4 months of work, groundbreaking AI profile picture app, and an eye-watering $40 revenue
I mostly spent September working on a profile picture AI app that the world hasn't seen yet.
I started working on its first version back in February. Right at the time when the AI avatar hype just died off.
I released it in May -- when even IBM had already launched and shut down an AI avatar app.
On this note, I want to emphasise how important it is to time the launch of your startup. I may give a TED talk on this topic if someone wants to give me such an opportunity.
The first version was bluntly asking user to upload 5 Vogue-grade pictures of themselves, pay $5 up-front, and wait 30 minutes to get the results. Not unique.
I humbly thought of myself as a genius for a different reason though.
1/ It required zero input from user except the Vogue-grade photoshoot. Email and name would be taken from Apple Pay data and respectively payment would be one-click easy too.
2/ I was targeting it at hardcore fans of a very popular anime that in my imagination were to beg me to take their money to turn them into their cartoon idol.
I was kind of wrong on both accounts.
1/ Once the hype was over, it was a B2C equivalent of a suicide to ask users to cherrypick 5 images and then pay before seeing any results.
In all fairness, I somehow convinced 8 people to do that, and raked in a whopping $40 over the course of one month.
2/ I was hoping to target them via Instagram and Reddit Ads as those allow to easily target specific communities. Plus, I wanted to target people specifically with iOS and Safari -- high chance for them to use Apple Pay.
The ads didn't perform as well as my brain painted them to do. Plus, I ended up getting banned for copyright violation.
Plastering the page with the anime name wasn't such a brilliant idea, as it turned out.
Cherry on top -- it was wildly expensive for me to run. According to my Goldman Sachs-grade unit economics, I had to rake in at least 250 paid generations per month to at least break even.
The reason? I rented a dedicated GPU server and wrote very obscure integration with it -- my Node back-end would basically connect to the GPU server via SSH, start the fine-tune + generation, and then take the output images back and clean-up.
Maybe there wasn't a pay-per-second API like Replicate back then? To be honest, I do not know and I do not want to know. At least like this, I have an excuse for why I spent two weeks inventing a wheel. Even if the wheel turned out squared.
After the projection of my CFO didn't fulfil, I shut the app down. And switched to a bunch of very different yet very similar projects.
And in September a brilliant idea struck me. (Modesty's always been my middle name -- only in writing, though.)
I knew how to re-do it, so it works with just one picture (vs 5-10 that such apps usually ask for), runs in ~1 minute (vs 30), and is very cheap for me to run.
I hesitated for a few days as for whether I should spend my time on it, but at the end, I jumped on it as it seemed like a fun bit of work to me.
The bit of work turned, of course, into a month of work... but the fun component was certainly there.
I relaunched it a few days ago and have felt very happy about my technical accomplishments -- cheap, proper B2C (free with watermark, pay to remove the watermark), one-shot, fast.
Technical accomplishments haven't helped me to acquire users yet. I'm still fighting with Reddit for approval of my ads. v2 is also targeted at a niche, but not Naruto anymore.
I love SEO and can do it fairly well, but it's useless for this kind of app. So I'm pushing myself out of comfort zone and trying to figure out paid ads.
Words ≠ actions, as Buddha said.
At the moment, I've started working on Version 3 -- just a few days after the Version 2 release.
Something didn't feel right about V2.
I want to make a few tweaks to the API now, so I can support multiple styles and have better generation quality. All while keeping it fast, cheap, and one-shot. And then pitch folks on Instagram and TikTok.
There's been a new hype with AI Yearbook going on just now, started by an app called EPIK.
The market needs to cool off a bit, but maybe Halloween's going to be a new trend. Or, Christmas. Do you want a picture of yourself turned into red-nosed Rudolph?
The app in question is rosa.lol.