A developer who failed 70 times is now earning $132K/month (~$1.7M/year) with a single AI photo generation app. Zero employees, $40/month in server costs. Let's break down how this works.
What Is This?
Photo AI is a service where you upload a few selfies and AI creates professional headshots, LinkedIn profile photos, and even Tinder photos. It launched in February 2023, built by Pieter Levels (levelsio), a solo developer from the Netherlands.
His track record is the interesting part. He started a "12 Startups in 12 Months" challenge in 2014, and has since built over 70 projects. Only 4 made money. That's a 5% success rate. Nomad List, Remote OK, Interior AI, and Photo AI. The other 66 all failed.
Right before Photo AI, he built Avatar AI. It made $150K in its first week, but when Lensa AI ate the market, he immediately dropped it and pivoted to Photo AI. That speed of judgment is the key.
The tech stack is shocking. Not React, not Next.js. 14,000 lines of PHP + jQuery + SQLite. The server is a single DigitalOcean VPS ($40/month). AI computation is delegated to the Replicate.com API, generating photos with Stable Diffusion + DreamBooth fine-tuning. In his own words: "The tech stack doesn't matter. Getting customers and getting paid comes first."
The revenue growth trajectory is impressive too. First week $5.4K → Month 2 $28.7K → Month 6 $61.8K → Month 18 crossed $100K → Currently $132–138K MRR. Costs run about $13K/month (mostly Replicate API fees), putting net margins above 87%.
What Makes It Different?
Levels' approach is fundamentally different from typical SaaS startups. Raise VC, build a team, use the latest stack, perfect it, then launch — that's the "playbook." Levels did the opposite.
| Traditional SaaS Startup | Levels' Approach (Solo AI SaaS) | |
|---|---|---|
| Dev timeline | 3–6 month MVP | 3–4 week MVP, launch immediately |
| Tech stack | React + Node + PostgreSQL + K8s | PHP + jQuery + SQLite |
| Team size | 5–15 people | 1 person (himself) |
| Funding | Seed $1–5M | Bootstrapped ($0 capital) |
| Free plan | Aim for freemium conversion | None. Paid from day one |
| Margin | 20–40% | 87% |
| Marketing | Paid ads + PR | Building in public + organic SNS |
The "no free plan" strategy is particularly impressive. Levels charged money from the start. He admits the early output was "honestly terrible," but people still paid. Why? Because paying users give real feedback. Feedback from 100 paying users is more valuable than 10,000 free users.
His growth channels are unique too. 50% of traffic comes from Twitter/X. The 600K+ followers built through 10 years of building in public are his marketing engine. He posts Stripe revenue screenshots daily, shares product demos, and doesn't hesitate to throw out controversial takes. $0 in ad spend.
When a TikTok influencer mentioned Photo AI, MRR jumped from $12K to $40–50K. Traffic also surged after his appearance on the Lex Fridman podcast. The key is that the product itself creates "results you want to share" — posting your AI-generated photos on social media is viral marketing by nature.
The Essentials: How to Start a Solo AI SaaS Like Photo AI
- Build on top of AI APIs
Don't train your own models. Build your product on top of APIs like Replicate, OpenAI, or Stability AI. Levels doesn't run Stable Diffusion himself — he uses the Replicate API. $40/month in infrastructure is enough. - Charge money from day one
Don't create a free plan. Accept payments even if the MVP is incomplete. "Only people who pay are real customers." Photo AI's early quality was poor, but it improved rapidly through paying user feedback. - Launch within 2–4 weeks
Perfectionism is the enemy. Ship at 70% done and fill the remaining 30% with real user reactions. Levels deploys dozens of times a day — 37,000 git commits in 12 months. - Focus on a niche
Not "AI generates all images" but "professional headshots + profile photos" — focus on a specific problem. Photo AI's ARPU is $60–70 because the use case is clear. - Build in public to create a distribution channel
Share your revenue, failures, and lessons publicly. Even 6 months of consistency makes a huge difference in early traffic, even if you don't have 10 years. Products without an audience typically earn $500–2K in the first month — Levels made $5.4K in his first week.
Levels' core philosophy
"The tech stack doesn't matter. Getting customers, getting paid, and iterating fast is all that matters." That's from someone making $132K/month with PHP+jQuery. Instead of learning the latest framework, go get one more customer.
There are risks too
The API dependency is high. Levels has shared experiences where API prices jumped from $3 to $20. GPU costs, model updates, and competitor entry — platform risks are ever-present. Just like Remote OK dropped from $140K/mo to $10K then recovered to $40K, going all-in on a single product is dangerous.
.png)
