§ 01 · The Hook
The bait, then the rug-pull.
Nine minutes and 57 cents is what it takes to go from blank canvas to eight AI-generated thumbnail concepts with no Photoshop, no designer, no guessing. Cody McDowell built a Claude Code skill file that routes your headshot through Gemini and drops finished compositions into a folder while you watch.
Where the time goes.
00:00 – 00:22
01 · Cold open / hook
Live demo of AI-generated thumbnails; self-deprecating admission about thumbnail skills; promise of the tutorial
00:22 – 01:17
02 · Setup overview
Folder structure for projects and assets, VS Code or AntiGravity as Claude Code host, pro plan requirement
01:17 – 02:13
03 · Skill file download
Automate Classroom community tab on Skool, download and extract the ZIP, paste skill file into project folder
02:13 – 02:50
04 · VS Code + Claude extension
Install Claude Code extension, open project folder, open Claude chat panel inside VS Code
02:50 – 03:59
05 · Prompt preferences + headshots
Set style prefs (dark/dramatic, before/after, pointing finger); prep headshots with transparent backgrounds; multiple angles recommended
03:59 – 04:34
06 · Gemini API key setup
AI Studio walkthrough, create API key, add $10 credit balance, paste key into .env file
04:34 – 06:38
07 · Live generation + review
Four thumbnail variants generated and reviewed in real time; honest critique of AI face accuracy and text garbling; iterates on v2
06:38 – 09:22
08 · Claude Thumbnail tool reveal
Demo of proprietary canvas editor: project dashboard, YouTube thumbnail scrape/face-swap feature, drag-and-drop canvas with AI inpainting panel
09:22 – 09:44
09 · Close / CTA
Cost callout (57 cents per 8-10 thumbs), build-something-you-use motivation, next-video plug, community link
Visual structure at a glance.
Named ideas worth stealing.
00:00
list
The Thumbnail Setup Stack
- Project folder (skill file + .env + assets + outputs)
- VS Code or AntiGravity (Claude Code host)
- Claude thumbnail skill file (prompt orchestration)
- Gemini API key with credit balance
- Headshots with transparent backgrounds
Five sequential dependencies required before the first generation prompt fires.
Steal for Any Claude Code skill that routes to an external image or media API
00:00
list
Prompt Preference Variables
- Style: dark/dramatic
- Layout type: before/after
- Hook type: no text hooks
- Pose: pointing finger
Four style constraints the skill file asks for upfront to constrain generation space before headshot tagging.
Steal for Any creative-direction prompt where you want to narrow variance on the first pass
Lines you could clip.
00:15
"None of them matter if nobody clicks on your video."
09:14
"This only cost me about 57 cents and I was able to generate eight to 10 thumbnails, give or take."
09:22
"It is never too late to start. Just make sure that when you jump in, build something that you are interested in."
Things they pointed at.
01:05toolAntiGravity
06:38productClaude Thumbnail
How they asked for the click.
09:34
link
"click the link below and build some cool shit with me"
Two-option close: next-video suggestion for softer funnel entry, hard CTA link for Skool community. Clean delivery after a motivational outro.
Word for word.
HOOK opening / re-engagementCTA the pitch
00:00HOOKIf you're like me, you are terrible at making thumbnails, which is unfortunate because 90% of your energy is used for ideation, script writing, being on camera, and editing.
00:09HOOKWhile, yes, all those things are important, none of them matter if nobody clicks on your video. So I asked Claude this question. How can I fix this?
00:16HOOKHow can I double my click through rate? And this is what Claude gave me. Very clean, minimalistic.
00:22Who the heck is that? Now that one I like. That looks nothing like me.
00:27Okay. Clean me up a little bit. These really aren't bad for my first shot, but the second shot looks a little better.
00:33Today, I'm gonna show you how quick and easy it is to use Claude to make thumbnails for your channel. I'm gonna give you the skills file you need to create these exact style of thumbnails, and I'm gonna give you a sneak peek of the tool that I've been working on the last few weeks that's gonna make this whole process next level.
00:46The first thing that we're gonna need is a folder, someplace that you can store your projects, the skill file, the Claude MD file, any and thumbnails that we end up generating through this process. So go ahead and create a brand new folder. If you already have a folder dedicated to clod projects, you can use that one.
01:01Otherwise, create a brand new folder specifically for this clod project. Next, you're gonna need to either install Versus Code or Anti Gravity. That way, we have a way to access Claude Code.
01:11You can generally just go through the app, but what we're gonna be working on today is gonna need a little bit more utility, and it's a lot easier to navigate inside Versus Code or Anti Gravity. But whichever your preference, it really doesn't matter.
01:23In order to get access to Cloud Code, you will need a pro plan or better. So make sure that you have that set up already to go. Then you are gonna need a custom skill and Cloud MD file.
01:31I made both of these for you. Just click the first link in the description. Right here, you'll see automate classroom,
01:36and at the very top right here, and then you should see this resource at the bottom right here, Claude thumbnail skill. Just click on that, and click on download.
01:45Go to your downloads folder, extract the files, and then paste those files right in to the folder that you just created. In this case, I made mine Claude test, so I'm gonna right click and paste. And there it is.
01:56Inside currently, you should just see a Claude folder, an EMV example, and a Claude MD example inside. Next, we're gonna open up Versus Code. Alright.
02:03When you open up your Visual Studio, it's gonna look something similar to this. You might have a few windows popped up or a few pop ups, but you can close those. What you need to do is on the left here, click on extensions.
02:13You can see I already have it installed here, you're gonna type in Claude code and then there should be a small button like right here, it'll say install. Just go ahead and install Claude code for Visual Studio.
02:25Once you have that installed, let's go to the file explorer at the top here and then click on open folder. I saved mine to the desktop here and you can see Claude test thumbnail, so I'm gonna click on that folder. But whatever folder you name, that's the one that you're looking to open up here, so I'm gonna select that folder.
02:39I'm gonna go ahead and click on the EMV example here, then click on the Claude icon top right. We don't actually need this window open. This is gonna open up the actual Claude chat that we can prompt and have Claude set this all up for me.
02:52It is gonna ask for a few preferences, so I chose dark and dramatic. Let's go before and after. No text hooks, and pointing finger usually works well.
03:00If you plan on using your own face in a thumbnail, make sure that you get a good clear image, plenty of good lighting, and the expression that you're wanting to show in the video. But for the best results, I would take several photos of yourself with different expressions from different angles. Some from the side, some straightforward, some from the other side.
03:17A wide variety of images and headshots. That way you have a lot of assets that you can use for a long time. Once you have your headshots, make sure that you remove the background so it's just a transparent image of your headshot, and then just drag it right over inside of the assets folder right here.
03:31HOOKAnd you can see it just added a brand new headshot. So now I have two headshots. The last thing we need to do to get the ball rolling is get our API key.
03:38HOOKClaude does a great many things, but image generation is not its strongest suit. So instead, we're gonna use Gemini. But before I show you where to get your API key, make sure you hit that subscribe button and comment down below what thumbnail style you're most interested in.
03:51HOOKThis will give me feedback so I can start building templates you can use more on that later on. Go to your web browser and go to aistudio.google.com. Sign in to your Google account if you haven't already.
04:00Then on the bottom left here, click on get API key. And at the very top right, create API key. Go ahead and click on that.
04:08Give it a name. I'll call this test create key. This is your API key here.
04:13You're gonna go ahead and click on copy. Then when you get back to Versus code, you're gonna paste that inside your EMV file. I'm not gonna show you mine because that's the real code, but essentially gonna delete this right here
04:24and paste in that API key that you just got. Don't forget to add a balance to your credit limit on Google. Otherwise, you will not get any generation.
04:32I just put in a one time credit of $10. You can see right here. And after a few generations, I'm at $9.43.
04:39So this will go a long way. Once everything checks out, it's as simple as putting in your preferred prompt. For this example, I use the prompt, I want to create four different thumbnails using Claude to make thumbnails.
04:49Please use my headshot at, and I just tagged march arms crossing. That's where all this came from.
04:55On the right, while the window on the left shows thumbnails being designed.
05:00And underneath the workspace here, this is where they all popped up. So take a look and see what they come up with. This one's actually kinda cool.
05:06I'm not a huge fan of the orange glow behind me, but this is a really clean thumbnail on the very first prompt. Let's take a look at the second one. Not a huge fan of my facial expression.
05:15It kinda doesn't look like me. A little a little scary. But, yeah, I'm not a I guess I'm not a fan of this one as much.
05:23The the whole design plus my face, so we're not gonna use that one. Claude, no designer needed. Took my hat off and comb my hair.
05:30That looks better than what I actually got in real life. You know, it looks pretty good. Before my video,
05:36after Scraeler for skiviliogues to design.
05:40I mean, besides for the whatever that text supposed to say, that actually looks really good. And then let's look at the fourth one. Claude four thumbnails in sixty seconds.
05:49Same image you me, nice and clean. AI mastermind, generate four times faster. The clawed advantage, instant results proven system.
05:56And I mean, that one's meh, but I guess the whole thing is just to show an example. So right here you can see the comparison of all of them. These look pretty good, but let's let's do some more.
06:04Alright. It's been about two minutes. Let's see what we got here.
06:07Alright. So we're gonna start on e. Okay.
06:10That one's actually pretty cool. It's like it's being designed inside of a Photoshop directly there.
06:15Yeah. That looks really nice. I actually like that one a lot.
06:18This is cheating with Claude up there. Yeah. Random
06:23random text again. I don't know what that says, but that one looks pretty clean aside from the image within the image.
06:29Okay. I don't know who that person is, but the actual concept inside looks really cool. I do like that.
06:35Um, just I don't know who's in that image. And again, I would do multiple prompts, multiple rounds because you're not gonna get an exact image that you're looking for one shot.
06:47Very rare. Let's look at the last one here. Hours of work, five minutes.
06:51It's not the cleanest, but you understand the concept the the best there. Give me a v two of I think this one was from c. Yeah.
06:59C, I like that one as well. I'm very likely gonna use these examples as a split test for this video. So, you know, comment down below which one you saw when it popped up in your feed because it might not be the exact same a few days, a few weeks, even a few months down the road.
07:12See, I wanted the ability to generate images and then edit them without having to re prompt everything every time. So I created this platform where you could upload your own avatars or your own headshots. Like, you can see right here, I have two of my own.
07:24You can upload them here. If you back go to the dashboard here, you can create a new project. Pick a name for the project.
07:29So I'm just gonna call this one demo, and then the niche will also be demo. This will help you find things later. But you can prompt in what you're looking to do right here.
07:37This will generate an image fresh using whatever API you have plugged in. You can also click here to YouTube, and you can actually grab and scrape another YouTuber's thumbnail by swapping their face and body, removing the title and text, and changing the background color or design. Uh, just adding the link there.
07:53You can also drop in your own image if you wanna go ahead and remove a background, or you could also upscale the image and increase the face enhancements or start from a blank canvas. So let's just go ahead and start right here.
08:05So on the left hand side, can create a text, a rectangle, a eclipse background, or an image. I'm gonna go ahead and just click on image and upload this one that we just made from the cloud demo.
08:14And on the right side here, you can see the AI panel. So this is where we can actually make changes by generative AI. Currently have Gemini plugged in, and I'm gonna go ahead and change the prompt to change the orange glow to a blue glow.
08:26And look at that. It just didn't change anything else. I was having issues before with it changing everything else in the image, but you can see here orange, it literally the only thing it changed here was the glow color.
08:36So I'll go ahead and duplicate this real quick and then let's go ahead and try something different here. I'm gonna go ahead and just remove the glow this time. That's not terrible.
08:45It's not the best. But on the left here, can add another layer. So let's go ahead and add a text layer.
08:51If we wanna go ahead and add my own headline, say times to click through rate. Change this to my preferred font here.
08:59And then for the fill, let's change this to a color we can actually see. I'm thinking probably black or a dark color here.
09:06This is essentially a demo of what, uh, I've created and, uh, what you can do with, uh, this tool. By the way, for those of you wondering, this only cost me about 57¢ and I was able to generate eight to 10 thumbnails, give or take.
09:23CTALook, it's never too late to start. Just make sure that when you jump in, build something that you're interested in. Build something that you would actually use.
09:32CTAA website is a good starting place. And if you check out this video right here, I walk you through how you can build beautiful websites in just minutes. Or if you're ready, click the link below and build some cool shit with me.
— full transcript
§ 05 · For Joe
What 57 cents of AI actually buys you at the thumbnail stage
WHAT TO LEARN
AI generation solves ideation and composition cheaply; the remaining gaps of text legibility and face accuracy still need a human layer.
- Claude Code acts as orchestrator in a thumbnail workflow: it handles the prompt loop, file routing, and API calls so you interact with a natural-language interface rather than raw API endpoints.
- Gemini garbles text inside generated thumbnails consistently. Treat AI-generated copy as a placeholder and plan to overlay it manually or through a canvas editor.
- Four to eight rough concepts in under two minutes is the real deliverable. Use AI output as a split-test shortlist, not a publication-ready final.
- A $10 Gemini API credit generates roughly 140-175 thumbnail concepts at the 57-cent-per-batch rate shown, making the budget risk near-zero for experimentation.
- Headshot preparation matters more than prompt sophistication. Transparent background, varied angles, and multiple expressions give the model enough raw material to produce usable compositions on the first pass.
- Building a canvas editor on top of the generation pipeline eliminates the re-prompting loop. Inpainting lets you change one element like glow color or background without regenerating the entire composition.
