Sam asks what your current workflow is for creating thumbnails, especially for automating this process for YouTube creators.
I get about 10 people each day offering to design thumbnails for me, often using some scraper to find channels in a niche and offering a free demo thumbnail. I take them up on the offer; for example, I worked with a designer named Hanan from India/Pakistan, paying $30 per thumbnail. My workflow is simple: after recording my daily update, I record a quick 10‑second clip of myself making ugly faces at the camera—only my thumbnail editor sees this. I then tell the editor to upload the finished thumbnail to the same Google Drive folder, I post the video, and the next day I update the thumbnail. For my other thumbnails (like the ones on Nix), I start with a Canva template, grab a smiling screenshot from the earlier clip, delete the previous face, paste the new one, add another copy, use the background remover to isolate the face, creating two layers so I can place hair over a green section for a 3D effect. I then add platform logos (Instagram, NN) and do light editing in Clarity to make my face pop without maxing it out. The result looks like a YouTube‑sized thumbnail when zoomed out. I don’t spend much time on thumbnails, but they’re effective because consistent, quality content makes viewers associate my face with value, turning thumbnails into a self‑fulfilling market signal.
Let me show you how I generate my thumbnails. There's no generation happening in my daily update thumbnails; it's me in Canva. I have a little template with logos I drag and drop, plus my face. If I delete it, you'll see it's me—updated me. What I do is apply a bunch of filters to make myself pop, because YouTube thumbnails are viewed at about this size, so this pops out more than that. The idea is to make it pop, adding foreground and background. You can see the same energy here; it's mildly similar but sometimes I flip it when I don't have a thumbnail picked that day. So here's how I do it: I have layers with logos, then I drag and drop myself in. I'll go straight to OBS, go full screen, stare at the camera, take a screenshot, go back in, paste the screenshot, grab the style from the previous one, and paste it onto this one. Think of it like that. Then I delete the old layer and I'm done. This takes me about three minutes each day. I know this because I've been tracking the time recently. This is all my info product stuff; I'm trying to track it religiously. The packaging takes 5, 3, or 6 minutes. I need to be better at tracking time spent on these things, but I figure daily updates are a bigger chunk of my work. That's going to be important. Okay, cool.
Yeah, sure. Let me do that right now. Canva.com. Copy my worst mistakes running an AI automations agency. Okay, let's just duplicate this. I guess I already have. So, check this out. In Canva, I have a BG or background remover when my screen doesn't lag like crazy. If I reset the tools, this is what the actual background looks like. It's just me looking damn sexy—I might add. Then I remove the background on the main photo. I don't know why there are two of me there; I didn't think that through. Then I go to edit, adjust, and crank up the vibrance and then the sharpness. Sharpness is important. Then the clarity as well. I find I don't know: if you have just two choices, this or this. What would you see on a tiny YouTube thumbnail (an inch by an inch)? Probably the clearer stuff. Does it look the best all zoomed in? No. Then I up the brightness a little bit, make the skin tone as bright as possible, make it stand out. So if I were to show you two examples: let's reset all the adjustments. That's the before and the after. The after hair looks kind of weird because it's white, but it pops more. That's what I do. And then, like the thumbnails, I mean, I don't really make them all myself now for the main channel. I do for this one here, although this was an exception. I made this one technically, but it was based off a template somebody else made for me. So that's how I do it. It's pretty quick.
I use AI to take the transcript of this video and generate candidate title and thumbnail options. I then apply constraints—no more than seven characters per line (so it doesn’t overflow my face) and exactly three lines—and select the best options from a training set of reasonable thumbnails that must confirm the title. The system then creates three thumbnails, we split‑test them, and pick the winner. It’s fully automated; I let the market choose high performers. The value is that I can produce a ton of content very quickly with zero post‑ or pre‑production—I just show up, answer questions, and deliver value. On the business side: we’ve hit about $500 K this month (still waiting for final numbers, but on track). We’ve landed several big affiliate partnerships that are crushing expectations and bringing in far more money than anticipated. There’s also a major insurance client in the works worth roughly $2 M per year—we’ll confirm once payment arrives. I run an annual promo for Maker School; while the MRR shown excludes annuals, the actual cash flow each month is higher because I front‑load the annual payments. Selling around 40‑50 annual spots at $1 104 each adds roughly $50 K, which combined with community revenue (~$280 K) puts us in a strong position. On a lighter note: the smoke from the Canadian wildfires is clearing a bit, it’s windy, and I was able to go outside briefly. I paused my daily updates after realizing we’d already hit the $500 K milestone in just over a month—a testament to consistent daily effort. My thumbnail system is “goated”: I tested many styles (cartoon, 3D, pixel) and found the 3D approach consistently outperforms, backed by AB‑test data. Faceless thumbnails are interesting too—when I use my face, growth stays steady; when I hide it, I see occasional double bounces, likely due to timing effects. Looking ahead, I’m inviting all the Maker School winners I’ve worked with to jump on calls like the one with Justin (which was rough around the edges—I love the guy but messed up some questions). My plan is to run many interviews and, through sheer volume, make the 100th one the best interview you’ve ever seen. On a personal note: I saw the Odyssey with my girlfriend; I’d expected a political rage‑fest over Elliot Page being cast as Achilles, but the film turned out to be a solid movie despite the American accents—totally fine. The whole experience reminded me how much online outrage is just stupid rage bait. Finally, regarding the Canadian wildfires: they’re tragic but also reveal potential upsides—melting permafrost could unlock arable land for new communities, assuming we improve forest management and stop burning the country down each year. I’m not looking to become a political commentator; that was just a joke. Thanks for watching—I’ll see you tomorrow.
No, the thumbnails are not made with Google Gemini. They are procedurally generated: I set a fixed height and width for the container, then I use a large library of photos of myself. AI slightly modifies my hair and outfits in each image, and the system places the text on top. The process relies on Claude code running on your phone. For the text itself, I limit it to three lines and ensure it doesn’t exceed the container’s width or cut into my head; if it does, I remake the thumbnail with different wording. The result does a pretty good job overall. The benefit is that I simply upload the thumbnail to a designated folder, trigger the process, and it automatically generates and publishes the thumbnail. I’ll elaborate later, but editing is minimal—I don’t edit the thumbnails at all. The intro I did edit once, but that was straightforward and took only a few seconds.