Kiyoro (@0xKiyoro)

A 19-year-old showed $51,026 from AI kids videos and accidentally showed the real product It was not the cartoon. It was the wrapper. Her pitch is simple: find a kids video with millions of views, copy the description, paste it into Creao AI, download the output, then post 5 to 10 times a day. The example starts with a cute 2D Hey Bear sensory video. The output becomes a 3D pineapple baby and a strawberry baby eating itself with a spoon. It sounds absurd until you remember the audience. A 3-year-old does not care if the plot makes sense. If the colors move, the song loops, and the character is simple enough, the video can get replayed until the iPad overheats. One hit 6.2 million views. Another passed 1.1 million. Then the dashboard gives away the backend. Claude Sonnet 4.6 sitting in tiny gray text. The “secret AI tool” is not some private model. It is a chat wrapper with a clean interface and an API key. That is the real lesson. The money is not only in making AI videos. It is in packaging the workflow so other people pay for the shortcut.

A 21-year-old Chinese guy spent $64 on tools and made $1,800 in one week from YouTube Shorts. Not from one channel. From 6 fresh accounts running the same system. The stack was simple: Claude, CapCut, AI voice, stock clips, an antidetect browser and six YouTube accounts. The work was not “make original videos.” The work was finding patterns already winning: first 3 seconds, caption style, loop, sound, thumbnail and comment bait. If a format died after 48 hours, he killed it. If one started moving, Claude turned it into 10 new versions and pushed them across the other accounts. That is the real system. Not one faceless channel. A testing machine where YouTube decides which format deserves more output.

13M views turned into a $20,000/month faceless YouTube channel. A 22-year-old did it by copying the structure of weird reaction videos already getting 500K-2M views. Not the content. The structure. Hook, cut speed, reaction timing, caption style, pacing and payoff. Claude turned one winning idea into 10 script variations. ElevenLabs gave them a voice. CapCut packaged every upload into the same repeatable format. The channel hit 220K views, then 480K, then one video crossed 1.3M. By month 7, he was posting 12-15 videos a month and letting the best few carry the revenue.

A 22-year-old is lying on cardboard, pointing at a MacBook and claiming he hit his first $9.7K month from AI animal Shorts. The method looks simple: ChatGPT gives crazy animal ideas, Kling generates 4-second videos, he posts twice a day to YouTube Shorts. Then he shows the dashboard. But it is not YouTube Studio. YouTube Studio is blue, tracks watch hours, RPM and ad revenue. His screen looks like Shopify analytics. That changes the story. He is probably not making $9.7K from Shorts. He is selling access to the tool that makes people believe they can make $9.7K from Shorts. The money is not in the animal videos. It is in the checkout page behind them.

A 20-year-old guy is making $38,000/month from YouTube without sitting in Premiere all day. The important part is not that he uses Claude. Everyone uses Claude. The important part is that he stopped treating it like a writing tab and turned it into the operator behind the workflow. Claude finds high-CPM niches and turns them into ideas. Python scripts turn those ideas into scripts. Premiere Pro turns the scripts into finished videos. The system runs all day, so the bottleneck is no longer staring at a blank timeline. In the first month, his channels crossed 1M views on Shorts. Then he turned the same machine into a service. He charges $400 per client video. Each video takes about 20 minutes of his actual time. Around 25 videos a week adds another ~$10,000. Most people are trying to make better prompts. He built the pipeline around them.

A 17-year-old does not make YouTube Shorts anymore. He runs video infrastructure. 12 channels. 24 uploads a day. 3 hours of human work. $100,000+/month. Each channel has its own niche, its own content rhythm and its own testing surface. Claude writes 30 ideas per channel per week. ElevenLabs creates the voiceover in 40 seconds. CapCut assembles the video. Python uploads when the RPM window is best. That is the part most people miss. He is not trying to be a better creator. He is increasing the number of times the system can test a video before the day ends. Month 1 was $0-$800. Month 3 was $4,000-$8,000. By month 12, the system was doing $100,000+/month. Most people still ask, “What niche should I start?” The better question is, “How many testing surfaces can I run before YouTube finds the winner?”

One laptop, three TikTok girls, $14,000 a month, and not one of them is real He runs all three at once because copies of a fake creator cost nothing > 45k, 95k, and 145k followers > over 6 million likes combined > three different girls, three audiences, one operator > the followers and the money are real, the girls aren't A human creator maxes out at one account and their own time he runs three on the same hours, and could run thirty

a girl cut her ai costs from $512/month to $19 by buying a $599 box once, the mac mini m4 is quietly the cheapest ai setup of 2026 i ran the math because the gap sounded fake. it holds up > a heavy stack of claude, chatgpt, cursor and api usage hits ~$512/mo, about $6,144 a year > the mac mini m4 starts at $599 and runs on roughly $4/mo of electricity > ollama plus open webui runs qwen, deepseek and llama locally > if it handles even 75% of the work, it pays for itself in about six weeks she still keeps one cloud model for the hardest tasks, that part's honest but once a small box on the desk does most of the work, paying full subscription price for everything starts to feel stupid

the $22k a year everyone posts about local ai is the loud number. the quiet one is bigger, the contracts you stop losing on data residency NVIDIA's DGX Spark is a $2,999 box that puts a 128GB ai supercomputer on a desk. it runs the same 70B models people have been renting in the cloud, except nothing crosses the network and no terms of service govern a machine you own that second part is the business. own-compute stopped being a budget optimization, it's a sales motion. clients who couldn't legally send data to a cloud model can buy now, and the freelancers who own the hardware are the only ones who can sell to them cheaper was never the point. ownable was

$18,796 of hardware replaces a cloud bill that runs $1,900 a month, and pays itself off before the third invoice the box is a DGX Spark, $2,999 each. 128GB unified memory, a petaflop of compute, 70B models at full precision, where a 4090 caps out at 24GB and sends you back to renting i've been reading into local setups and the scaling is the part people miss > two boxes over the 200Gbps ConnectX-7 link, 256GB, up to 405B parameters > four boxes, near-linear, around 700B after payback it's about $22,000 a year staying in the business instead of routing into someone else's data center. fine-tunes that ran $400 a piece cost nothing, and client data that can't touch a public model never leaves the room cloud pricing teaches you to ration ai. ownership deletes the hesitation the edge wasn't a faster model, it was noticing the rent