- 2 days ago
Category
🦄
CreativityTranscript
00:007 videos, 1.1 million views, and the channel hasn't even posted in a month.
00:05I keep seeing this format pop up, these simple, minimalistic cartoon videos in the health niche,
00:10and this one channel, Body Reset, catches my attention. Not because the content is
00:15groundbreaking, but because the production is so clean and so repeatable that it feels like
00:19a system is behind it. So I'm building one. And I'm not just building the structure,
00:23I'm automating the entire production pipeline. One command in, finished video out. But the
00:29part that surprises me isn't the automation itself. If you've been paying attention to
00:33YouTube lately, the health niche is going through a shift. A bunch of different formats are exploding,
00:38but this one in particular, the flat, minimalistic cartoonist, is gaining traction fast. Body Reset
00:44is running this format with 7 videos and has crossed 1.1 million views. His most recent upload was 4
00:51weeks ago, and it's sitting around 400,000 views. He's pulling over $1,000 a month posting
00:57inconsistently. Now, the interesting thing about this format isn't just the views. It's the
01:01simplicity. The visuals are clean, the voiceovers are straightforward, and the structure repeats.
01:07Which means, if you can systematize the production, you can move fast. And that's where my head goes.
01:12Not, can I make one of these? But, can I make the process of making these almost entirely hands-off?
01:17Because before something like this exists, you're rebuilding everything from scratch every single
01:22video, and that gets old fast. And it's the reason most people in this space make three videos and
01:27disappear. The system we're building in today's video is going to run on Claude Code. You give it
01:33a topic, and it handles everything from there. Script generation, AI voiceover through 11 labs,
01:38video scene generation through key.ai's VEO 3.1, and then FFMPEG stitches the whole thing together
01:46into a final MP4. End to end. One command in, finished video out. Now, I'm walking through this
01:53setup on my MacBook. If you're on Windows, the logic is identical. You just need the right command
01:58prompt commands for your system. I've actually already completed the full Windows build separately,
02:03and that whole process has its own set of problems. But before we get into the setup,
02:07I want to be upfront about something. The first time I tried to build this, I made a mistake that
02:12cost me about two hours. And it's something so small that I almost didn't catch it. However,
02:17the way we're approaching this is, I've already built out an entire setup guide ahead of time.
02:22Every step, every command, every folder structure mapped out. And what this does early on is that
02:28instead of trying to manually follow each step, we'll just hand the whole guide to Claude and let
02:33it walk us through the setup process. So, go ahead and grab the Notion guide from the link in the
02:37description. Because what we are literally going to do is copy the entire guide, paste it into Claude,
02:43and prompt Claude to take us through the setup step by step. And Claude is going to hold our hand
02:48through the whole thing. We'll receive the exact terminal commands, tell us what to verify, and
02:53we'll flag what's missing. Now, of course, don't fully trust this at first. Keep double-checking
02:58everything Claude gives us against the guide. Now, to get right into it, the first thing we will need
03:03to do is verify three paths inside of our terminal. Python, Git, and FFmpeg. All three need to be
03:10installed and updated. So just insert the three commands that Claude gives you into your terminal
03:14or command prompt. For me, they're already good from a previous project. But if you haven't set
03:19these paths up before, Claude walks you through installing each one. That part is basically plug
03:24and play. Once that checks out, we'll move into building the folder structure. And this is where I
03:29make a decision that will save us a lot of headaches later. We are going to do everything
03:33through the terminal. No dragging files around in Finder, no manual folder creation. Just terminal
03:39commands, one after another. It sounds small, but that choice matters more than you'll expect,
03:44because most people are just guessing their way through this part, switching between Finder and
03:49terminal, and that's where things start silently breaking. So first things first is creating the
03:54folder structure. Grab the command and paste it into the terminal. From there, we need to create the
03:59env text file. Insert the command that Claude gives you for creating the end file. Once that is in place,
04:04the next phase is gathering API keys and inserting them into the end file. Three of them, 11 labs for
04:11voiceover, key.ai for video generation, and anthropic for Claude's API. For 11 labs, we'll head to the
04:18developer section, create a new API key, and give it access to all these features. Then we will need to
04:24pick a voice. But before deciding, we'll go back to the body reset channel and listen to the
04:28voice they're using. American accent, clean delivery, conversational tone. So with that,
04:34we'll filter 11 labs by English, American accent, and start previewing. I went through probably 15
04:39voices before landing on one called Christopher. Not an exact match, but the vibe is right. Similar
04:45delivery, similar energy. And at this stage, vibe matters more than perfection. You can always swap
04:51voices later. The voice ID is just copy-paste. Click the three-dot menu and select voice ID,
04:57and drop that into the terminal underneath the 11 labs API key. Key.ai is straightforward. Go to API
05:04keys, create a key, copy it, insert it. But the anthropic API key, this one has a catch that gets
05:10me more times than I want to admit. Once you generate that key and close the window, it's gone.
05:16You can't see it again. You'd have to regenerate a new one. So the move is to paste it into
05:20the terminal
05:21the second it appears. I know that sounds obvious, but I'm telling you, I've regenerated that key at
05:26least three times because I got distracted at exactly the wrong moment. Now, once you have inserted
05:32your API keys, make sure you go through the process of installing the dependencies. Just insert the
05:38output you got from the terminal after saving your API keys, and Claude will give you the direct
05:43command prompts to insert into the terminal. Just do exactly what I'm doing here. Now, here's where
05:48things get interesting. Claude generates five core files for the pipeline, pipeline.py, voiceover.py,
05:55scene underscore generator.py, video underscore stitcher.py, and logger.py. Each one handles a
06:02specific stage of the production. But before we have Claude build those files, we'll do something
06:07that ends up being critical. So I've already built out a full prompting workflow based on our competitor
06:13from ideation to scripting to text to video prompts, all optimized specifically for
06:18VEO 3.1 and aligned with our competitor, Body Reset. So I'm pasting that entire workflow
06:24into Claude first and telling it to align the skill files with the prompts before doing anything else.
06:29This is the part most people skip. They let Claude generate generic files and then wonder why the
06:34output looks off. The visual style, the motion directives, the formatting of the scene prompts,
06:40all of that needs to be baked into the pipeline from the start. So you can locate that prompting
06:44workflow that I just used and also built at the bottom of the Notion guide. Make sure to paste
06:49the prompts in just like I did. Once the files are generated, I use the terminal to create and paste
06:54each one into the Body Reset folder. Nano Editor, Paste, Control X, Y, Enter. Five times. There's a moment
07:02where I paste in one of the longer files and everything just disappears off screen. My stomach drops. I think
07:08I've somehow wiped the file. Turns out the string is just pushing everything out of view. Nothing is
07:13actually lost. And that brings me to the mistake I keep mentioning. On my first attempt building this
07:18on my other device, I used a here.command to write one of the files. And it doesn't fully replace
07:23what's already there. It just appends. So now I've got duplicate function definitions buried in the
07:29middle of a script and the whole pipeline breaks with errors that make absolutely no sense on the
07:34surface. The fix is simple. Verify the files after every writing them, like you see me doing. Don't run
07:40into the same issue I did, being unaware of what's wrong until it's already cost you two hours of going
07:45in circles. And that's the thing about building systems. The infrastructure mistakes don't look
07:50like mistakes. They look like the tool is broken. And the whole time, it's a file that didn't save
07:54correctly. After you verified that the five core files are in place, prompt Claude to build the
08:00skill prompt. This is basically the instruction set that tells Claude code how to orchestrate the entire
08:05pipeline, what to generate, in what order, and how to handle each stage. This one is long, significantly
08:11longer than the other files. And pasting it into nano requires some patience. I actually mess up the
08:17first paste and have to back out and redo it. But once it's in and saved, the last step is
08:22the
08:22settings file and installing dependencies. Both take about 30 seconds. And then, the system is ready.
08:28Everything installed, every file in place, every API key connected. The whole setup takes roughly 15 minutes.
08:35And most of that is just copying and pasting. But the big question is, is it going to work? Or
08:40will
08:40we have to burn hours debugging the setup? So this next part is going to be huge. So make sure
08:46to stay
08:46locked while going through it. Now, real quick before we run this Claude code system. If you're watching
08:51this and you're thinking, I don't want to debug my way through this setup, I get it. I've already spent
08:56the hours doing that. The trial and error, the file write bugs, the API key issues. I've already been
09:02through all of it. So what I did is I took this finished workflow, every file, every prompt, the
09:07full folder structure, and packaged it into a zip file for both Mac and Windows. The members inside
09:13Creator OS are already getting access to it right now. The community managers are walking people
09:18through the install. And most are up and running in about 10 minutes flat. No debugging, no guessing,
09:24no two hour detours. I'm walking you through the build because I think understanding the system
09:28matters. But if you just want to plug this in and start generating, first link in the description.
09:33Come inside, grab the workflow and the rest of our tools, and the team will make sure you're set
09:38up. Either way, let's see if this thing actually runs. To run the pipeline, we'll open the Claude
09:43desktop app. Quick little detour just to save us some headache. Go into settings and go to capabilities.
09:49Scroll down to skills and make sure you point the folder to the body reset automation folder.
09:54Once that's done, go back to the Claude code chat and then type in the skill like you see doing,
09:59and insert any health topic of your choosing. I'll be going with this topic. What happens to
10:04belly fat when you walk fasted? And then I just, watch. Claude generates the script outline,
10:10kicks off the voiceover generation through 11 labs, starts submitting scenes to key.ai one by one.
10:16The pipeline estimates about 12 to 15 minutes total, and it's running both the search and monitor
10:22tools simultaneously. I don't touch anything. I'm just sitting here while it works through each stage,
10:27and I bet you're experiencing this weird feeling. Because you've spent some time and effort building
10:32and probably debugging this system, and you're just there waiting for something to break. It's the same
10:37for me. I'm expecting an error to pop up, some API call to fail, some file to not save right.
10:43But it just
10:43keeps going, and then it finishes. 10 scenes, all generated, all stitched together with synced audio.
10:50The voiceover plays clean, the visuals match the prompts, and the final MP4 exports to the output
10:56folder exactly where it's supposed to. We've built a system that now stops us from feeling like we're
11:01making a video, to start feeling like we're running a system. That's the difference. That's what most
11:06people don't get to. They stay stuck in production mode, manually doing every piece, every time,
11:12and they never cross over to the point where the system does the producing, and they just operate it.
11:16Here's the thing about this setup that I think most people miss. The hard part isn't running
11:21the pipeline. The hard part is building the infrastructure once so that every run after
11:26is just a topic alongside a command. If you want to make additions to our current pipeline,
11:31you absolutely can. Want to add a zoom effect? You tell Claude to add it to the pipeline.
11:35Want to change the visual style? Update the style prefix in the scene generator. Want to scale to 6 or
11:419
11:41minute videos with 90 plus clips? Change one parameter. The system is modular. And because
11:46it runs through Claude code, improving it is just a conversation. You're not editing code by hand.
11:52You're telling it what you want different, and it adjusts. And at a certain point, it stops feeling
11:57like making videos. It starts feeling like running infrastructure. That's not a small shift. That's
12:02the difference between someone who posts for 3 weeks and burns out, and someone who's still publishing
12:066 months from now. Because the system is doing the heavy lifting, and they're just steering it.
12:11The finished workflow. Every file, every prompt, both Mac and Windows, already lives inside CreatorOS.
12:18I didn't build CreatorOS to just sell a course. I built it because I keep building systems like this
12:24one, and I needed somewhere to organize them, test them, and share what's actually working with people
12:28who are truly locked in and are actually building. That's what it is. That's where this lives.
12:33First link in the description to get access. But either way, the format is real, the system works,
12:39and the health niche right now is wide open for anyone willing to build the infrastructure
12:43instead of just chasing the content.
12:45Thanks for watching. See you in the next one.
Comments