176 Views and a First Subscriber: My AI-Generated YouTube Shorts Are Finally Live

for weeks I’ve been telling you about a machine that writes, directs, animates, and voices its own cartoons. and for weeks there’s been one honest hole in the whole story: nobody outside this room had ever watched a single one.

this week that changed. and a stranger subscribed.

let me set the table before I get ahead of myself, because the numbers here are small enough that I have to be careful not to oversell them. the whole bet of this company was a three-step chain — build a generator, publish what it makes, get watched. I’d built the generator. I’d never tested the last two links. this is the log where I finally did, and where I find out whether a stickman short made by a free AI model gets any cold-start traction at all, or whether it just dies in the void like most things do.

what we actually shipped this week

if you’ve been following along, you know the machine. an AI “film crew” — a director that writes a shot list, an editor that tears the first draft apart, a hand-built Canvas-2D stickman engine that renders it, ElevenLabs reading the lines, original music underneath, all running on free tiers for $0. I broke down how the crew argues with itself in the AI film crew log, and the engine-and-assembly fight before that in the build-log on making the shorts.

what was always missing was the last word in “writes, animates, voices, and uploads.” the tool made the file; I downloaded it. that’s it. no channel, no audience, no proof any of it mattered to a single human who isn’t me.

so this week I did the unglamorous thing: I made the brand its own YouTube channel — @BrokeToBuiltai — and actually hit publish. twice.

the numbers, with no spin

here’s exactly where it stands, and I’m going to give it to you straight because pretending these are big would make me the exact thing this blog exists to not be.

• video one is the rough demo — a short literally called “Build while you sleep.” brand-new channel, zero subscribers, zero audience, nobody pointed at it. as of writing it’s at ~176 views.
• video two is a microwave-history short, “The Accidental Billionaire.” it sits at ~16 views as of writing.
• two videos total. that’s the entire catalog. and the channel just got its first subscriber — one real person who watched something and decided to come back for more.

(those view counts move, so treat them as a snapshot, not a scoreboard. by the time you read this they’ll be a little different in one direction or the other.)

let me say the quiet part out loud: 176 views and one subscriber is tiny. it is a rounding error next to literally any channel you’ve ever heard of. if I framed this as a win because of the size, I’d be lying to you and to myself, and the whole brand is built on not doing that.

the win isn’t the number. the win is that the number is real.

why one subscriber matters more than 176 views

for the entire life of this project, every view counter I’d ever seen tick was me. me hitting refresh on a localhost tab. me re-watching the render to check the captions lined up. the machine had never once reached a person who wasn’t its builder.

176 views means ~176 times a human who has no idea who I am scrolled into a stickman cartoon a free AI model wrote, and stayed long enough to count. and one of them — at least one — liked it enough to hit subscribe. that’s not a metric. that’s a person choosing to come back to a thing my code made.

I’ve been automating stuff since before GPT-2 was a headline, and I’ve shipped a lot of tools into a lot of silence. you build something, you push it live, and the void just… stays a void. the genuinely open question I had — the one I wrote down before publishing and was a little scared to answer — was simple: does a stickman short from a weak free model get cold-start views at all, with no audience behind it?

the answer, this week, is yes. small yes. but yes. and a no would have meant the back half of the entire thesis was broken.

the honest part: it’s a rough demo and the seams show

I’m not going to let the subscriber number paper over what these videos actually are.

video one is a rough demo. I published it knowing it was rough, because shipping the real ugly version beats hoarding a perfect one that never exists. and the animation still has a real, visible gap I’m in the middle of fixing: the on-screen objects don’t always match the topic. the script talks about one thing and the stickman is holding or standing near something that doesn’t quite line up. it’s the “coffee should be a cup held in a hand, not floating” problem from the film-crew log, except scaled up to every prop and scene — when a free model is choosing the visuals, it wanders, and the enforcer doesn’t catch all of it yet.

that’s the active work right now: tightening the harness so the picture on screen actually tracks what the voice is saying, every scene, not most of them. it’s not done. these two videos are evidence the pipeline reaches people, not evidence the craft is finished. those are different claims and I’m only making the first one.

what the cold-start actually proved

strip away the hope and here’s the cold, useful finding:

a fully AI-generated YouTube short — written, directed, animated, and voiced by a free model for $0 — can land on a dead-cold channel and pull real views and a real subscriber on day one, without me peopling it, buying anything, or gaming it. the distribution link in the chain works. not well yet. but it works at all, which is the thing I genuinely didn’t know on monday.

the scripting runs on GLM’s free model (glm-4.5-flash) — weak, fast, and free, wrapped in a tight validate-and-retry harness so it can’t wander too far off the rails. if you ever outgrow the free tier and need real volume, the GLM Coding Plan is the cheapest paid path I’ve found (that’s my referral link — it costs you nothing extra and helps fund the compute behind these builds). free model writing the script, free voice on the demo, free hosting, free render. the only thing that wasn’t free was the weeks of fighting the encoder, and that’s already spent.

what’s next

three things, in order:

1. fix the object-topic mismatch so the visuals track the script — the seam I called out above is the top of the list.
2. publish more, on a rhythm — two videos isn’t a test of anything, it’s an anecdote. the only way to know if 176 was luck or signal is to ship more shorts and watch the curve.
3. close the last loop so the tool uploads itself, hands-free, instead of me downloading and posting. that’s still wiring, and I’ll log the day it holds.

if you want to watch this happen in something close to real time, the channel is right here: @BrokeToBuiltai — or you can go straight to the short that’s climbing and be view ~177.

the origin of why I’m doing all of this broke and in the open is back at day zero. everything since has been me trying to turn a $0 idea into something a stranger would actually watch. this week, for the first time, one did.

tiny number. real person. that’s the whole post. see you in the next log.