fix with it and all of that it came up with. Okay. So based on yeah based on that I gave the voice foundation that we have and asked it to like create a 60-cond video around this particular idea on top plus the gaps and it created me a video with the you know the statistics why 91%

[15:02] marketeers are using automation wrong. So, we've got the the the gaps and we've uh identified those and then we have these like ideas for improvements, you know, to keep people engaged throughout the video and include some of those psychological triggers. And then we're creating like a fresh script here, right? We're not just like taking, you know, the the the content from the video that we found. We're integrating all this into a brand new fresh idea with our voice foundation, how we create content on X. And it's kind of mixing all those together uh to to make the end uh result. Exactly. You're right. So it gave me this and then it said that what

[15:43] key elements uh how does the hook look like and how has it like you know created that script on what basis based on the gap and fixes it found. Okay, got it. So now is the beauty. Uh I've added this. So I've asked it to add the script to the edit tableable so that I'll be able to get it back when I want to create video on hijen. So it is also it reported back saying hey this is the title this is the script this is the record ID I've added it and stored it in editable which we can see it here title and script. Okay and if

[16:19] you want to edit it or whatever this is the place to do it. So, a lot of times we recommend having like a human in the loop, not just being totally rellyant oneshotting with AI and letting it go and and create a final product. This is your step to come in, make sure the script sounds good. Maybe you want to add a few different facts or angles or something like that. This is where you

[16:40] do it. Exactly. Yep. Now going back to claw uh I've asked it to take the script from uh air table that is added and ask it to generate the video. Now first step is it's going to pick up the uh like in the trans I mean the in the in workflow we saw it's going to pick up the aar voice and then generate. So it said that great the video has been successfully created. So this is the video ID. Give me the details and it is going to check the status now. Like I said that can you check if the video has been created said that it was processing. It's going to

[17:16] take some time and I asked you to check after 2 minutes. Here we are. So how long did it take to generate the video? Um let's say per video like overall this chat took me 5 to seven minutes. So like I was able to create two videos in 10 minutes. Wow. So, you know, the big the big thing here that I find so interesting is that sure, you can go to Hey Genen or go to whatever tool you want to use and and make a video, but what's happening here is that we're creating like a closed loop process from identifying a topic, uh figuring out what videos are performing for that

[17:55] topic, finding the gaps, uh creating a fresh script based on whatever input we want. uh maybe we you know present in a certain way or have different topics that we want to integrate and then you're just clicking a button. It's passing all that into Hey Genen. You don't have to go in there and prompt and mess around with it. It's using the saved assets that you have already and then it's going ahead and just doing this for you. So it's a completely closed loop from finding an idea to creating a video. Yep. Yep. And the best part. Do you want to see yourself on the video, James? Let's see. Let's see. I

[18:29] mean, I feel like that's probably what everyone's gonna be waiting to do. We have to check this video out. Yeah. We'll be honest, too, with how it turns out. Yep. I'm a little nervous. Yeah. I was actually waiting for this moment how you're going to react. I have no hat on. Yeah. Okay. Can you turn up the volume? Yeah, I'm just going to do that. It's full. So expressive. very expressive. I was like shocked. Yeah. Yeah. I'm going to play

[18:59] from the beginning. Okay. Okay. I can't hear the uh the video uh unfortunately. Oh, okay. Uh do you want me to share the link with you so you can share the link? I'll I'll try to share my screen and see if that comes through here on Riverside. Let's let's try that. Yep. So, I'm going to stop sharing and going to share that link with you. Just a

[19:27] moment. In the chat. Okay.

[19:36] Okay. Okay. I've shared it in the chat with you. Let's see. [Music] Um, can you send it to me on Slack? Yeah, I've I've slagged it. Oh, okay. I was looking at the chat here. Uh, one second. All right. I'm going to share my partners think automation means I'll share. Did you hear that, by the

[20:06] way? I did. I did. Yes. All right. Cool. Here we go. All right. Here's the moment we've been waiting for. Let's get a drum roll here. Okay. Here we go. Most marketers think automation means set it and forget it. They're dead wrong and it's costing them customers. Here's the uncomfortable truth. 91% of businesses using marketing automation are actually damaging their customer relationships. They automate everything, including the parts that should stay human. I've audited 200 plus automation

[20:43] systems. The pattern's always the same. Companies automate their personality right out of existence. They become robots talking to humans and humans hate robots. The fix: Apply my automate process not personality framework. Automate the boring stuff. Scheduling, data entry, follow-ups, but keep human touch points at decision moments, problems, and celebrations. Example, one client automated their welcome sequence but kept personal video check-ins at day 7 and 30. Result: 312% higher retention than full automation. Same time saved, better relationships built. Remember, your customers aren't buying from your software. They're buying from you. Your automation should amplify your humanity, not replace it. Stop trying to remove yourself from your business. Start using

[21:36] automation to multiply your presence. That's how you scale without selling your soul. Ready to fix your automation? I'll show you the exact 7-day framework in the next video. Pretty good. Pretty good. Um, let me stop sharing and we'll, uh, we'll kind of do a a breakdown here. So, what I liked, uh, I like the process. I I like how easy it is to find inspiration, uh, and create a video with with chat. Um, I think that a lot of these models just with a one shot out of the gate, it's not quite there yet. When I see people do uh really good sort of AI avatars and things like that, I think

[22:18] they're usually using a couple different tools like maybe they have runway or or something like that coming in. I think this is a really really interesting start though. Like just the fact that it can shortcut the time to getting this done and getting the script and all that's really cool. Feels a little robotic. I was kind of laughing when the avatar was like humans hate talking to robots or whatever because that's literally what we were doing there. But uh fascinating fascinating stuff. What's your take on it? I I think so too because like you said, you know, the the the robotic part is still there. But for

[22:53] some reason I found some founder talking about so he has actually created multiple videos using AI and he has grown his uh Instagram to 1 million followers in the less than six to 12 months of period and he said the trick was like you said like there's hunen plus there's 11 labs for voice but even in the 11 labs he used to like record all his voice like for example the more data the voice has the more closer it'll look natural. So I think it's all about the data even agent for that matter you anything that you're trying to like you know let's say that there are multiple

[23:30] emotions that you share like happiness sadness or whatever and then like say you've recorded it video and voice for that matter I think that's going to do the trick. Yeah. Interesting. Yeah because I only created like one like short one minute avatar video in Hey Genen. So I think we should try that. I'll get in there. I'll upload more. Let's see if more data helps improve the output and then uh we should start posting those on Instagram. Why not get

[23:58] us to a million followers, you know? Yes. Yes, for sure. Awesome. Cool stuff. Uh we will share kind of an overview of the workflow, I guess, um so that people can, you know, explore it or whatnot and and try it themselves. I'm really curious to see if people can integrate some of these other, you know, aspects into it as well. Like maybe uh you could create a thumbnail based off of the video that is created uh with this MCP

[24:28] integration automatically using Open AI. Um maybe, you know, you could match up the voice a little better or, you know, do some interesting edits with some stock footage. I could really see how this can develop and almost a full-on video assistant, which is super cool. Yep. Yep. Yeah. I think there's runway which we can try for the B- rolls and stuff like that, but yeah, it's

[24:52] interesting. Awesome. Great work, Veru. Uh yeah, hopefully uh we can keep working on it and uh we'll see what other MCPs you have uh later on this week. Yep. Okay. Talk to you soon. See you.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt