19:03JOEY · Tutorial
These New Claude Skills Are Saving Me Hundreds
A 19-minute deep-dive into the updated AI cinema pipeline — three changes that killed the plastic look for good.
May 27thThe argument in one line.
Building repeatable Claude skills for character and scene generation produces better AI videos at lower cost than one-off prompting, and the system itself matters more than any single output.
Read if. Skip if.
READ IF YOU ARE…
- A content creator with 6+ months of Claude or ChatGPT experience who wants to build repeatable AI video workflows instead of one-off prompts.
- A solo creator or small team running low on API credits who needs systematic ways to generate consistent multi-shot videos with locked characters.
- Someone building narrative video content (music videos, shorts, scenes) who's frustrated that every AI generation requires starting from scratch in a new chat.
SKIP IF…
- You're looking for a beginner's guide to prompting — this assumes you already know how to work with Claude and focuses on systems-level thinking.
- You work in film, animation, or VFX professionally — this targets creators with limited budgets and credits, not production-scale workflows.
The full version, fast.
Repeatable systems beat one-off AI flex posts, and the path to consistent AI video is owning the prompting pipeline underneath the tools. The creator built two free Claude skills out of an actual two-week music video production: a banana pro director skill that writes image prompts, character sheets, outfit references, and scene plates using real camera and skin-detail language, and a cinema world builder skill that handles video prompts across five cinema modes, each with its own lens stack and color grade. Use them by building a character sheet first, then a scene, placing the character inside it, and uploading that reference image back into Claude to direct single or multi-shot sequences while tracking per-scene seconds to control credit spend.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:04
01 · Cold open
AI music video teaser shot — a white-haired girl in a G-Wagon, cinematic and clean.
00:04 – 02:29
02 · The gap and the promise
Joey identifies the two camps of AI video content (flex posts vs. courses) and promises to give his system away free in 10 minutes.
02:29 – 04:00
03 · CTRL: Run Up the World (music video)
Full embedded playback of the AI-generated K-pop music video Joey built with the pipeline: dressing rooms, concert stages, cyberpunk environments, AI characters with speaking roles.
04:00 – 05:12
04 · Post-video debrief
Joey clarifies that CTRL is not a brand exercise or merch play — he built it to see if he could, and the pipeline that emerged is the real deliverable.
05:12 – 07:23
05 · The two skills
Banana Pro Director (image prompts, character sheets, hyper real stacks) and Cinema World Builder (video prompts, 5 cinema modes, lens stacks, color grades). Brief Claude UI demo shown.
07:23 – 08:07
06 · Why free
Joey argues that giving the skills away protects his time (no course to run), aligns with his taste (users will build their own thing, not clone CTRL), and is simply the right move.
08:07 – 09:10
07 · Mira delivers the actual tutorial
AI character Mira takes over and explains the pipeline step by step: character sheet, scene reference, Claude prompt, multi-shot or single take, credit cost estimate. Joey forgot this part.
Lines worth screenshotting.
- Most AI video content online is either one-off flex posts or courses about screenshots — nobody is teaching the repeatable system underneath.
- The pipeline is the actual product; the music video is just a demo of that pipeline being used.
- Claude skills beat one-off prompts because every new chat forces you to rebuild context from scratch.
- Building two specialized Claude skills — one for image prompts, one for video cinematography — is what compressed a full music video production into two weeks.
- If your prompts keep failing, the problem is the absence of a system, not your ability to describe what you want.
- A cinema-language skill that knows camera, lens stack, and color grading vocabulary produces generations a generic prompt never will.
- Giving away the system freely is partly altruism and partly self-protection — running a course would actively make life worse.
- Skills trained on your actual production taste produce your work; the same skills in someone else's hands produce something entirely different.
- Tracking credit cost per scene by shot length is a practical discipline that separates amateur AI video makers from serious ones.
- Building a fictional world as a creative experiment — with no Patreon, no merch, no community play — is a legitimate reason to make something.
- Most AI video tutorials online teach the output of a process, not the repeatable process itself.
- A character sheet skill that uses real camera language and hyper-real stacking parameters does more work than any single prompt can.
- The goal of sharing a system is seeing what other people build with it — not replicating what you built.
Ship the artifact. Explain the system. Give away the tools.
Build-in-public format to steal
The video is the demo; the pipeline is the product — and giving the pipeline away builds more trust than selling it.
- Make the thing first. The music video gave Joey something real to point at. The tutorial only works because the artifact exists.
- Use output as proof of system. Do not lead with the system — lead with what it produced, then reveal the system underneath.
- The AI character handoff (Mira takes over for the tutorial) is a format innovation worth testing: let an AI persona deliver the practical how-to section.
- Frame the gap before the solution. Two camps of AI content = one clean positioning sentence that disqualifies both competitors.
- Give away the tools to protect your time. Anti-course positioning lands because Joey explains the self-interest: no course means no support overhead.
- Credit cost framing is an underused hook for AI tool videos — telling users how many seconds each scene will take is a practical pain point nobody else addresses.
Terms worth knowing.
- Claude skill
- A Markdown file that packages a set of specialized instructions for Claude Code, allowing it to perform a specific task consistently without re-prompting each time.
- Higgsfield
- An AI video generation platform that produces cinematic AI video from prompts, used here to generate dance and performance footage for a music video.
- multi-shot video
- A video composed of multiple separate AI-generated clips edited together to tell a continuous story, as opposed to a single unbroken generation.
- character locking
- Maintaining visual consistency of a character's appearance — face, clothing, body — across multiple separately generated AI image or video clips.
- Higgsfield Canvas
- An advanced Higgsfield feature for building complex multi-scene AI video projects with granular control over characters, environments, and camera movement.
- credits (AI generation credits)
- The consumable currency used by AI image and video platforms to charge for each generation — limited credits require efficient prompting to avoid running out quickly.
- LLM (Large Language Model)
- An AI model trained on large amounts of text data that can understand and generate natural language — used here to translate creative direction into structured image and video prompts.
- prompt engineering
- The practice of crafting precise natural-language instructions to reliably get a desired output from an AI model.
Things they pointed at.
05:12toolSeedance 2.0
02:29linkCTRL: Run Up the World (music video)
Lines you could clip.
04:04
“The pipeline is the actual product. The video is just a demo showcasing it being used.”
01:23
“Every tool is a fresh argument. Every chat is a new thing. What you need is a system.”
07:39
“Selling you a PDF would actually actively make my life worse.”
08:08
“Joey left out the most important part. As per usual, he shared the skills but not how to use them. Classic.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogystory
I'm about to do something no one in the AI space is doing right now. I spent the last two weeks building Claude skills to help us prompt dance, nano banana pro, everything in the Higgs field to waste less credits and make better content.
And in about ten minutes or less because I don't want my editor to spend so much time editing Thank you. I'm going to give all of it away for free to you, the Internet, whoever's watching right now. And I wanna be upfront as to why.
It might be because I hate money. It also might be because I'm a bit unwell. And honestly, it's probably both.
Okay. Real quick. Joey's gonna go on and on, but stay till the end because I'll be sharing the real tips.
Okay. Bye for now. Here's what I mean by no one is doing this.
You look at what's online around AI image and video right now. Most of it falls into two camps. Camp one, somebody made one cool image.
They post it. Look what AI can do. Cool.
You have camp two and you have somebody selling a course about how to make those screenshots, master mid journey in thirty minutes, unlock the hicks field cheat code.
I've scoured YouTube entirely trying to find videos that actually give helpful advice on how to build scenes, build characters, put the characters in those scenes, and then build multi shot videos, creative performances, keeping the characters locked.
Nobody is showing you the system the actual repeatable thing underneath. The reasons your prompts keep failing you isn't because you're bad at prompting or isn't because you're bad at describing to Claude or ChatGPT or your LLM how to prompt for you. .
Every chat is a new thing. Trying to have Claude remember stuff, it's difficult.
What you need is a system. Now, Higgsfield released something called Canvas, but that is way too advanced and not what we're trying to get into right now.
I'll do another video on Canvas once I learn it a bit more and how to help people understand it a little bit more because it's super specific and very descriptive. And I'm really excited about building out more scenes and videos and full length feature films with Canvas. Main thing I want you to take away from this is we are the common folk.
We don't have 10,000,000 credits, and I want to help us build cool things with less generations because you don't wanna be doing this all day and getting this all day.
You know what I mean? Let's dive into it.
Imagine Joey's actually upset when he walks in here. Yeah. He's all like fifteen minutes.
Those shriekers kept you down there fifteen minutes blah blah blah. Actually, seventeen minutes, but who's counting? You girls have about thirty seconds to hit the stage.
Come on girls. The fans are waiting.
Mira, have you told them yet?
Hey. They're big girls. They can handle it.
If the Shriekers already hit the mainframe, they'll be three times as big, and I I Let them do their thing, Kay. You need my meds more than I do. Sheesh.
Now who's in the building?
Hi. They're just doing some last touch ups on me right now. We're filming the big jump scene next, so I'm honestly kinda kinda nervous.
Oh my god. They're crazy.
Duh. That was almost it made me drop my croissant. I was like, no.
Hey. Come in. Get in here.
We just filmed the dumbest scene at Shreeper pink gum or whatever gets everywhere.
So what we just watched, that is a music video. It's my music video that I built for this group called Control that I've also built.
It's a real video. You can go watch it. Don't do that now.
Do it after this. I built every scene, every prompt, every character build, every outfit change, every sequence, um, and I built the entire pipeline to build it in about two weeks.
And the pipeline is the actual product. The video is just a demo showcasing it being used. I'm gonna say this is not great.
Okay. I like it. I like it.
I liked it enough to post it, but that was also a few days ago. I'm always getting better and I'm always learning more. There's obviously things that I want to I'm a perfectionist, whatever.
Uh, the control girls soul mirrors are a day. I'm building a whole world for them just to as I continue to learn and build with AI. All of this is for pure entertainment purposes.
What I can do pushing the boundaries. Uh, I do wanna be clear about something. Control is not a brand exercise.
It's not a content marketing strategy. There's no Patreon. There's no merch coming.
I'm not trying to build a community around it. Although if that happens My man.
I'm not going to turn it down if there are a ton of people that love AI, love the video direction of k pop girls being in the world of AI, and they're all about it. Awesome. We'll we'll continue doing it.
But I made it because I wanted to see if I could. The skills I'm about to show you come out of trying to make it. The video came from the skills.
So here's what I built. I built two skills. First skill is the banana pro director skill.
Second skill is the cinema world builder skill. First skill, image prompts, character sheets, outfit references, scene plates.
It writes them in my voice. It knows what hyper real stacks are. It knows what kind of skin pores look good, what doesn't look good.
It uses real information, real camera language that we'll be using in video production. The cinema world builder skill, video prompts, five cinema modes, each one its own camera, its own lens stack, its own grade.
I can say shoot this scene in a parking lot at night with anamorphic flares and it knows the camera language, it knows the color grading, it knows everything. The technical layer underneath all of it, that's what we've built.
That's what helps build the prompts to give you really good, really real generations. So now the part where I lose my mind, if I haven't already.
Leaving you with these two skills for free because if you can dream anything, you can build anything. Take them, use them, modify them, build your own off of them. I don't care.
That's actually a good point. If you want to take these and build off of them, do that. But here's the thing.
The reason this stuff is good is because it came out of my actual production pipeline, my actual taste. If you take my skills and use them, you're not going to suddenly make the entire control world that I'm building. You're going to make your own thing, and that's what I'm excited about, and that's what I wanna see.
Candidly, also, I just don't wanna run a course. I I want to make cool movies. I wanna make cool films.
Selling you a PDF would actually actively make my life worse. So this is partially all truism and partially me protecting my own time, and both can be true.
If you build something with the skills, send it to me. I'd love to see it. That's the actual experiment.
So that's it. I hope I don't get any hate on this video. But if I do, you can leave a comment because it helps engagement.
And then if you loved it, also leave a comment. It helps engagement. Thank you for a 100 subscribers.
I went from, like, I think 40 subscribers to a 100 in the span of two weeks or a week and a half. If a 100 turns into 200 by next week, that's just what happens.
Alright. I'm gonna give even more because I like to give back. Alright.
This is a long video. Did I write anything else on this script? I have to figure out this week what an alien sounds like when it gets hit by a stick microphone.
That is a real problem I have to solve. So peace, love, and AI.
Hi. Hi. Real quick, Joey left out the most important part.
As per usual, he shared the skills and the pipeline, but not how to use them. Classic. So pipeline, you wanna make a video, you need a character and a scene.
Then you build the character sheet, the skill will help you do that. Next, the scene and placing your character or characters in that scene. Now you got your reference image.
Upload that image to Claude and it will help you build the scene you want. If it's a multi shot, tell it. If it's one take, tell it.
It should also tell you how many seconds roughly each scene will take. Important for credit costs. Okay.
That's all for me. If you have any questions, ask for one of us, not Joey. He's so scattered.
Don't forget to sub and like and leave a comment. We read every single one. See you in the next love.
Bye.
The Hook
The bait, then the rug-pull.
Joey opens with a market diagnosis: most AI video content is either a one-off flex post or a course selling screenshots. Nobody is teaching the repeatable system underneath. So he built one, used it to make a real music video, and is giving the whole thing away.
Named ideas worth stealing.
01:00concept
The Two Camps of AI Video Content
- Camp 1: one cool image flex post
- Camp 2: course selling screenshots
Joey frames the entire market as either shallow flexing or gatekept education, and positions his free system release as a third path.
Steal forany positioning video where Joe wants to separate himself from courses and hype posts
08:16model
Character + Scene Pipeline
- Build character sheet (director skill)
- Place character in scene
- Get reference image
- Upload to Claude
- Build scene prompt
- Specify shot count
- Estimate seconds for credit budgeting
Mira delivers the clean 7-step pipeline Joey forgot to include. Character and scene are always the two inputs everything else flows from.
Steal forany build-in-public content where Joe shows a repeatable production workflow
How they asked for the click.
VERBAL ASK
08:07link
“Take them, use them, modify them, build your own off of them.”
No link shown on screen — skills dropped in description/comments. The AI character delivers the usage tutorial instead of Joey, which is a format-within-a-format CTA.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
