Pieter Levels: Photo AI to $100K+/mo with One PHP File
Solo founder behind Nomad List, Remote OK and Photo AI. One large index.php, no employees, still shipping.
- MRR
- $86K–$105K
- ARR
- ~$1.2M+
- Users
- Thousands of active subscribers
- Founded
- 2023
- Team
- Solo
- Stack
- PHP, SQLite, jQuery, Replicate
How did Photo AI start?
I had already been building in public for years with Nomad List and Remote OK. When Stable Diffusion became usable in late 2022 I started experimenting. I first tried generating houses, then interiors. Interior AI showed that people would pay for AI design. Then I trained a model on my own face for fun. It worked better than I expected. People started asking how they could get the same thing.
I did not build a polished product first. I put up a simple HTML page with a Stripe payment link and a Typeform for photo uploads. Early customers paid, uploaded selfies, and for the first few hundred I manually trained models and generated photos. It was chaotic, but it proved demand. Only after the payments started did I automate the pipeline.
Photo AI officially launched in February 2023. The first week brought several thousand dollars. Month two was already in the high twenties of thousands. It kept climbing through 2023 and 2024. At various points the public numbers crossed $100k and even higher months. The exact figure moves with annual plans, churn, and model costs, but the product has remained a core part of the portfolio for years, still run by one person.
Why such a simple stack?
Complexity kills solo founders. Photo AI is essentially a large PHP file with SQLite, some jQuery, and Replicate handling the heavy model training and inference. No React, no microservices, no large team. I can understand the entire system, deploy it myself, and fix it at 2 a.m. if something breaks.
People mock the stack. Revenue does not care about the stack. The goal is to ship features that customers pay for, not to collect technology badges. When a new model or a new capability appears I can test it quickly because there is no large architecture to rewrite.
The same philosophy applies to the rest of the portfolio. Nomad List, Remote OK, Interior AI, and other experiments all run on simple, durable setups that one person can maintain.
What is the hardest part at this scale?
Keeping the product useful while the underlying models change. AI companies will try to absorb everything that is pure software. Features that used to be differentiating can be copied or offered for free by larger players. That is why I am increasingly focused on remaining moats: proprietary data, network effects, audience and distribution, and anything with a physical or real-world component.
Churn and cost management matter too. Training and inference are not free. Margins are still high, but they are lower than pure software products without AI costs. Annual plans help with predictability. Clear pricing and honest communication about what the product can and cannot do reduce support load.
How do you think about the rest of the portfolio?
Photo AI is important, but it is not the only thing. Nomad List and Remote OK still exist and still generate revenue and attention. Interior AI continues. I experiment constantly. Some experiments die. Some become the next meaningful line. The portfolio approach means that if one product slows, others can carry the load.
The audience is the real long-term asset. People who followed the nomad and remote-work journey were ready to try the AI products. The same distribution engine that made Nomad List work still helps every new launch.
What does a typical week look like?
I still write a lot of the code myself. I still post publicly about revenue, experiments, and lessons. I travel. I try to keep the work focused rather than reactive. Support is handled with good documentation and clear expectations. I do not try to build a traditional company structure. The point of the solo model is leverage and freedom, not the appearance of a startup.
When something new appears — a better model, a new use case, a complaint that repeats — I test it quickly. If it works, it stays. If it does not, it is deleted. That rhythm has not changed much since the early days.
Advice for indie founders watching from the outside?
Solve a problem you understand. Charge from day one. Share the journey even when the numbers are small or embarrassing. The audience compounds slowly and then suddenly becomes the main distribution channel.
Do not over-engineer the first version. The first version of Photo AI was almost embarrassingly manual. The payments told me it was worth automating. Most people do the opposite: they build for months in private and then discover nobody wants it.
Keep the stack simple enough that you can still move fast years later. And never assume any single product will last forever. Build the next one while the current one is still working.
More stories
Meta
The Portfolio Pattern: Why Top Indie Makers Run Multiple Products
Product Origin
Marc Lou: The Week ShipFast Changed Everything
Mature
Pieter Levels (Part 2): Nomad List & Remote OK — The Original Audience Engine
Profitable
Jon Yongfook: Bannerbear — Automated Image Generation to Solid ARR
Educator + Builder
Arvid Kahl: From FeedbackPanda Exit to Teaching Bootstrappers
Early Growth
Fernando Mendes: ScanQueue — QR Queues for Real Businesses
Scale
Danny Postma: HeadshotPro — AI Headshots to Six-Figure Months
Scale
Tony Dinh: TypingMind from 5-Day Build to $130K+/mo
Portfolio Scale