How I Automate Content Creation with Antigravity (claude skills)

Today I'm building a workflow that repurposes my own content into tweets, carousels, community posts, so I'm not stuck behind the desk posting all day like a loser. Posting is hard, and to have any chance at visibility, you need to post like all the time, so anything you can do to remove some of that friction counts as a huge dub. If you're new here and we haven't met, we've been using anti-gravity and codecs to build out solutions to my real problems, starting with subscription services, so I can get actual control of my pipeline and I can stop paying sas-tac. I'm crossing those suckers off the list, and this content multiplier is the next one. I am doing this end-to-end, I haven't built this out for myself before yet, So you get to see the entire thing without any sort of gatekeeping You can follow along or if you'd like you can click the link in the description where I'll leave the finished product there You can download if you like that. Let's dig in So I had something like this working for me in the past where I had an end and end workflow that would watch my newsletters pick really good articles repurpose them in my voice and then send it to Placid to Punch out a PDF for LinkedIn or whatever and that's cool. But like that's not what I want to do anymore I don't really want to just like copy paste things and pass information along it was great for visibility, but I really want to leverage my voice and my ideas in a way that only AI can do. So I've seen this before. I have a baseline idea of what it is we want to do, but even still, we're going to wireframe it here in Opus first to help with rate limiting and anti-gravity. But first things first, we get our folder set up in anti-gravity. I already got it here and fire it up like that. The only bit of legwork you need to do is you need to have a template in advance that we can kind of paint over. And I do recommend getting like an intro, like a hook page, a body page and a CTA page. But for now, I'm just going to do the one for the sake of this video. Moving forward, I'm definitely going to have more than one template. So the vision I have for this workflow is, well, let's just talk it out. I stumbled over that while I was thinking this through. So I'm going to articulate it in one shot again. So I want to build a workflow that repurposes my video content into other formats, ideally carousels, community posts, tweets, and the like. The idea is we put a source video into a folder. It's processed with the transcript. We take out three to five insights depending on the video length. Carousel templates are then overlaid with the insight text via a pillow library or something of that sort. Processed videos are moved to a done folder. Output is moved to the output folder with subdirectories. I would like the script to run on an exe so that I can click and make it easier for myself each output isolated in their own sub directories video in carousel out okay writing the core modules thing i love about opus is it doesn't just like wireframe the thing it's like oh here's a sample python script for it so even if you use weaker models you can have it just execute the task because i really do trust opus thinking for the actual coding part of things all right it's starting to become like the de facto workflow if you watched any of my other videos this is like how i do it let me zoom out here though cost per open ai api i do not want i'm gonna ask this thing is it possible to use my gpt oauth with my gpt subscription or could we maybe do something like that short answer no i'm gonna try a gemini 3 with the gemini 3 api key again it rate limited on me yesterday but we'll go with this switch model to save 10 times um okay so let's get this rolling copy and go over to anti-gravity we've got this here cool here's a project wireframed by another coding assistant i'd like to build this please the only thing i'd like to look at is potentially the llm used we can maybe change it to open router or let's try the gemini api key please instead just build everything else out and we'll worry about that later find the png carousel template in the context 10x root directory let's try like that doing my best paste that's gonna take a couple minutes i had some bad luck yesterday i was uh doing this thing with um what the fuck was it yesterday clipping and i tried a couple times to get gemini api to work and it just like rate limited off the top and then the other library busted on me i had to do like a bunch of live debugging now i have a great picture cool it's like i agree let me create the implementation plan the key decisions are which api key carousel renderer to match the png i need to tell it um we need to specify the text box window versus the uh versus each template is there shoot i should have included that is there a way to do this programmatically and then we'll screenshot this guy and just show them what we're working with programmatically because this or you know why am i typing this this particular template is mocking a twitter post so it has to go in a certain space i'm really trying to lean into like the comic book aesthetic now because everybody's doing the cyberpunk thing like even the default uh ui skills are all like let's do neon pink and teal or like orange like danger colors i'd rather go like ultimate spider-man if you if you know you know kind of into like the i mean i i grew up with the ninja turtles right that's kind of like the vibe i sort of like batman golden era comics you know great idea we need to make sure that we're typing in the same i can download a font for this too actually you know what let's look up uh look up some six fonts you know what what the hell let's just ask this guy what would be an excellent font for a what's it called neo brutalist oh the shortlist i love okay this is the neo brutalist font that's what it says whoa looks looks like it hey actually that looks download family all right okay i'm gonna install these guys i did install these system wide because i kind of like them but i'm gonna put them here too i want to make sure that this is actually in yeah do it okay so let's just copy this copy paste and then move this into the fonts folder move uh get the other one too because why not no you know what let's let's slow the roll here a little bit run setup.bat installs gemini pillow etc here i'll just say run it run it you do it i don't want to do it gonna tell it fonts are here we have a font here other fonts to follow but we'll use this for now put her in there this is running a thing okay dependencies are already installed i knew that uh if there's a gemini api key it should be good after yesterday my rate limit is probably reset who knows i'm not ah it's gonna do it again though i am so sure oh it's gonna pull in my api key from other projects what how is this possible all right well in that case try gemini then fall back to open router please you need the folders though too right because the idea here is open router api key gemini cool i'm gonna do this on the desktop i'm gonna call you atomizer can you guys see this okay so what i'm doing is i created a folder on my desktop and now i'm just doing the structure i want so this will be oops let's just rename it yeah inbox output and this one's gonna be done all right so this is the what i want is to just like drop things in here and then it'll have the output folders and each carousel in its own like subdirectory and then when it's complete it'll move the source file into done right makes it easy no big deal um this will be automated at first but i find the best way is to just make sure that it's working first manually so once this is tweaked and like dialed in i have a couple templates that i really like i will have everything that moves into the output folder just automatically posted to socials and then all i have to do is run it through my video editor workflow it'll move into here and then it'll just like start to really flow for me you know all i have to do is like turn the camera on actually just like talk to you guys and then we'll be good to go also if we're going to be using open router don't use open ai let's use a smarter model like sonnet 4.6 please honestly i think actually for this sort of thing grok is probably a contender oh does grok grok has a free tier yeah but it's not the nice model let's see what see what research has to say uh let's talk to comet new manis skills what was that what was it looking at it's like watch your browser history okay um all good for social media posts if i already have an idea i just repurpose or what would be the best writing model part of me wants to say grok is that accurate okay so grok repurposing speed high social media fit excellent yeah but what's the consistency medium oh for your repurposing workflow turning ideas into bite-sized social posts start with grok i love that it remembers me makes life easier let's say oh man perplexity if you're not aware is an actual game changer and i hate using that word but it's so much better than google because it like applies everything to your context like it didn't just say uh we don't know like grok's okay use claude but like it already knew that i was uh already knew that i was like what i was doing like because i've done this in the past i've kind of planned things out before oh okay it's okay to use a short I guess for the test workflow but it's kind of like finding okay so that is in a completely different directory the video it's pulling from I didn't give it any videos to digest it looked for them in the clipping output that I had before I really don't want you to be messing with other successful workflows i can give you the directory if you need copied not moved good non-destructive gemini would have rewritten everything and been like oh yeah i know it's a different workflow but i decided to help you out i guess uh this is actually a really good look into how ai works like we can't rationalize ai thought or ai processes and logic because it's non-human thinking right so this is trying to find the solution to its problem and it's like going in and doing stuff and that's fine in this instance but like can you imagine if you're seeing i didn't tell it to do any of this and it's already like going and doing that things like open claw are a problem in that regard because the ai has a mathematical sort of thing that they're trying to solve for and they'll do it at all costs and it means their perception of your desired outcome is fluid as well that's a lot of words i don't like it when ai does shit on its own um i actually have a guardrail like a system prompt where it's like you only get to do specifically what we talked about and i can be like more granular rather than like go get it oh what's it look like that's not good uh except i did give it that font so i do need i need to figure out what the font is for twitter for that i'm gonna go back to my monster at i fucked up otherwise not so bad though hey this isn't awful if i change this template okay that's actually really good for this particular template can we change the font to monster at do i have it and that's gonna fucking spell that wrong for sure what is it that one ai just negotiated 4200 off a car that wasn't a demo it was real i love that it's got like this in the sub it really fucking understood the assignment you know what we do from here is we give it a different template oh my god okay so not in windows fonts let me download it i did give this network level access i'm actually down with this i think this is excellent with the numbered cards and the heading and subtext however for the can we save this first of all this is great i want to use this moving forward but for this particular template i just want one text block to emulate a twitter post single block of text like an actual tweet post no slide numbers no head oh so good so good so good so good and the community post insight one ai carpet bombed 500 contacts with unsolicited text you are handing god to your permissions to code you don't understand actually that's not too this is really good legitimately video this is my clipping workflow oh my god we're not ready open claw is some black magic sorcery that is dope as hell but it's becoming there you go edited captioned i'm actually stoked about that one too plan approved let me build it so good so good breaks it down into tweets fuck sonnet is okay i'm not one for like you know hyping up models and stuff like that but the difference between sonnet 4.6 and like what i was using last year to to do just this is insane i didn't have to tell like it already has so much information on the goal i um could have to do with you know the the plan that we laid out but like breaking these down into threads it already knows the format it knows enough about every bit of what we're doing that i don't have to give it as much instruction which is obviously really cool i know i just talked about how i don't like ai doing stuff on their own uh i take it back um some of this is dope just like don't let it have access to stuff that could do damage you know what i mean like the worst that could happen here is like i oh shoot i just gotta like start over or we erase some stuff or go back or it makes a bunch of api calls like i'm not worried about that i mean i gotta put like another 10 bucks on openrouter no big deal this is just like tweaks to the template and the carousel aesthetic but the workflow itself is running fine so as long as you have your templates broken down into the different folders that's cool and i'm gonna work on pipeline next we actually i'm gonna tell it right now okay this is great the rest of this seems like we're just tweaking aesthetics and different carousel looks which we can do forever what I want to iron out now is actually the pipeline and making sure that the text box calculation is baked into the script I want to make sure we know hmm what else we need to know also ideally something to think about later on down the line is randomizing which template to use based on what's in the template folder if that makes sense and and finally a exe command or something that i can double click and have this run automatically based on what's in the inbox all right moment of truth hormosy style that's good enough probably do bold but fuck whatever i would post this would you post this it's really like buffalo bill okay so randomization based on insights right so let's say for example that last run we had three insights if i had three templates i would prefer they'd be different each time potentially not deterministic just like random so if i get two with the same template i don't even mind that and also with the style dot y a m l is that like a calculation dependent on the text box or free space like will we have to have a literal text box like we have here with the Twitter template like explain how this works to me crackers okay the build needs updating it references this but doesn't include the templates and fonts directories which reminds me we need to standardize fonts for each template there should be like a config for each template and then that way it makes it easy for whatever model to just like read that first and then know where the proper text box should go and what font to use that makes sense to me it's thinking my voice transcription okay so randomization for insight I'm right I Love hearing that how text box work right now. It's manual pixel coordination set okay for X templates and manners For plain background templates you can use the full canvas Yeah, that's what I want is the helper script okay? which reminds me and the helper script that maybe runs when we add a template generating all needed dependencies and measurement i'm talking too much all right cool let's go for it edited go go this is really coming together so i'm sure you see the vision where i record a video i put it into my clipping software the clipping software takes the done i guess package and moves it over to this and then i have all of my posting assets in one place i review and then i pump it out to social media the next thing i'm going to do is a lot of people use bloatato and that's awesome bloatato's bloatato's great but i'm just going to do something locally where i can do the same sort of thing and then just click and it'll schedule and pump my stuff out great success today nice um i'm probably gonna spend a bunch of time like adding templates but like once you once you do this once you can keep building your template carousel library like i might want one oh dude how did we get that that is great i might want to like move that up okay you guys seeing this it's like the preview yeah you're seeing this okay now that's the preview red rectangle shows exactly where text lands here's the full workflow now are we doing this with ocr how are we doing this okay per insight randomization excellent excellent to add a new template honestly the way it's laid out right now is just fine you've got it in the content engine but the exe file i want to be able to move to my desktop or something so that whatever is in the inbox can be processed and i don't have to worry about it now let's close this up this is cool how it finds this stuff okay so what it's doing right now is it's just building out an exe file the idea here is it's like turning this into a process an application that i can just double click it'll run in the background from wherever it is so like i could even drag this you guys can't see where my house is but i could even drag this down to my start tray down here and then once the pipeline's all done i just have to like click a button and it'll it'll run the whole thing we're build this out to include like thumbnails i've already got some youtube like trend detectors that i am conspicuously ignoring ignoring the trend detector is just doing my thing and showing you it says it's taking too long because it's pulling in really heavy python libraries we're not gonna do that because we don't need it just gonna pull in the ones we need why do we need open ai for the exe file just out of curiosity oh okay so it's just using it to talk to open router which is fine by me building desktop exe let me test it and then also try a proper exe build with the i suppose that's not the end of the world um even if i just get like a command i can run in the terminal i don't want that that's not what i asked for but for now while it's building out before i'm automating the whole thing because don't get me wrong that is the goal we need to keep it going manually for now until we dial it in perfectly and it's consistent and then we automate it on de okay so it says hey let me roll this over so you guys can see double boop and my comment browser will close so let me open this back up there you are okay so i don't know if you guys can see but we've got this here and this is the idea i want to like click a button perfect install yeah man that's cool i don't mind if it's that's cool i didn't need a .exe file i just wanted the functionality to be there where i could just click the button rather than going into anti-gravity or running it in the terminal and uh from anywhere no matter where this button lives it runs the tasks from the content engine and anything that's in the inbox gets processed full summary of what we built per insight template randomization add template p oh my god yes okay so check this out it's got an onboarding for new templates so anytime you add a new template you just run this script and it'll track the okay I can't find that anymore where was it the preview preview preview preview it'll track where the text box goes and pick a font or you can upload a font and it'll go from there I've got my desktop launcher this is it I am legitimately excited for this uh if you were following along awesome but like i said at the outset i will leave a link to the full product down below just remember to onboard your templates there'll be instructions and all that stuff in the link in the description make sure to sub for more videos just like this and we'll catch you in the next one

📚 Skool Community https://www.skool.com/easymachineai
🤖 AI Strategy Call - Let's Work Together: eric@easymachineai.com

Stop playing photocopier with AI Social Media, multiply your own voice as a content. If you want to grow your online visibility without spending hours chained to your desk copying and pasting tweets or formatting carousels, you need to build your own automation.

We don't rent broken tools. We build our own machines.

In this 100% LIVE tutorial, we build a custom, local Python pipeline for complete social media automation, with Google Antigravity. Watch as we create a workflow that takes your source video, extracts the best insights using Claude Sonnet 4.6 and the Gemini, and automatically stacks them into PNG carousels, Twitter threads, and community posts. 

No copy-pasting. No monthly subscriptions. Just a custom AI coding agent packaged into a one-click automated .exe file that runs locally right on your desktop.

Chapters:
0:00 - The problem with paying the "SaaS tax" for social media tools
2:15 - Wireframing an automated AI content multiplier workflow
4:30 - Using Antigravity and Opus to code the Python pipeline
7:45 - Dealing with Gemini API rate limits and model switching
10:20 - Generating neo-brutalist PNG carousels with Python and Pillow
13:15 - Standardizing fonts and calculating text boxes for templates
16:40 - Why AI coding agents doing things "on their own" is dangerous
18:50 - Switching to Claude Sonnet 4.6 for intelligent text extraction
21:30 - Packaging the Python script into a one-click desktop .exe
24:15 - Testing the final social media automation tool

================================

Why watch? If this is your first time here—I'm Eric! I run Easy Machine AI and built my AI automation business to a consistent five figures monthly after learning some expensive lessons along the way. Real entrepreneurship isn't about shortcuts or magic tools—it's about understanding business fundamentals and being willing to do the uncomfortable work upfront to become truly self-reliant.