How to make $150,000 in a week - story of a viral AI website | Pieter Levels and Lex Fridman

Summary

The speaker recounts their journey with generative image models, starting with Stable Fusion, which they used to create a website generating non-existing houses. They then pivoted to interior design with Interior AI, achieving significant financial success. Experimenting with image-to-image technology, they trained models on their own face, leading to the creation of Avatar AI. Despite competition from larger companies, they found their niche in photorealistic AI, culminating in Photo AI. The narrative highlights the challenges and triumphs of leveraging AI for creative and business purposes.

Takeaways

• 😀 The speaker experimented with generative image models, specifically Stable Diffusion, to create images from text prompts.
• 🏠 They discovered that the model excelled at generating images of houses and architecture, leading to the creation of 'thishousedoesnotexist.org'.
• 🔄 The speaker pivoted to interior design, launching 'InteriorAI.com', which utilized image-to-image technology to modify existing photos into different styles.
• 💸 The venture into interior design proved profitable, generating significant monthly revenue within a short period.
• 🤖 The speaker explored fine-tuning AI models to specialize in interior design, which improved the quality and realism of the generated images.
• 👤 They also experimented with generating avatars of themselves, which unexpectedly became popular and led to the creation of 'AvatarAI.me'.
• 📈 The avatar project went viral, earning a substantial amount of money in a short time, but the speaker felt it was too 'cheesy' and not sustainable.
• 📸 The speaker then shifted focus to creating a more realistic photo AI, aiming to produce photorealistic images without the need for a photographer or studio.
• 🛠️ The process of training the AI on a limited set of photos and manually curating the results was labor-intensive, prompting the need for automation.
• 🔗 The speaker collaborated with AI model platforms to automate the training process, which eventually led to the development of more specialized AI models for various applications.

Q & A

What is the first generative image model the speaker started playing with?

The first generative image model the speaker started playing with is Stable Diffusion.

How did the speaker discover Stable Diffusion?

The speaker discovered Stable Diffusion on Twitter.

What was the speaker's initial experience with generating images using Stable Diffusion?

The speaker found it amazing and was able to generate images in any style by writing prompts.

What was the speaker's observation about the model's ability to generate images of people?

The speaker noted that the model was really bad at generating images of people.

What type of images did the model excel at generating according to the speaker?

The model was good at generating images of houses and architecture.

What website did the speaker create to showcase the model's ability to generate house images?

The speaker created a website called 'thishousedoesnotexist.org' to showcase the model's ability to generate house images.

How did the speaker monetize the interior design AI model?

The speaker monetized the interior design AI model by launching a website and charging users for generating interior designs.

What was the monthly revenue the speaker achieved with the interior design AI model?

The speaker achieved a monthly revenue of 40K to 50K dollars with the interior design AI model.

What did the speaker do to improve the interior design AI model?

The speaker fine-tuned the existing AI model on a specific goal by training it with a gallery of beautiful interior designs.

How did the speaker's experience with generating avatars lead to the creation of a new startup?

The speaker's experience with generating avatars led to the creation of a new startup called 'Avatar AI' after realizing the potential of the technology to generate images of oneself in various styles.

What was the speaker's reaction to a big VC company entering the avatar market and outperforming their app?

The speaker felt it was amazing and saw it as a validation of the market potential, despite the competition.

What was the speaker's approach to handling customer orders for AI-generated avatars before automating the process?

Initially, the speaker manually took customer photos, trained the model, and emailed the generated avatars back to the customers.

What challenges did the speaker face when scaling the avatar generation service?

The speaker faced challenges such as the need to automate the process due to the high volume of customer orders and dealing with increased costs for training the AI models on the platform they were using.

Outlines

00:00

🖼️ Generative Image Models and AI Startups

The speaker discusses their initial experience with generative image models, specifically Stable Fusion, which they used to generate images on their MacBook. They discovered the model's proficiency in creating architectural designs, particularly houses, leading to the creation of 'this house does not exist.org'. They later pivoted to interior design with Interior AI, which became a successful startup. The speaker also experimented with image-to-image technology and fine-tuning AI models for specific tasks, like interior design, and eventually created an AI that could generate images of themselves in various styles, leading to the creation of Avatar AI, which went viral and earned significant revenue.

05:01

💸 Monetizing AI-Generated Avatars

The speaker shares their journey of monetizing AI-generated avatars, starting with a simple HTML page and Stripe payment link. They manually processed customer orders by training AI models on uploaded photos and emailing the generated images back. The demand was high, with even famous individuals and billionaires using the service. The speaker reflects on the rapid growth and the need for automation, which led to the development of a more sophisticated website. They also discuss the challenges of scaling and the importance of finding the right AI platform for model training, highlighting their positive experience with Replicate.

10:02

🔧 Fine-Tuning AI Models for Personalized Avatars

In this section, the speaker delves into the technical aspects of fine-tuning AI models to create personalized avatars. They discuss the process of training the AI with a limited number of photos and the challenges of ensuring high-quality results. The speaker also shares their experience with different AI model training platforms, the impact of their business on these platforms' pricing, and the eventual shift to Replicate, which accommodated their needs and contributed to the platform's growth. The speaker emphasizes the importance of finding the right tools and strategies for fine-tuning AI models to achieve the desired outcomes.

Keywords

💡Generative Image Model

A generative image model refers to a type of artificial intelligence algorithm that can generate new images from existing data. In the context of the video, the model is used to create images based on textual prompts, showcasing its ability to produce a variety of images in different styles. The script mentions the user's experience with the first generative image model, Stable Fusion, which they used to generate images on their Mac.

💡Stable Fusion

Stable Fusion is a specific generative image model mentioned in the script. It is highlighted as the first model the user experimented with, which they could install on their MacBook. The user's experience with Stable Fusion led them to create various projects, including 'this house does not exist.org', demonstrating the model's capabilities in generating images of houses and architectural designs.

💡Prompt

In the context of AI and generative models, a 'prompt' is a textual description or command that guides the AI to generate specific content. The script describes how the user would write prompts to guide the AI in creating images, such as houses or interior designs, in various artistic styles. This process showcases the interactivity and creativity involved in working with generative AI models.

💡Interior AI

Interior AI refers to the application of AI in the field of interior design. The script details how the user pivoted to using AI for generating interior design images after noticing the model's proficiency in creating images of houses. They created a website, Interior AI, to offer AI-generated interior design solutions, which became a successful business venture.

💡Image to Image

Image to image technology is a process where an AI model takes an existing image and transforms it into a new image based on certain parameters or modifications. In the script, the user experimented with this technology to modify interior photos, changing elements like设计风格 to create new designs. This showcases the adaptability of AI in altering and enhancing visual content.

💡Fine-tuning

Fine-tuning in AI refers to the process of adjusting a pre-trained model to better perform a specific task. The user in the script fine-tuned an existing AI model to specialize in interior design, creating a more efficient and targeted tool for generating interior design images. This highlights the customization and optimization possible with AI models.

💡Avatar AI

Avatar AI is a term used in the script to describe a project where the user trained an AI model on their own face to generate various风格的 avatars. This project went viral and led to significant financial success. It demonstrates the potential of AI in creating personalized and engaging content that resonates with users.

💡Photorealism

Photorealism in AI refers to the ability of AI models to generate images that closely resemble real photographs. The script mentions the user's shift towards creating a more photorealistic AI application, indicating a move towards higher quality and more realistic visual outputs in AI-generated content.

💡Replicate

Replicate is mentioned in the script as a platform for fine-tuning AI models. The user leveraged this platform to train models for their various AI projects, including Avatar AI. The platform's role in the script underscores the importance of accessible tools and services in the development and scaling of AI applications.

💡API

An API, or Application Programming Interface, is a set of rules and protocols for building and interacting with software applications. In the context of the script, the user utilized APIs to automate the process of generating images with AI models, allowing for the creation of a more streamlined and efficient service for users.

💡ML Model

ML Model stands for Machine Learning Model, which is a type of AI model that learns from data and improves its performance over time. The script describes the user's journey in training ML models for various purposes, such as generating images of houses, interiors, and avatars, highlighting the versatility and power of machine learning in AI applications.

Highlights

Stable Fusion is a generative image model that can be installed on a Mac.

The model generates images from text prompts in various styles, including artistic styles like Picasso.

Architecture, particularly houses, is an area where the model excels, unlike human figures.

A website called 'this house does not exist' was created to showcase non-existing houses generated by the model.

The website featured a voting system for users to upvote their favorite house designs.

Interior AI was launched to generate interior designs, capitalizing on the model's strengths.

Image-to-image technology was used to modify existing photos into different interior styles.

The speaker made a significant income from Interior AI, reaching 10K to 20K a month within a week.

Fine-tuning AI models was explored to specialize them in specific tasks, such as interior design.

Avatar AI was created after the model was trained on the speaker's face, leading to a viral response.

The speaker made a substantial amount of money quickly with Avatar AI, but felt it was too cheesy and not sustainable.

Photo AI was developed to focus on realistic photography without the need for a photographer or studio.

The speaker had to learn to automate the AI model training process due to the high demand for Avatar AI.

Replicate was used as a platform to fine-tune AI models, which helped scale the speaker's business.

The speaker trained over 36,000 models on Replicate, showcasing the platform's scalability.

The speaker's journey with AI models highlights the rapid evolution and commercial potential of generative AI.

Transcripts

so stable Fusion is like this the first

like generative image model a model and

um I started playing with it like you

could install it on your Mac like

somebody forked it and made it work for

MacBooks so I downloaded it and uh

cloned the repo and started using it to

generate images um and it was like

amazing like it would I found it on

Twitter because you see things happened

on Twitter and I I would post what I was

making on Twitter as well and you could

make any image you could write a prompt

to essentially write a prompt and then

it generates a photo of that or image of

that um in any style like they would use

like artist names to make like a Picasso

kind of style and stuff um and I was

trying to see like what is it good at is

it good at people not it's really bad at

people but it was good at houses so

architecture for example I would

generate like architecture houses um so

I made a website called this house does

not exist. org and it generated like

they called like house porn and that one

like house porn is like a subreddit so

and this was stable Fusion like the

first version so it looks really you can

click for another photo so it generates

like all these kind of non-existing

houses it is house porn but it looked

kind of good you know like especially

back then now things look much

better um that's really really well

done wow and it also generat like a

description and you can up vote is it

nice up vote it man there's so much to

talk to you about like the choic is here

it's really well this is very Scrappy in

the bottom there's like a ranking of the

most up voted houses so these are the

top voted if you go to all time you see

quite beautiful

ones yeah so this one is my favorite the

number one it's like kind of like a um

how is this not more popular it was

really popular for like a while but then

people got so bored of it I think cuz I

was getting bored of it too like just

continuous house porn like everything

starts looking the same but then I

thought it was really good at interior

so I pivoted to Interior ai.com

where um I tried to like upload first

generate interior designs and then I

tried to do like there was a new

technology called image to image where

you can input an image like a photo and

it would kind of modify the thing MH so

you see it looks almost the same as

photo it's the same code essentially um

nice so I would I would upload a photo

of my interior where I lived and I would

like change this into like a I don't

know like maximalist design you know and

it worked and it worked really well so I

was like okay this is a startup because

obviously interior design Ai and

nobody's doing that yet so I launched

this and I was successful and made like

within a week made 10K 20K a month and

now still makes like 40K 50k a month uh

and it's been like two years so then I

was like how can I improve this interior

design I need to start learning

fine-tuning and fine tuning is where you

have this existing AI model and you fine

tune it on a specific goal you wanted to

do so I would find really beautiful

interor design make a gallery and um

train a new model that was very good at

interior design and it worked and I used

that as well and then for fun I uploaded

photos of

myself and here's where it happened uh

and to train myself like and this would

never work obviously and it worked and

actually it started understanding me as

a concept so my face worked and and you

could do like different styles like me

as a like very cheesy medieval warrior

all this stuff so I was like this is

another startup so now I did Avatar ai.

me I couldn't get to do com and this was

uh this was yeah Avatar I me well now it

forwards to phot because I pivoted got

it but this was more like

cheesy thing so this is very interesting

because this went so viral it made like

I think like 150k in a week or something

it's the most money I ever made and um

and then big this is very the big VC

companies like lenza which are much

better at iOS and stuff than me I didn't

have IOS app they quickly build the IOS

app that does the same and they found

technology and it's all open technology

so it's good and I think they made like

$30 million with it yeah they they

became like the top grossing app after

that um and it was how do you feel about

that I think it's amazing honestly and

it's not like you didn't have like a

feeling like ah [ __ ] no I was a little

bit like sad cuz all my products would

work out and I never had like real

Fierce competition and now I have like

Fierce competition from like a a very

skilled High Talent like iOS Developer

Studio or something that and and they

already had an app they had an app in

the app store for like um I think

retouching your face or something so

they were very smart they add these

avatars to there it's a feature they had

the users they do a push notification to

everybody who have these avatars Yeah

man they made CRA I think they made they

made so much money and um I think they

did a really great job and uh I also

made a lot of money with it but that was

I I quickly realized it wasn't my thing

because it was so cheesy it was like

kits you know it it's kind of like um me

as a Barbie or me as a you know it was

too cheesy I wanted to go for like

what's a real problem we can solve

because this is going to be a hype this

is going to be and it was a hype these

avatars um it's like let's do real

photography like how can you make people

look really photo realistic and it was

difficult and that's why the offs wor

because they were all like in a Cheesy

you know Picasso style and art is easy

because you interpret the the the all

the that AI has with your face are like

artistic you know if you call it Picasso

but if you make a real photo all the

problems with your face like it just you

look wrong you know so I started making

photo AI which was like a pivot of it

where it was like a photo studio um

where you could take photos without

actually needing a photographer needing

a studio you don't just you know you

just type it and I've been working on it

for like the last yeah it's really

incredible that journey is really

incredible let's go to the beginning of

photo AI though cuz I remember seeing a

lot of re really hilarious photos I

think you were using yourself as a case

study right yeah yeah so what uh there's

a tweet here

sold

$100,000 in AI generated avatars and

it's a lot like it's a lot for anybody

it's a lot for me like uh making 10K a

day on this you know that's

amazing that's amazing and then the

nested tweet like this the launch tweet

and then the before there is like the me

hacking on it oh I see so

that okay so October 26 2022 I trained

an ml model on my

face sure because my eyes are quite far

apart I learned when I did YouTube I

would put like a photo of like my DJ

photo you know my mixure and people

would say I look like a hammerhead shark

was like the top comment so then I

realized my eyes are far apart uh yeah

the internet helps you realiz how you

look you know boy do I love the interet

first

trap what what is is this wait it's

water from the waterfall but the

waterfall is in the back you know so

what's going

on so this is how much of this is real

it's all AI it's all AI yeah that's

pretty good though for the early days

exactly so but this was hit or miss so

you had to do love curation cuz 99% of

it was really bad mhm so these are the

photos I uploaded how many photos did

you use only these I will try and more

up to date picks later these are the

these are the only photos you uploaded

yeah

wow wow okay so like you were learning

all this super quickly what uh what are

some like interesting details you

remember from that time for like what

you have to figure out to make it work

and for people just listening he

uploaded just uh just a handful of

photos that don't really have a good

capture of the face and he's able to F

is cropped it's like a cropped by the by

the the layout but they're they're

Square photo so they're 512 by 512 mhm

because that's sa fusion um but

nevertheless not great capture of the

face like it's not it's not like a uh

collection of several hundred photos

that it's like exactly like I I would

imagine that too when I started I was

like a this must be like some 3D scan

technology right yeah so I think the

cool thing with AI it trains the concept

of you so it's literally like learning

just like any AI model learn it learns

how you look so I I did this and then I

was getting so I was getting DMs like

telegram messages like how can I do the

same thing I want these photos my

girlfriend wants these photos so I was

like okay this is obviously a business

but I didn't have time to code it uh

make a whole like app about it so I made

a HTML page um register the domain name

and this was not even it was a stripe

payment link which means you have a

literally a link to stripe to pay but

there's no code in the back so all you

know is you have customers that paid

money then I added like a some uh a Ty

form link so Ty form is a site where you

can have make create like your own um

input form like Google forms so they

would get a email with a link to the

type form or actually just a link after

the checkout and they could upload their

photos so enter their email upload the

photos and then um and I launched it and

I was like here first SE so this October

2022 and I think within like the first

24 hours was like I'm not sure it was

like a thousand customers or something

um but the problem was I didn't have

code to automate this so I had to do

manually so the first few hundred I just

literally took their photos uh trained

them and then I would generate the

photos with the prompts and I had this

text file with the prompts and I would

do everything manually and this quickly

became way too much so but that's

another constraint like I was forced to

code something up uh that would do that

and that was essentially making it into

a real website right so at first it was

the type form and they uploaded through

the typ form strip check out form yeah

and then you were like that image is

downloaded did you write a script to

export download the images myself was a

zip file unzip file you unzipped it

unzipped yes and then I no I'm because

you know do things don't skill Paul

graah says right so uh and then I would

train it and I would email them the

photos I think from my personal email

say here's your here's your avatars you

know and they liked it they were like

wow it's

amazing you emailed them with your

personal email I didn't have an email

address on this domain and this is like

a 100 people yeah and then you know who

signed up uh like like man I cannot say

it but really famous people like really

really like billionaires famous Tech

billionaires did it and I was like wow

this is crazy and I sent I was like so

scared to mess them so I said thanks so

much for using my sites you know he's

like yeah amazing app great work so it's

like this is different than normal

reaction you know it's Bill Gates isn't

it can say

anything just like shirtless PS gdpr you

know like privacy European regulation

cannot share anything but I was very I

was like Wow and uh but this shows like

so you make something and then if it

takes off very fast you're like it's

validated you know you're like here's

something that people really want but

then also I thought this is hype this is

going to die down very fast and I did

because it's too cheesy but you have to

automate the whole thing how you

automate it so like what's the AI

component like how hard was that to

figure out okay so that's actually in

many ways the easiest thing because

there is all these platforms already

back then there was platforms for uh

fine tune stable diffusion um like now I

use replicate back then I use different

platforms um which was funny because

that platform when this thing took off I

would tweet because I tweet always like

how much money these websites make and

then so the the you called vendor right

the the G the platform that did the gpus

they increased their price for training

from $3 to $20 after they saw that I was

making so much money so immediately my

profit is gone because I was selling

them for

$30 and I was in a slack with them like

saying what is this like can you just

put it back to$ they say yeah maybe in

the future we're looking at it right now

I'm like what are you talking about like

you just took all my money you know and

they're smart well they're not that

smart because like you also have a

large uh platform and a lot of people

respect you so you can literally come

out and say that I think it's like kind

of dirty to cancel a company or

something I I prefer just bringing my

business elsewhere but there was no

elsewhere back then right so I started

talking to other um AI model ml platform

so replicate was on of those platforms

and I started dming the CEO say can you

please creates like it's called dream

Boo this fine tuning of you yourself can

you add this to your site cuz I need

this cuz I'm being price gged and he

said no because it takes too long to run

it takes a half an hour to run and we

don't have the gpus for it I said please

please please and then after a week they

said we're doing it we're launching this

and then this company became it was like

not very famous company it became very

famous with this stuff because suddenly

everybody was like oh we can build

similar apps like Avatar apps and

everybody started building Avatar apps

and everybody started using replicate

for it um and it was from these early

dmss with like the CEO like Ben Fish

very nice guy and he was like they never

price Gage me they never treated me bad

they always been very nice it's it's a

very cool company so you can run any ml

model any AI model llms you can run on

here and you can scale yes they scale

yeah yeah and I mean you can do now you

can click on the model and just run it

already it's like super easy you log in

with GitHub um that's great and by

running it on the website then you can

automate with the API you can make a

website that runs the m

generate images generate text generate

video generate music generate spech find

two models they do anything yeah it's

very cool company nice and you're like

growing with them essentially they grew

because of you because it's like a big

use case yeah like the the website even

looks weird now it it started as like a

machine learning platform that was like

I didn't even understand what it did it

was just too too ml you know like you

would understand cuz you're in the ml

world I wouldn't understand now it's

newb friendly yeah exactly and I didn't

know how it works and um but I knew that

they could probably do this and they did

it they built the models and now I use

them for everything and we trained like

I think now like 36,000 models 36,000

people already but is there some tricks

to fine tuning to like the collection of

photos that are provided like how do you

like yes man it's so many Hecks the hack

it's like 100 hex to make it

work for