One Person, One 40,000-Line index.php: Pieter Levels' AI Photo Studio Photo AI Hit $161K a Month After a Podcast and Still Makes $86K a Month Three Years On
In October 2022, indie developer Pieter Levels launched AvatarAI.me, a small AI avatar tool that made 2,943 sales and $100,000 in 10 days. Soon well-funded companies with iOS skills built similar apps, so he moved to something harder: photorealistic AI photo shoots. On February 10, 2023, after two months of work, he launched Photo AI: upload your photos to train a model, then generate photos of yourself in any setting. It's a website written in PHP, jQuery and SQLite, sold as a monthly subscription, with AI inference running on Replicate. By his own published numbers, it made $61,808 a month with 1,872 paying customers in July 2023; after his September 2024 appearance on the Lex Fridman podcast, Photo AI briefly hit $161K a month, though he said that lasted about one day; in March 2026 it was a 40,870-line index.php making $105,000 a month in revenue and $80,000 in profit; in October 2026, his X bio says $86K a month. He describes himself as one founder with zero employees, using automation scripts and AI instead of staff.
Process
If you follow indie developers on X, you've probably seen Pieter Levels (@levelsio) posting how much a product made this month, what tech it runs on, or what he just launched. He's Dutch, has lived as a digital nomad around the world for years, and has built dozens of websites by himself, from the digital nomad community Nomads.com to the remote job board RemoteOK.
Photo AI, an AI photo studio, is one of his most profitable products of recent years. It has no funding and no employees, and the whole site is a single PHP file of more than 40,000 lines. All the numbers below come from his own public posts and blog.
Stage 1: From an Electronic Music Channel to "12 Startups in 12 Months"
Pieter didn't study computer science. On Lex Fridman's podcast he said, "I started in business school." Early on he ran an electronic music YouTube channel, Panda Mix Show, posting dubstep and drum and bass mixes, which earned about $2,000 a month from ads. Building websites for the channel taught him a bit of coding.
In 2014 he launched a public challenge: 12 startups in 12 months. Many failed, but one became Nomad List (now Nomads.com). Ever since, he has built websites with the simplest tools: PHP, jQuery and an SQLite database.
Stage 2: AvatarAI, $100,000 in 10 Days
In 2022, the AI image model Stable Diffusion took off. That October, Pieter launched a small tool, AvatarAI.me: upload your photos and AI generates avatars in all kinds of styles.
It exploded. On November 7, 2022, he posted that in 10 days since launch it had made 2,943 sales and $100,000, averaging $33 a sale.
It didn't last. On the podcast, he recalled that companies like Lensa, "much better at iOS and stuff than me," quickly built the same thing as an app; "I think they made like $30 million with it," becoming the top-grossing app. He admitted he was a little sad: "I never had real fierce competition."
Something else happened around then. He always tweets how much his websites make. The GPU platform he used for training saw that and raised its training price from $3 to $20, while he was selling for about $30. "Immediately my profit is gone." He later moved his AI inference to Replicate.
Stage 3: Two Months to Build a Photorealistic Photo AI
With avatars too crowded, Pieter decided to do something harder: photorealistic photo shoots, not cartoon avatars but images that look like real photography.
By his account, building a photo studio was the idea of his friend and fellow indie developer Danny Postma, and the two talked about it for a month. The project was so complex that he kept procrastinating; his friend Marc Köhlbrugge pushed him to ship.
On February 10, 2023, Photo AI launched. He wrote that he had "spent the last 2 months w/ blood, sweat and tears building" it. Here's how it works: upload a few photos of yourself, train your own model, then type a description, such as wearing a black hoodie, working in a coworking space at golden hour, and it generates photos of you in that scene; or upload a reference photo for the AI to copy.
Unlike AvatarAI's one-off payments, Photo AI was a monthly subscription from the start. In August 2024 the plans were $19, $39, $99 and $299 a month; today they're $19, $49, $99 and $199.
Stage 4: One Person, Revenue in Public
Almost every step of Photo AI's growth has been posted publicly on X.
- In March 2023, photoai.com's banner showed about $28.7K MRR (web archive).
- On July 3, 2023, he posted that Photo AI was almost 14,000 lines of raw PHP mixed with inline HTML, CSS and JavaScript: "I did not use TS, flexbox or frameworks except jQuery." It had 1,872 paying customers making $61,808 per month. The post got nearly 5 million views.
- In February 2024, the site showed about $64K MRR (web archive).
As for marketing, he hasn't spent zero. In September 2023 he publicly shared a tactic: search your industry's hashtag on TikTok, list 30 creators whose videos get 100K+ views, and pay each $100 to $300 to post a video about your product.
Stage 5: One Podcast, Triple the Revenue
In August 2024, Pieter appeared on episode 440 of the Lex Fridman podcast for more than three hours. On air, he demoed Photo AI: he had pulled 10 or 20 photos of Lex from Google Images to train a model, and after typing a prompt it generated a set of photos in about a minute.
The effect was immediate. On September 22, 2024, he posted that his products together had hit a record $420,000 a month at about 80% profit, with Photo AI alone at $161K a month. The biggest cost was GPUs, about $60K a month for Photo AI and InteriorAI combined.
He gave the podcast full credit: total revenue went up 2.5x after the episode, with Photo AI up 3x. By comparison, he said, going on national TV gets him "like 3 sign ups."
He was also clear that the record lasted "like ONE DAY" before dropping way down.
Stage 6: A 40,000-Line index.php and a One-Person Company
On March 6, 2026, Pieter posted that photoai.com is a 40,870-line file called index.php, making $105,000 a month in revenue and $80,000 in profit. By October 2026, his X bio lists Photo AI at $86K a month.
The whole company is him. In November 2024 he shared his "minimal solo founder setup": one founder who is also developer, designer and marketer (with ChatGPT's help); zero humans; scheduled scripts that call the ChatGPT API to do work humans would normally do, like moderation and sales; and a self-service dashboard where users change plans, cancel and even refund themselves.
In September 2026, he said his businesses together were steady at $200K–$250K a month, and that he keeps pushing costs down and margins up. He also mentioned that in 2023, on Danny Postma's advice, he invested in Replicate and FAL, the AI inference platforms his photo apps rely on.
Thinking
1. The last product is market research for the next one. AvatarAI making $100K in 10 days proved that many people will pay for AI-generated photos of themselves. Photo AI wasn't conjured from nothing; it took a validated demand one step further, toward something harder and more valuable.
2. Easy hits get copied fast. AI avatars were too easy to build, and well-funded companies better at apps caught up in weeks. Pieter's answer was technically harder photorealistic shoots, plus a shift from one-off payments to subscriptions for steadier revenue.
3. Relying on one vendor is risky. When the GPU platform saw his revenue and raised prices from $3 to $20, his profit vanished overnight. For AI products, inference costs and vendor choice decide margins, so be ready to switch.
4. Public revenue is marketing in itself. Every revenue tweet is a free ad. The Lex Fridman episode drove the point home: one in-depth interview brought far more users than national TV.
5. Pick tech one person can maintain. PHP, jQuery and SQLite look dated, but one person can understand and change the whole system. That lets him avoid hiring and keep the profit.
6. A peak isn't the norm. $161K lasted one day; today it's $86K. Exposure creates short spikes, but long-term revenue depends on how many users the product keeps.
Action
1. Test demand with a small tool first. Don't start with a big, all-in-one product. Like AvatarAI, build the fastest thing that can take money and see if people pay.
2. Once demand is proven, move up to something harder. If a small tool sells but is easy to copy, think about a harder, better version, and consider subscriptions for steadier revenue.
3. Share your progress and numbers publicly. Post on X or elsewhere what you built, what you earned and what you learned. Real numbers draw more attention than ads.
4. Go after podcast interviews. Find podcasts your target users listen to and try to get on, ideally demoing the product live.
5. Try paying creators on a small budget. Borrow his tactic: find 30 creators in your niche with 100K+ video views, pay each $100–$300 for a video, and see which channels actually bring users.
6. Hand repetitive work to scripts and self-service. Moderation, follow-ups, refunds, plan changes: automate whatever you can, and spend your time on product and growth.