Avatar AI Made $150K in a Week — So Why Did Levels Shut It Down Himself? A Record of Choosing to Abandon a Winning Hand
Pieter Levels' Avatar AI recorded his personal fastest revenue ever — about $150K in one week — in the early days of the AI avatar boom. Yet instead of chasing the boom, he declared it "too cheesy — I want to solve real problems" and shifted his weight to the utility-focused PhotoAI. A rare record of withdrawal in the middle of success.
This is a translation of the Japanese original. The Japanese version and cited primary sources are authoritative for all figures.
Note: yen conversions in this article are rough estimates at ¥150 to the dollar.
The numbers and the sequence of events
| Item | Figure |
|---|---|
| Peak burst | About $150,000 in one week (in his words, “the fastest I’ve ever made money”) |
| Market conditions | VC-funded players like Lensa entered immediately after, reportedly “earning $30M” |
| The decision | Declined to chase the boom; pivoted to PhotoAI for practical use cases |
What this case teaches
“It’s making money” and “it’s worth continuing” are separate questions. Levels’ reason for exiting was not declining revenue but that it was “too cheesy” — he judged that entertainment-style avatar generation would lose novelty and turn into a capital fight. And indeed, the market became an ad war among VC-backed companies. Exiting mid-boom became the best possible running start for the next product (PhotoAI, at $132K/month).
The speed of the exit is itself part of the “4 hits out of 70” portfolio strategy. Fold the misses (or the ill-fated hits) at the same speed you double down on the winners. An indie developer’s only resource is time, and slowness to exit is the largest hidden cost.
Anatomy of a boom: the window where an individual could win lasted weeks
Avatar AI’s $150K came from the few-week vacuum between the discovery that Stable Diffusion fine-tuning could produce AI avatars and the entry of capital-backed companies. Levels sprinted from the technology’s emergence to a shipped product in days, and captured that vacuum whole. Immediately afterward, Lensa entered (VC money, ad budgets, top app-store placement), and the phase where an individual could win an advertising war was over.
In other words, the individual’s only advantage in an AI boom is speed — and that advantage is structurally short-lived. Individuals should build in from the start the plan 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.
The three conditions of a “well-executed exit”
What makes Levels’ withdrawal admirable is that he didn’t hesitate once the conditions aligned: (1) he quit while revenue was still flowing (not a slow-bleed retreat), (2) he articulated the reason for exiting (“too cheesy” = novelty will fade + it becomes a capital fight), and (3) he immediately redeployed the technical assets he’d learned into the next product (PhotoAI).
The build-70-hit-4 portfolio strategy tends to draw attention for its “build a lot” side, but what actually does the work is the speed of discarding. Only those who can let go even of their winners can be first to stand in the next vacuum.
Related cases
Sources
This article summarizes and analyzes the public sources above. Please refer to the primary sources for details.