In short
- It didn't launch from zero — it launched from an existing paying audience. Photo AI was built as an evolution of Pieter Levels' earlier product AvatarAI.me, which 'still runs/sells really well' — so day one had a warm user base and revenue stream most launches spend years earning.
- 5 months to $61.8K/mo, no ad spend, no team. By July 2023, Photo AI had 1,872 paying customers on a codebase that was still under 14,000 lines of raw PHP with no frameworks.
- A single feature ($29/mo try-on mode) opened an entire new buyer segment. Adding AI clothing try-on in June 2023 repositioned the product as a 30x-cheaper alternative to a Shopify product photo shoot — without touching the core mechanism.
- In 2026, 'AI made moats disappear' became the reason to ship more, not less. Levels used the same AI tools that commoditized his product to vibecode a full video editor into it in days — treating faster building as the response to easier copying, not a threat to route around.
The mechanism
Photo AI's growth chain starts with a shortcut most teardowns don't get to assume: it launched into an audience and revenue base a prior product had already built, then let its own shareable output do the advertising a normal consumer app has to buy. The one deliberate lever Levels pulled — the $29/mo Shopify try-on mode — didn't create a new channel, it repositioned the existing engine for a buyer with more money and a clearer reason to pay, which is most of the gap between $61.8K/mo (2023) and $105K/mo (2026).
Launching as a sequel, not a startup, skips the cold-start problem entirely
Photo AI wasn't a blank-slate launch. Pieter Levels positioned it explicitly as an evolution of his prior product AvatarAI.me, which he says 'still runs/sells really well' at launch time. That meant Photo AI inherited an audience already primed to pay for AI-generated photos of themselves, plus a founder-brand (Levels' X following) that would amplify the announcement — the two hardest things a new consumer app has to build from nothing.
Photo AI is an evolution of my previous project AvatarAI.me which still runs/sells really well. But I really wanted to double down on high quality photorealism instead of just avatars.
The product's output is inherently shareable, so usage doubles as promotion
Photo AI generates photorealistic images and videos of the person using it — content people naturally want to post, send, or show off. Unlike a backend tool or dashboard, every output is a piece of content the user is motivated to distribute themselves, without Photo AI having to build a referral mechanic or ask for shares. That built-in incentive to spread the output is the same class of loop that drove revenue from launch through the 2023 milestone.
Once you like a result, you can [ ♻️ Remix ] it to get more photos like it
One narrow feature reframed the whole product for a segment that pays more
In June 2023, adding an AI clothing try-on mode wasn't a random feature — it retargeted the product at Shopify sellers who need model photos and would otherwise pay $1,000s for a shoot. Pricing it at $29/mo (explicitly marketed as '30x cheaper') let the same core photorealism engine serve a commercial buyer with a much clearer willingness to pay than a consumer wanting personal photos, widening the revenue base without a new product.
Great for @Shopify sellers who have products and need model shots but don't want to spend $1000s (only $29/mo, so at least 30x cheaper)
How it went
Launch as a sequel to an existing product
2023-02 → 2023-06After roughly two months of building, Pieter Levels launched Photo AI in February 2023 as a photorealism-focused evolution of his earlier product AvatarAI.me, which he says was still generating revenue at launch time [S1]. The product let users upload photos, train a personal AI model, and generate photo shoots in any setting [S1] — inheriting rather than building an initial audience.
Feature expansion opens a commercial buyer segment
2023-06 → 2023-07In June 2023, Photo AI added an AI clothing try-on mode explicitly aimed at Shopify sellers needing product model shots, priced at $29/mo and marketed as at least 30x cheaper than a traditional shoot [S3]. One month later, Levels reported 1,872 paying customers generating $61,808/mo — the first public revenue milestone, on a codebase still under 14,000 lines of raw PHP [S2].
Multi-year quiet growth to $105K/mo
2023-07 → 2026-03Between the July 2023 milestone post and the next public update in March 2026, Levels published no revenue figures for Photo AI in the archived sources here — a multi-year gap consistent with his broader pattern of posting milestones only occasionally rather than on a fixed schedule. By March 2026 he reported $105,000/mo revenue and $80,000/mo profit (~76% margin) on a codebase that had grown to 40,870 lines in the same single index.php file [S4].
Using AI-shipping speed to add a video editor
2026-03 → 2026-08In August 2026, framing 'AGI destroying moats' as a reason to ship bigger rather than retreat, Levels vibecoded a full video editor into Photo AI — letting users generate and edit short videos using their trained AI models, including a planned agent sidebar that would let users describe a short film in plain language and have it scripted and generated automatically [S5].
Milestones
- 2023-02Launch — evolution of AvatarAI.me, photorealistic AI photo studio
- 2023-06AI clothing try-on mode added, $29/mo, targeting Shopify sellers
- 2023-07618081,872 paying customers, $61,808/mo revenue, ~14,000-line PHP codebase
- 2026-03105000$105,000/mo revenue, $80,000/mo profit, 40,870-line PHP codebase
- 2026-08105000Vibecoded video editor with AI-generated video and model likeness added
Whether it fits you
This playbook depends on a couple of structural advantages that most first-time founders won't have on day one. Run it if these hold for you; the tradeoffs below are real costs, not hypothetical ones.
What it needs
You already have a warm audience or predecessor product to launch into
Photo AI's cold-start problem was solved before it existed — AvatarAI.me supplied users and revenue, and Levels' personal following supplied distribution. Run it if you have an existing audience, waitlist, or sibling product to launch from; without one, expect this loop to start much slower.
Your product's output is something people want to share on its own
A photorealistic photo or video of yourself is inherently postable content — no incentive design required. This works when the output is personally flattering or useful enough that users want to distribute it themselves; it doesn't transfer to products whose output has no audience beyond the user.
You're comfortable running the whole thing solo, in one codebase, indefinitely
The entire product has run as a single index.php file (14,000 lines in 2023, 40,870 by 2026) with one founder, no framework, and no team. Run it if you're willing to accept the maintenance and scaling ceiling that implies in exchange for speed and full control.
What it costs
No visible playbook for the actual acquisition math
Levels' public posts share revenue and profit milestones but never publish CAC, conversion rate, or churn — so this teardown can describe the mechanism (audience carryover, shareable output, feature-led segment expansion) but can't verify how efficiently it converts to paying customers.
A single-file, no-framework codebase is a deliberate technical debt bet
Going from 14,000 to 40,870 lines of raw PHP in one file over three years is a bet that founder velocity matters more than maintainability. It works for a solo operator who wrote every line, but it doesn't transfer to a team, and it caps how fast new hires (if any) could contribute.
The mechanism depends on already owning distribution most founders don't have
The launch-into-an-existing-audience move is the single biggest lever here, and it's also the least repeatable part: it requires either a prior product with real usage or a founder audience large enough to make announcement posts do real work. Without one of those, expect the same tactics to produce a much slower ramp.
The numbers we could verify
- profit margin
- ~76% ($80K profit / $105K revenue, Mar 2026)
- try on pricing
- $29/mo, marketed as ~30x cheaper than a professional model shoot
- revenue growth
- $61.8K/mo (Jul 2023) to $105K/mo (Mar 2026) — no interim figures published in these sources
Channels it actually used
- existing audience carryover from a prior product (AvatarAI.me)
- X/Twitter build-in-public milestone posts
- output as its own advertisement (shareable photorealistic images)
- feature expansion into adjacent buyer segments (Shopify sellers)
Our read
The popular read on Pieter Levels' post-AI launches credits raw shipping speed or AI hype — but Photo AI's actual advantage at launch was inherited distribution: an existing paying audience from AvatarAI.me, not a clever new acquisition channel. Founders without a predecessor product or following should expect a much slower version of this same playbook.
The ~2.5-year gap between the 2023 and 2026 revenue posts in these sources is worth naming directly: it means this teardown's growth curve is a start-and-end snapshot, not a continuous timeline. The mechanism (audience carryover, shareable output, segment expansion) is well evidenced at both ends, but the specific path of MRR growth in between is not documented here.
Levels' framing of AI as removing moats — and his response being to ship more complex features rather than defend the existing one — reads less like a general growth tactic and more like a founder-specific bet: that continuously out-building competitors on features is more durable for him than trying to protect a static product from being copied.
Sources
- My new project Photo AI a photorealistic AI photo studio2023-02-10
- Photo AI is almost 14,000 lines of raw PHP making $61,808 per month2023-07-03
- Photo AI adds try on mode for AI clothing models2023-06-14
- Photoai.com is a 40,870 line index.php making 105,000/mo revenue and 80,000/mo profit2026-03-06
- I vibecoded a full video editor into Photo AI that generates and edits videos with your trained models2026-08-01
