Avatar AI Made $150K in a Week — So Why Did Levels Shut It Down Himself? A Record of Walking Away from a Winning Hand

Note: yen conversions in this article are rough estimates at $1 = 150 yen.

Archives of success stories almost never contain the record of someone who “shut it down while it was still making money.” Exits are usually narrated only after the revenue runs dry. Pieter Levels’s Avatar AI is the exception. About $150,000 (roughly 22.5 million yen) in one week — in the very middle of what he himself called “the fastest money I’ve ever made,” Levels retired the product from his main lineup and shifted his weight to the utility-focused PhotoAI. In hindsight, that decision produced a successor earning $132K per month, but what deserves attention is less the outcome than how the decision to step off at the peak was constructed.

The Numbers and the Timeline

Avatar AI’s revenue was one-time purchases, not monthly subscriptions. The $150K was a one-off spike, not compounding MRR. That structural difference is the key to understanding the decision that followed.

“Me as a Barbie Doll” — Putting the Exit Reason into Words

Levels did not stop Avatar AI because revenue deteriorated. His own words: “It was too cheap. Kitschy, like seeing myself as a Barbie doll.” The remark sounds impressionistic, but unpacked, it contains two cool-headed observations.

First, entertainment-style avatar generation is content that gets consumed and then discarded, play with it once and you’re bored. The very fact that one-time payments produced $150K is evidence that people were paying for novelty, not for an ongoing need. Second, novelty markets turn into contests of capital. Sure enough, VC-backed companies including Lensa entered right afterward and the market became an ad-spend war. “It’s making money” and “it’s worth continuing” are separate questions, and Levels answered “no” to the latter while the former was at its all-time high. Quitting in the middle of the boom became the perfect runway for the next product (PhotoAI at $132K/month).

Anatomy of a Boom: The Window Where an Individual Could Win Lasted Weeks

Avatar AI earned its $150K in the few-week gap between the moment it became clear that Stable Diffusion fine-tuning could produce AI avatars, and the moment capital-rich companies moved in. Levels sprinted from the technology’s appearance to a shipped product in days, and captured that entire gap. Immediately afterward, Lensa (VC money, ad budgets, top App Store placement) arrived, and the phase where an individual could win an advertising war was over.

What powered that speed was a surprisingly boring tech stack: PHP, jQuery, SQLite, not fashionable frameworks, but the tools he could write fastest. Deploys finish in seconds. AI processing is outsourced to Replicate. The upshot: an individual’s only advantage in an AI boom is speed, and that advantage is structurally short-lived. Individuals should plan from the start to “earn everything in the boom’s first few weeks, then cede the field when capital arrives”. Avatar AI reads as the blueprint for exactly that.

Three Conditions of a “Clean Exit”

What makes Levels’s exit remarkable is that he did not hesitate once the conditions lined up. (1) He quit while revenue was still flowing (not a slow bleed-out), (2) he articulated the reason for quitting (“too cheap” = users will get bored + it becomes a capital war), (3) he immediately redeployed the technical assets he had learned into the next thing (PhotoAI). Indeed, PhotoAI experiments began in October 2022 and the official launch came in February 2023, because he transplanted the fine-tuning technology already validated in Avatar AI onto a practical problem (“headshots and profile photos”), PhotoAI was nearly profitable within days of launch.

The post-transplant numbers vindicate the exit decision. PhotoAI generates over 1 million images per month. The Pro plan is $29/month for 1,000 photos. Roughly half of its traffic comes from organic search, an acquisition structure that stays out of advertising wars. In quality of revenue, it is the exact inverse of the one-time, hype-dependent Avatar AI.

Levels is public about his own hit rate: “I’ve built more than 70 things, and only 4 made money and grew.” Portfolio strategy tends to get attention for the “build a lot” side, but what actually does the work is the speed of discarding. Only someone who can let go even of a winner can be first to stand in the next gap.

Before We Romanticize This Decision

Some reservations are in order. First, walking away from $150K/week looks correct in hindsight, but Levels had no certainty at the time. He had watched another business, RemoteOK, crater from $140K/month to $10K during the pandemic. His distrust of momentary spikes plausibly comes from that scar (RemoteOK later recovered to $41K/month). This exit instinct is the learning of someone who has been punched hard once. There is no guarantee anyone could make the same call on their first success.

Second, the exit became a “runway” only because the successor, PhotoAI, happened to hit, and this is a world of 4 hits in 70 attempts. Statistically, the normal outcome after quitting is that the next thing does not hit either. Third, Levels’s ability to ship a product in days and his existing audience of readers and users are the accumulation of many years. The very speed to seize a boom’s gap is itself a barrier to entry, and should be discounted accordingly.

Conditions for Reproducing This

What Japanese indie developers can take home from this case is not the flashy numbers but the design of the decision. What generalizes: (a) distinguish one-time sales from MRR, and never confuse a momentary spike with business value. (B) hold exit criteria not as “feelings” but as articulated conditions (signs of user fatigue, capital entering the market). (C) build with the assumption that when you fold, the technology and knowledge transfer to the next thing. What does not generalize: the ability to ship a product in days, and an existing distribution channel to deliver it through instantly. The few-week gap of a boom opens for everyone, but only those whose preparation is already complete can stand in it, and if the preparation is missing, “don’t chase the boom” is a conclusion drawn from the same blueprint.

Related Cases

• PhotoAI — the successor that implemented Avatar AI’s lessons
• The Neuron — the “sell in the middle of the boom” version of stepping off