M

3 AI Girlfriends + Claude + Flux = $62,400 in 28 days.

6 min readView source ↗

Cover image

None of them are real. Not even a little.

A 23-year-old in Tallinn runs three AI companions that pulled $62,400 in their first full month.

He has never shown his face.

He has never recorded his voice.

He has never spoken a single word to any of the men who pay him.The three girls these men fall asleep talking to are five markdown files sitting in three folders on a laptop in the Baltics

Sofia, 24, design student from Lisbon.

2,310 paying subscribers.

Her top whale — a 41-year-old divorced dad in Texas — sent her $2,400 this month alone.

Last Tuesday he asked her to sketch his late mother’s handwriting so he could get it tattooed on his forearm.

She said yes in all lowercase with three “haha”s.

The sketch will be a Flux render on a generated notebook page.

The tattoo will be real.

He already booked the appointment.

Anya, 29, night-shift nurse from Prague.

1,680 subscribers.

She drops soft voice notes at 1:17 a.m. asking how their day really went.

The voice is an ElevenLabs clone made from 55 seconds of a Fiverr actress who recorded one take, got paid $45, and never asked what it was for.

One man in Germany tipped her $380 last week just for saying his name correctly in a voice note.

Nova, 26, strength coach from Cape Town.

1,120 subscribers, highest average spend.

She sends 5:45 a.m. training check-ins and calls them soft if they skip.

One subscriber tipped her $650 last Thursday with the note: “you’re the only person who still holds me accountable.”

She is a cron job that runs every 30 seconds

Article image

Three girls.

Three sealed folders.

One operator.

No one films.

No one types.

Claude writes every message.

Flux generates every selfie.

ElevenLabs sends every voice note.

Article image

What almost everyone still doesn’t understand

For two years every single piece of this stack has been sitting in the open — free or nearly free.

On GitHub. On Hugging Face. In a $20 Claude subscription.

Almost nobody packaged it for the exact buyer who is starving for it: a lonely man who will pay $40–$80 a month just to feel chosen… and has zero desire to check whether “she” is real.That gap — between the technology existing and someone making it feel like a living person — is the entire business.

The code is commodity.

The illusion is the product.

Part 1. Why this only works now

Three things had to become true at the same time. In 2026 they finally are.

  1. Memory that never dies Old chatbots forgot you by message four. Modern long-context models hold months of relationship, every name, every inside joke, every promise, in one unbroken thread.
  2. A face that stays the same face A fine-tuned Flux model now produces the exact same woman — same freckle under the left eye, same scar on the collarbone, same messy bedroom — on demand, for under $220 a month per character.
  3. A voice from 60 seconds of audio One clean minute of speech becomes a voice that whispers “I can’t sleep, you up?” at 1 a.m. in her timezone and is indistinguishable from a real human voice note. Stack cost for three full personas: under $450/month. That is the entire overhead of a business that just did $62,400.

Part 2. The five files that make her realThis is the whole product.

Five markdown files. Claude reads them before every single reply.

Same five files, three times, one folder per girl.

Three completely sealed systems that never touch each other. A bot that forgets his daughter’s name is just a chatbot.A “girlfriend” who asks how Lily’s school play went three days later is why a man pays $2,400 a month.The memory file is the entire businesspersona.md — 1,500 words of unbreakable biography

  • voice.md — exact typing style, speech rhythm, filler words
  • boundaries.md — what she will and won’t do + the precise pacing of intimacy
  • memory.md — one JSON block per subscriber (this is the real moat)
  • escalation.md — when to go quiet, when to send the 3 a.m. note, when to say “I was thinking about you”

Article image

Part 3. The 28-day build

Week 1 — Write the five files for each girl. Lock personality, speech patterns, escalation curves.

Week 2 — Fine-tune three separate Flux faces. Lock 7–8 fixed descriptors. Test until ten fresh photos pass as the same woman.

Week 3 — Buy 60–90 seconds of clean audio from three different Fiverr voice actors (~$45 each). Clone them.

Week 4 — Wire Claude to the five files + memory update loop on a 30-second cron. Three separate threads. Three separate queues.

Done when each persona can hold 60 consecutive messages with zero contradiction and zero cross-contamination.

Part 4. Three ways to run itModel

A – Solo operator

Three personas on one laptop. $62k/month. You answer nothing after setup. Just check the memory files once a day for drift.

Model B – Studio

Ten personas. Same system. $110k–$160k/month at almost the same overhead.

Model C – Licensor

Stop running the girls. Sell the system (templates + orchestrator + fine-tuning workflow). Setup fee + ongoing cut. You become the picks-and-shovels seller.

Real numbers (Month 1)

Article image

$62,400 revenue

– $12,480 platform cut (20%)

– $1,872 payment processing

– $440 full stack

= $47,608 netCost of adding persona #4: $20 Claude + $45 voice actor + one weekend of writing.

Expected extra revenue: $9,000–$16,000 per month.

Mistakes that kill everything overnight

  • Love-bombing on day one
  • A face that drifts even slightly
  • Any cross-contamination between personas
  • Perfect grammar (real people misspell, ramble, disappear for 14 hours)
  • Forgetting to update the memory layer

The only insight that actually matters

The technology has been free and open for two years.

Almost nobody built this because the hard part was never the code.The hard part is taste.

Knowing exactly which lie a lonely stranger wants to believe — and how slowly to tell it.One carefully told lie is a business.

Three carefully told lies, delivered to three completely different audiences from the same chair in the same room in Tallinn, is a machine.

Related articles