07 — Product · the failure/Shut down
InstantArt.io
Multi-model image generation, free at the front door. The growth was real and so was the GPU bill, and only one of them was survivable.
- Users
- 1M
- Time to get there
- 3 months
- Social reach
- 1M+ views
- Funding raised
- None
Role
Founder, engineering, growth, support
Stack
- Stable Diffusion
- multi-model
- consumer web
A million users in three months, then I turned it off
What it was
Image generation across several models in one place, aimed at people who were never going to install anything. No local setup, no model wrangling — type, generate, share.
The growth
It hit a million users in three months. On the social side the campaigns cleared a million views in reach, and that traffic converted straight into sign-ups, because the product needed no explanation and no credit card to try.
That part worked exactly as designed. It was the only part that did.
Why it closed
Every free generation ran on a GPU somebody had to pay for, and that somebody was me. The compute bill scaled with the success and the budget did not. I burned through what I had on inference.
I went looking for funding to bridge it and did not land a round. With no runway and a cost curve pointing up, I shut it down rather than degrade it into something users would resent.
What I took from it
Distribution was never my weak point. Unit economics were. A free tier on generative inference is a marketing budget that grows fastest exactly when you are least able to pay it, and a crowd that arrived because it was free is not the same thing as a business.
That lesson is why everything I have built since runs on hardware the user already owns. Local inference does not send me a bill at three in the morning.
