As part of recording myself everyday until I make a profitable startup challenge, here is the next project:
A website where you can order a printed canvas with AI generated art.
For this one, I’m only to validate it before making the app.
- Step 1: Technical validation:
- Step 2: Market validation
Gonna start by playing around with picture generation A.I to see if I can manage to make great quality art. If so, I’ll then build a landing page and see if people would pay for that.
Was able to generate really cool stuff!
Currently writing the landing page
Here is the result after a day of tryharding:
Been working since 04:30 this morning, and it’s currently 20:40. Not gonna lie, my eyes are burning rn but I need to make the app as fast as possible.
@levelsio just posted the release of his AI portrait app, it’s crazy !!
I just found a tutorial to generate portraits. If I can nail this, it could be insane, I’m 74.82% sure people would love buying portrait of themselves.
Health: 0 – Ambition: 1
As soon as I’ll have the tech part validated, I’ll add Stripe and start looking for pre-orders. I have 0 back end in place yet, only scrappy stuff I found online / made myself. Let’s make it work!
Spoiler: It didn’t worked 🤡
6:12: Today, I’m gonna find a way to make that stuff work.
I found a way to make that stuff work:
Why do I look like a villain tho? 🥲
Been tryharding like crazy today, still feeling a bit sick but it will all be worth it. There is VERY little tutorial as I’m writing this, and most of them had bugs.
Now, I’m not done yet, I need to find a way out of Google Colab, as the memory is very limited and I don’t seem to be able to generate one person more than every few hours, and I won’t be able to automate that later via an API (not a problem for now, I’ll do everything manually at first).
Got one working but it’s not as good, I think probably because it runs on Stable Diffusion 1.4 ?
Here is the plan:
- Find a way to generate custom model in a consistent way
- As soon as it’s done, ship a prototype + update the landing page
- If proto is good (which it will be), focus on distribution channels.
20:51: Still stuck af.
21:40: Found what seems to do an API to fine-tune SD. No fucking way I spent the whole day trying to make Dreambooth works while there was an API available.
Looks like it is actually the case 😭
Well, I’ll see tomorrow if it’s working that well. If it is, wonderful, will help me make a MVP.
It costs $5 per picture tho, which is insanely expensive if I add the product cost. I guess it should be ok just to test the concept when I can’t use Colab.
Gonna get some sleep.
Huh, are we really in November already
But it worked!!
Expansive af but great to test MVP.
Still going to make tests to run it locally, if at noon I don’t have it working, I’ll move on and use what I currently have to start.
Ended up making the landing page anyway. Gonna use Shopify to make things simple.
Noice, so the app I found to make portraits increased the prices by 4, so it’s out of the picture.
That’s why I hate relying on 3rd parties. Anyway, shouldn’t be a problem until I make a shitload of sales, as my own system is working, but I will focus on making the non-portraits for now.
MyAIPainting.com is live!
No back-end atm, I’ll just use Typeform or something.
But first, the social media sharing option:
Sentence sound stupid simple right? But I did not wrote this randomly. Every word communicate the benefits:
- “Create“: It’s something YOU make
- “An unique“: No one else will get it
- “A.I”: It’s a new tech
- “Delivered in two weeks”: It’s something you receive at your home
Added Stripe pages to collect the payment AND…. The app is live!
Going to take a shower and enjoy my sleep! Tomorrow I’ll finally go back to the gym, then I’ll make the form and everything and then I’ll start ads.
Alright gonna make some tests this morning to get the portraits working, then switch to marketing for the current one I have + add automation.
Got 3 different GPU running full speed trying to make it work atm, so much for ecology 🥲
k, got one somewhat working but not really scalable.
Thing is, rn I’m very limited, as the tech is new, so the best thing I can do imo is to gather an audience through a waitlist, and made portraits manually one by one.
That’s what I’m going to do now before automating it. That way I can also make sure to hand pick the best quality for customers in the hundred of pictures I’ll make.
Alright, I think I’m done!
Wow, first 15 minutes break since at least two weeks!
Let’s try ads. Will start on Twitter to see how it goes, since people seems to already be hyped about the whole thing there. Never used it so let’s see how it goes.
Lol, just realized there is no message or anything when people join the waitlist, gonna fix that.
Almost 10PM and haven’t eaten yet, damn
Alright, 6:00AM, time to work
Goal of today: Having ads running by tonight on Twitter.
First gotta install some tracking to measure the ad’s results. Installing Google Tag Manager to group everything.
Now gonna select an audience, will start very broad to see how it goes, then I’ll split test different options.
No way, Twitter allow to target people who look like an account’s followers, that’s cool!
Obviously since those two started the hypetrain, I’ll try that:
Small audience but let’s see how it goes! Gonna put $20/day, a minimum for ads
Didn’t made the same mistake I did on my personal profile, this time I want to make it very clear it’s a physical product.
7:42: Ads are launched, time to take a shower and get a breakfast!
9:27: Facebook & Instagram accounts ready! Ads launching later today.
12:00 Shit results, cut everything 🤡 Will dive back to it later.
Ads are launched on FB, I’m starting with some engagement ads to get people to like the page to build credibility.
Both Pieter and Danny launched in producthunt, I’m def going to do it for the meme.
I guess I have 24 hours left to make the whole PH description, including training a model with better pictures of me so people can get a better preview.
Wrote a description and a mini-description using a very ironic tone, hopefully people will get it. Not really looking to make sales on PH anyway, will mainly allow me to see how it works
Now, time to prepare the pictures. I need to train a model with better pictures of me (I look grumpy af on the other ones).
Pictures: Taken, now let’s train the model.
Now the logo. It’s crazy that most people don’t use a .gif on PH.
THIS IS LITERALLY FREE ADVERTISING SPACE PEOPLE!!!!
Alright, Product Hunt page is done.
Not really expecting anything, I did it for the challenge, but I hope people will find it funny!
Busy day today, gotta make the form, get a haircut, launch on product hunt, run ads on Facebook and then meet my friend who arrived in Bali.
💡 Expert tip: Do not book a barber appointment the day of your Product Hunt launch
Alright, I think everything is ready. I can finally eat
Somehow the Stripe coupons are not working and I’m launching in less than one hour, nice.
Someone just subscribed to the waitlist before the launch, cool! I have no automation in place to answer email so I guess I’ll do it manually!
Not like if I’ll get dozens of people anyway.
Been very intense 24 hours but I think everything is ready!
That’s funny, I stayed in position #6 for almost one hour
Will go down tonight but it’s ok, was fun!
Now I can finally relax a bit
22:40: Nevermind, launching ads, I’ll relax later.
Ended up with 75 votes haha, that’s pretty cool! Not that I really care about the position, I mainly did it for fun, but it still give me some social proof with the nice badge!
Today going to launch ads and the funnel.
Alright, so, turns out, the ads results are absolute garbage. I’ve actually rarely seen results that bad. I thought I could use a channel I know to promote a proven app, but it did not worked out as planned.
- Total spent in ads: $103.99
- Sales: 0
- Waitlist: 18
There are likely a few reasons for this:
- Advertising costs are incredibly high at the moment. That’s because it’s before christmas, Black Friday etc. To give an idea, it was costing between $40 and $60 to reach 1000 people.
- The product is extremely broad so it’s hard to target an audience and craft a good message for them.
- The general public (Facebook) is not early adopters. They need something already well setup before buying, which was not the case here, as I have no social proof at all.
- Sometimes, you just get unlucky
The last one is probably not going to be popular, but unfortunately, a big part of entrepreneurship is luck. Luck is different from randomness though. For me, luck is when opportunity meet preparation.
Back then, I know two guys who made tests by launching the exact same eCommerce products at the exact same time with the same marketing. Every time, one worked and one did not.
Here, I also had limited preparation, so it did not helped either.
I’m happy I did it tho, I managed to launch on ProductHunt and get almost 80 votes:
That’s huge for something I made in 24 hours, without preparation.
I was trying to find “lessons” I learned with this failure, but honestly, I can’t find any. As Dwight (The Office) says it, not everything is a lesson, sometimes you just fail.
I guess it’s better to fail after one week than after spending five months making the product tho.
Not gonna lie, I got a bit depressed for a day or two (which made me write this, on how to move on fast after failure), but that’s part of the job I guess.
I have 4 other projects in the pipeline at the moment. 3 are not public yet, the 4th one is an app to restore old pictures. Competitive af but I saw someone on Twitter asking for it so I made it in 3 hours and got the sale (story here). Gonna do some marketing next week.
I guess that’s the end of this article!
I’ll let the the website alive until January 2023, in case people want portraits for christmas, then I’ll see if I shut it down or try to re-promote it with cheaper ads.