So last year @Wayfair, an $8B company, launched their own AI interior design tool

This week someone at Wayfair asked me to cancel their subscription of , which is my AI interior design tool launched 2 years ago

I was confused why would they sign up to it if they have their own? Well, because they signed up to it before they had their own and used it to build their version of it! 😂

They had 8 different accounts on Interior AI that ran for months 🤯

I learnt a bit about competitive spying today!

Told ya the big companies were checking what indie hackers did, that was my entire point when I said "indie hacking is dead", we're part of the big market now where billion dollar companies are INSIDE our apps!InteriorAI.com

If I did competitive spying I'd at least use a different domain name though, not my own company @wayfair.com

Any statistical way to solve this?

I test parameters on Photo AI on a % of photos taken

Then I check what % of photos are ❤️ favorited

Based on that I can figure out what parameters is better on 700,000 photos taken per month

But sometimes you get 1 or 2 users ❤️ favoriting lots of photos with one specific parameter which affects the data and makes one parameter choice come out as a winner when actually it's just a few users doing it

Any stat solution for this? Right now my SQL is smth like this

SELECT upscale_strength,count(upscale_strength) as cnt FROM photos WHERE upscale_strength IS NOT NULL AND (saved IS NOT NULL) GROUP BY upscale_strength ORDER BY cnt DESC LIMIT 10

Biggest benefit you have starting out in anything is being a noob and not knowing anything

If you’d share that true honest story it’s super authentic and people love to hear that

But nobody does

Instead I see people on social media faking being succesful with 100 followers I never understand why people on here aren’t being more radically honest and transparent about their life and work and business situation

Easiest way to get people interested in you because EVERYBODY is still too scared to do it

Way more engagement than being another threadboi… twitter.com/i/web/status/1…

Only losers go to networking events

The only people you'll meet at events are people who want something from you or aren't meaningfully engaged in a pursuit of their own

The easiest way to meet interesting people is by having accomplished something yourself

https://twitter.com/jonclemonsjp/status/1657611954033397760

All the interesting and unique people I met are after I accomplished something myself and kept doing so or trying to repeatedly

Millionaires and then billionaires will message you if you do interesting stuff

And they become your friends

You won’t meet them at networking events

I see a big shift in perception of Dubai among me and my friends vs years ago

Years ago: ehh weird place for ppl with no taste born into money driving sports cars

Now: everywhere else it’s getting harder for startups to do business, while Dubai is making it easier, maybe nice

https://twitter.com/balajis/status/1657619914855956482

And inb4 it’s about tax, surely it is but more so I think is how easy it is to do business which means easy bookkeeping and more time to focus on product/biz

See also crypto companies moving there

Not saying it’s all roses but interesting

Does Midjourney have an API yet? Cause it’s obviously way superior to Stable Diffusion now

Hard for me to sell products built on tech if there’s obviously superior tech out now

SD2 is worse than SD1.5 and both don’t come close to Midjourney 5 and 5.1

✅ $7,000 MRR w/ 225 customers on photoai.io

50,000+ photos taken now New feature I thought of today: when I have a better base model I can retrain people's models on it and auto upgrade them so your model will keep looking better every month

So @TheEconomist won't let you cancel your subscription, they force you to contact them

Thought this was illegal by now in US and EU right?

So shameful Holy shit they really won't let me cancel

People making AI startups seem to be stumbling over each trying to add every single new AI tech/feature/paper that comes out into their app first thinking it'll make a massive difference for customers

But not sure if customers even notice so much, they just want a problem solved Like ControlNet is obviously a breakthrough for AI imaging but for customers it's an incremental feature update that makes some images a bit better

Portugal ends Golden Visas (GV)

"attracted 6.8B euros in investment since its launch in 2012"

6.8B / 11Y = 618M/y
Min inv = 280k
618M / 280k = 2207 GV houses sold/y

Avg amt sold/y = 150k
foreignbuyerswatch.com/2021/05/11/tot…

Golden visa share = 2k/150k = only 1%

reuters.com/markets/europe… Golden Visa is the investment visa (you invest min EUR 280k into real estate or funds in Portugal)

Not related to the NHR which is used to attract foreigners to Portugal and still active

My point is it seems disconnected from reality, hard to see how 1% has effect on market

Just in time, my new project!

📸 PhotoAI.io

I spent the last 2 months w/ 🩸blood, 🥵 sweat and 🥲 tears building a photorealistic AI Photo studio and it finally works:

✨ Upload photos
🧠 Train model of person
🐈 Copycat any photo
📸 Create a photo shoot anywhere

https://twitter.com/levelsio/status/1623612426448158722

📸 PhotoAI.io is an evolution of my previous project 👩‍🎤 AvatarAI.me which still runs/sells really well

But I really wanted to double down on high quality photorealism instead of just avatars, which means the tech is very different!

Here's my model:

Pics of my first FULLY self trained Stable Diffusion (not Dreambooth) with my own data set

These are just samples as it's not even fully done training

Will train on men next! Then a combo model Few more I liked:

A lot more human than before, more imperfections = good

✨ Made $300,000+ with AI in the last 2.5 months w/ avatarAI.me + interiorAI.com

Started 77 days ago!

🦾🤝😊 Thanks robots Haven't had this much fun coding since 2014-2016 Nomad List + Remote OK days, everything is new, I have no idea what I am doing, perfect life

Figured out FS auto pilot so now I can code index.php while the ✈️ plane flies itself in IFR #automation Also, finally got CONTRAILS!

Actions taken now from mistakes learnt:
✅ set up my own GPU
✅ running ads from Monday
✅ celeb outreach started
✅ talking to acquire iOS app

Let's try + learn! ☺️

https://twitter.com/levelsio/status/1600863365181583360

me irl on the gpu vps

@marckohlbrugge @dannypostmaa They out executed @dannypostmaa and me really well, lessons:
- we built on top of APIs, limited by resources and pricing, they seem to went straight to running their own GPU servers, so less limits and lower cost
= don't build on APIs

1/n @marckohlbrugge @dannypostmaa - because they presumably run their own GPUs, they could go 15x cheaper, we were paying $20 per training when they launched, and sold for $30-$40, they sold for $2
- real cost of a training is probably $0.50 (and API providers moving to that fast now)
= again don't build on APIs

I'm getting obsessed with posting all my AI fails, here's my robot extending avatarai.me's 512x512 pic of @jelmerdeboer into a photograph and it adds a random guy drowning After 8 tries, it does this though, very nice!

Here we go....Why did this change?

It neuters all previously useful Customer objects to useless Guests for no reason and by default?

Another crazy Stripe default making our lives more difficult 😭 Maybe Elon can clean house?

https://twitter.com/levelsio/status/1600314461021122560

I always feel bad tweeting stuff like this about Stripe, because I have so many friends there. But if I stop tweeting about it just because my friends are there I'd feel it'd be even worse and dishonest and nepotist. So I'll keep tweeting about it

✨ Outpainting avatars to entire photographs is now fully automated and live

🖼 I made a visual explainer on the landing page to show how it works

avatarai.me What I really need is an AI model on @replicate to judge an image like a human does:
- do the arms, legs, hands, face etc look weird?
- what's the best pic in a collection?

Right now outpaint results can be unpredictable so I generate a few pics so users can pick best

Tried upgrading Stable Diffusion from 1.5 to 2.0 for interiorai.com

Left: SD 1.5

Right: SD 2.0

Maybe no 😂 Tried both

stable-diffusion-512-v2-0
stable-diffusion-768-v2-0

Both same bad results

Using Stability AI's API to gen with init_image

Managed to fix the hands problem So my fix is:

Auto detect hand position X,Y and size

Then find a photo similar to those hands (could automate this)

Put it in there and paint/mask it in there (also could automate this)

This seems a quicker fix then figuring out the hands model for the short term