It was a Tuesday night around ten-thirty and I was sitting at the kitchen table with my sticker-covered MacBook, a reheated cup of coffee, and a YouTube rabbit hole that had started with "how to grow short-form content" and ended — somehow — with a guy in a ring-lit office explaining that he had not filmed a video himself in four months. His AI avatar did it. He just typed the script.

I had been doing the Reels thing inconsistently for a while — filming from the car, mostly, thirty seconds of me being either charming or tired depending on the night. One of them went further than it deserved. Most went nowhere. The problem was never ideas; it was the hour it took to film something that looked like I had spent ten minutes on it. So when I saw HeyGen, the AI video tool that lets you create an avatar from a short clip of yourself and then generate videos from a text script, I did what I always do: I opened a new tab and started a free trial before I had thought it through.

I told myself this was research. I told myself that if it worked, I could batch a whole week of content in one sitting. I was wrong about the second part, and only partially right about the first.

Why a Tired Software Engineer Decided His Face Needed to Be a Product

My short-form content habit has always had the same ceiling. I get a good idea, I film it, I edit it on my phone standing in the driveway, and I post it. Sometimes it lands. More often I spend forty minutes on something that gets eighty-three views, eleven of which are probably mine from checking the analytics too early. The bottleneck isn't creativity — it's that I have a day job, I have kids, and some nights the last thing I want to do is turn a camera on my face and perform.

I'd been thinking about this more honestly after I sat down and actually counted the hours that don't show up in any payout — the setup, the editing, the re-filming because a kid walked through the background. The number was not flattering. Content creation is mostly overhead with occasional moments of actual content.

HeyGen's pitch was basically: what if you only did the overhead once? You film a two-minute consent clip, the platform builds a digital version of you, and from that point on you type a script and your avatar delivers it. Adjustable aspect ratio, your voice, your face, no ring light required after setup. That sounded like something worth a week of experimentation, even if I knew from experience that "worth experimenting with" and "worth anything" are two different sentences.

The HeyGen Setup, the Aspect Ratio Question, and the Part Where My Avatar Sounded Drunk

The avatar creation process was genuinely straightforward. You record a short clip of yourself speaking — they want good lighting, a neutral background, you looking more or less at the camera. I did mine in the living room at eleven PM with the overhead light on, which was not ideal, but it was fine. Processing took a few hours overnight. By Wednesday morning I had a digital version of myself staring back at me from a browser tab, and I want to be clear: it was uncanny in a way that I was not prepared for at ten in the morning.

The first thing I had to figure out was the HeyGen aspect ratio situation. The platform lets you choose — 16:9 for landscape, 9:16 for vertical Reels and TikTok, 1:1 for square posts. For what I was doing, 9:16 was the obvious call. That part worked fine. You set it before you generate the video and the avatar is framed accordingly. No complaints there.

Then I tried the HeyGen voice cloning. This is where things got interesting in the wrong direction. The platform can clone your voice from the same consent clip and use it to deliver whatever script you type. In theory, it's your voice. In practice, the first three videos I generated sounded like me after two beers explaining something I only half understood. The cadence was slightly off — a little flat, a little slow in the wrong places, pausing on syllables that don't need pausing. My oldest walked by while I was playing one back and said, without stopping:

"Dad, that sounds like you but sadder."

Which, honestly, was fair. I spent another evening adjusting the script punctuation — adding commas to force pauses in the right places, removing them where the avatar kept sighing — and got it to about seventy percent of normal. The voice cloning feature also has a singing option buried in there somewhere. I tried it once out of pure curiosity. I will not be describing what happened. It has been deleted.

I generated seven videos over the week. Thirty seconds each, 9:16, scripted by me, delivered by my slightly-sadder digital twin. Posted all seven across the same platforms I normally use. I tracked the numbers the same way I track everything: a Notepad file from approximately 2010.

What the Free Trial Didn't Cover and What My Time Was Actually Worth Here

The free tier on HeyGen has generation limits. I hit them by Thursday. To finish the week's experiment I upgraded to a paid plan for the month — I'm not going to state the exact price because these platforms change their pricing roughly every time I check, and I don't want to send anyone in with wrong numbers. What I'll say is: it was more than I expected for a week of testing something I wasn't sure about, and I did what I always do, which is spend an extra twenty minutes hunting for a promo code before paying. I found one. It helped a little.

Time-wise: the avatar setup was maybe ninety minutes total, spread across two evenings. Writing and adjusting seven scripts was another three hours — because it turns out writing for an AI avatar that has to sound like you requires more precision than writing for yourself, where you can just improvise in the moment. Each generation took a few minutes. Reviewing playback, cringing, adjusting, re-generating: another couple of hours I didn't account for going in.

The seven videos collectively did about the same as my normal Reels — which is to say, unspectacularly. One did slightly better than average. The others were average or below. There was no meaningful difference in performance between the avatar videos and the ones I film myself in the car. If the algorithm noticed I wasn't physically present, it did not seem to care either way. The one that performed best was the one with the tightest script, which is probably the lesson, except I don't want to call it a lesson.

What I'm Still Using, What I Cancelled, and the Specific Thing I Should Have Done Differently

I cancelled the paid plan at the end of the month. Not because it was bad — the technology is genuinely impressive in a way that still weirds me out a little — but because the thing I was trying to solve didn't actually get solved. I wanted to save time. Instead I traded filming time for scripting and QA time, and those turned out to cost roughly the same number of hours. The avatar doesn't save you from having to think. It just moves where the thinking happens.

What I kept: the habit of writing a tighter script before I film anything, even when I'm filming myself. That part transferred. The avatar's limitations forced me to write more precisely, and it turns out more precise scripts make better videos whether a human or a slightly-sad digital clone is delivering them. That's worth something, even if I can't put a dollar figure on it.

What I dropped: any idea that AI avatar tools are a shortcut for someone at my level of content volume. This is one of those things that probably makes more sense if you're already producing at scale — which I am not, and the gurus who swear by it tend to leave out that detail. They're running operations. I'm a guy with a Notepad file and a week of evenings.

The voice cloning singing thing I mentioned — I want to reiterate that it is gone. I don't know what I expected. I think I expected something funny I could post. What I got was something I cannot explain in a way that would make you understand how wrong it was. Some tools exist and should not be used. That is the whole story there.

The AI tool side of my life has a consistent pattern: I build or find something that nearly works, get genuinely excited about the nearly, and then spend too long in the gap between nearly and actually. HeyGen is a real product. The gap was just wider than I thought on a Tuesday night at ten-thirty.

The avatar is still in there, technically. The account is on the free tier now, sitting idle. Every couple of weeks I think about trying another video just to see if I've figured out the voice cadence issue. I haven't gone back yet. Maybe I will. Maybe the slightly-sadder version of me is just waiting patiently in a browser tab, script unwritten, aspect ratio set to 9:16, ready to post the Ls too.

Note: Everyday Hustle Jeff is the written side of a digital series. These accounts are composite and dramatized. Any figures are what they were on the day, for one person, and are not a projection, a track record, or advice.