How I made $3000 in 3 weeks selling AI-generated art? πŸ’°

Last week I showed you how you can use VQGAN+CLIP to generate interesting images based on text prompts.

Now, I'll tell you how I sold some of these as NFTs for more than $3000 in less than 3 weeks.

Let's go πŸ‘‡
Background

I've been interested in NFTs for 2 months now and one collection I find interesting is @cryptoadzNFT. What's special about it is that the creator @supergremplin published all of the art in the public domain. This spurred the creation of many derivative projects.

πŸ‘‡
The Idea πŸ’‘

My idea was to use VQGAN+CLIP to create interesting versions of the CrypToadz. So, I started experimenting with my own toad #6741.

I took the original NFT image as a start and experimented a lot with different text prompts. The results were very promising!

πŸ‘‡
Monetizing πŸ’°

But how to sell these? I got inspired by the GelToadz project by @DannyZareto. People that own a CrypToad would be able to commission an AI Toad from me.

I also wanted people to be able to add some personal touch to the AI Toad by specifying some custom theme

πŸ‘‡
Website 🌐

I hacked a minimal single-page website with @tailwindcss and launched it on @Netlify in a day. For orders, I used a free Airtable form where people could enter their CrypToad ID, wallet address, and their custom theme. Keep it simple!

aitoadz.xyz

πŸ‘‡
Price 🏷️

I decided to go with an incremental price:
β–ͺ️ First 10 Ai Toadz cost 0.05 ETH ($230)
β–ͺ️ The next 10 Ai Toadz cost 0.1 ETH ($460)
β–ͺ️ After that each Ai Toadz cost 0.2 ETH ($920)

The update of the price on the website is manual by me changing the HTML πŸ€·β€β™‚οΈ

πŸ‘‡
Launch πŸš€

I launched it on Twitter and a couple of Discord servers and started getting 1-2 orders per day.

An important step was to also get my project listed on toadztoolz.com - a website acting as a directory of all CrypToadz related projects.

πŸ‘‡
Creating NFTs 🎨

I use a popular Colab notebook modified so I can easily run multiple experiments. For each order, I try different prompts and styles until I have 2-3 variants I like and that fit the theme provided by the client.

Experiments for #6067 "War never changes"

πŸ‘‡
After that, I contact the client and show them the variants I have. In some cases, I make some additional experiments based on the feedback. Some people want specific features to be visible, while others like particular styles more.

This is the artistic part of the process!

πŸ‘‡
In some rare cases getting the desired results wasn't easy, so I had to do some tricks. I either modified one of the intermediate images to enforce a particular look or created two separate versions. In the video, I had to blend both versions using DaVinci Resolve.

πŸ‘‡
Compute πŸ’»

Some toadz are easy, but some require a lot of experimentation with the prompt. To be efficient, I define multiple experiments and let them run for hours in the background. I subscribed to Google Colab Pro+ so I can run 2 notebooks in parallel on a P100.

πŸ‘‡
NFT ⛓️

After an AI Toad is ready I create a video NFT using OpenSea. It is really easy and free! You need to provide:
β–ͺ️ The video
β–ͺ️ A thumbnail
β–ͺ️ Name
β–ͺ️ Description
β–ͺ️ Optional properties (I didn't specify any)
β–ͺ️ Collection

πŸ‘‡
Collection 🏞️

Creating a collection for the AI Toadz NFTs is also easy - you need a description and some images for banner, logo, and social media previews.

Check out the AI Toadz collection on OpenSea here:
opensea.io/collection/ai-…

πŸ‘‡
Sale 🀝

I then list the NFT for the agreed price as a private sale - only for the wallet address holding the original CrypToad.

OpenSea uses a process called lazy minting, so there are no gas fees for the creator! The NFT is minted on the blockchain with the first sale.

πŸ‘‡
Orders πŸ“”

Now some statistics on the current state of the orders:
β–ͺ️ 15 valid orders for a total of 1 ETH ($4600)
β–ͺ️ 11 AI Toadz actually sold for a total of 0.7 ETH ($3300)
β–ͺ️ 4 AI Toadz in progress
β–ͺ️ Current price 0.1 ETH ($460)

πŸ‘‡
Summary 🏁

To summarize my process:
β–ͺ️ Create a simple website and AIrtable form to collect orders
β–ͺ️ Experiment with different prompts for VQGAN+CLIP
β–ͺ️ Iterate with the client if needed
β–ͺ️ Create NFT on OpenSea and sell it

πŸ‘‡
Next ▢️

I don't plan any big marketing effort. If orders stop coming I may post more about them on Twitter/Discord and also create some free ones for members of the community to boost the visibility. This is supposed to stay a small side-project.

No roadmap - only !vibes 😁
The AI Toadz 🐸

I shared some examples across the thread, but if you want to browse all of them check out the collection on OpenSea:

opensea.io/collection/ai-…
Ressources πŸ”—

If you are interested in how you can use VQGAN+CLIP to generate your own images, check out my thread here

I regularly post threads on machine learning, but now I'm increasingly looking at the intersection of machine learning with wen3, NFTs, and crypto.

Follow me @haltakov for more!

β€’ β€’ β€’

Missing some Tweet in this thread? You can try to force a refresh
γ€€

Keep Current with Vladimir Haltakov

Vladimir Haltakov Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @haltakov

8 Nov
Why is AI bad at math? πŸ“

Machine learning models today are good at generating realistic-looking text (see GPT-3), images (VQGAN+CLIP), or even code (GitHub Co-Pilot/Codex).

However, these models only learn to imitate, so the results often contain logical errors.

Thread πŸ‘‡
Simple math problems, like the ones 10-year-old kids solve, usually require several logical steps involving simple arithmetics.

The problem is that, if the ML model makes a logical mistake anywhere along the way, it will not be able to recover the correct answer.

πŸ‘‡
@OpenAI is now working on tackling this issue.

In their latest paper, they introduce the so-called verifiers. The generative model generates 100 solutions, but the verifiers select the one that has the highest chance of being factually correct.

openai.com/blog/grade-sch…

πŸ‘‡
Read 13 tweets
3 Nov
How to create art with Machine Learning? 🎨

You've probably seen these strangely beautiful AI-generated images on Twitter. Have you wondered how they are created?

In this thread, I'll tell you about a method for generating art with ML known as VQGAN+CLIP.

Let's jump in πŸ‘‡
Short History πŸ“œ

In January @OpenAI publicly released CLIP, which is a model that allows matching text to images.

Just days after that, some people like @advadnoun, @RiversHaveWings, and @quasimondo started experimenting using CLIP to guide the output of a GAN using text.

πŸ‘‡
OpenAI published an image generation model together with CLIP, called DALL-E, but without the full code and the pre-trained models.

The results from guiding StyleGAN2 or BigGAN with CLIP aren't as accurate as DALL-E, but they are weirdly artistic.



πŸ‘‡
Read 18 tweets
2 Nov
Creators only get badges πŸ…

There is a problem with how value is distributed in online communities today. It seems we take the status quo for granted and don't discuss it much.

The people that create most of the value, get none of the money! Only badges...

Thread πŸ‘‡
Online communities

I'm talking about platforms like Twitter, Reddit, Stack Overflow etc. They're wonderful places, where you can discuss interesting topics, get help with a problem, or read the latest news.

However, the people that make them truly valuable receive nothing πŸ‘‡
It usually looks like this:

β–ͺ️ Company creates a web 2.0 platform
β–ͺ️ Users create content and increase the value
β–ͺ️ Company aggregates the demand
β–ͺ️ Company monetizes with ads and subscriptions
β–ͺ️ Company gets lots of money
β–ͺ️ Creators get badges, karma and virtual gold

πŸ‘‡ Image
Read 25 tweets
13 Oct
Machine Learning Formulas Explained! πŸ‘¨β€πŸ«

This is the formula for the Binary Cross Entropy Loss. This loss function is commonly used for binary classification problems.

It may look super confusing, but I promise you that it is actually quite simple!

Let's go step by step πŸ‘‡
The Cross-Entropy Loss function is one of the most used losses for classification problems. It tells us how well a machine learning model classifies a dataset compared to the ground truth labels.

The Binary Cross-Entropy Loss is a special case when we have only 2 classes.

πŸ‘‡
The most important part to understand is this one - this is the core of the whole formula!

Here, Y denotes the ground-truth label, while ΕΆ is the predicted probability of the classifier.

Let's look at a simple example before we talk about the logarithm... πŸ‘‡
Read 15 tweets
21 Sep
There are two problems with ROC curves

❌ They don't work for imbalanced datasets
❌ They don't work for object detection problems

So what do we do to evaluate our machine learning models properly in these cases?

We use a Precision-Recall curve.

Another one of my threads πŸ‘‡
Last week I wrote another detailed thread on ROC curves. I recommend that you read it first if you don't know what they are.



Then go on πŸ‘‡
❌ Problem 1 - Imbalanced Data

ROC curves measure the True Positive Rate (also known as Accuracy). So, if you have an imbalanced dataset, the ROC curve will not tell you if your classifier completely ignores the underrepresented class.

More details:

πŸ‘‡
Read 19 tweets
20 Sep
How to spot fake images of faces generated by a GAN? Look at the eyes! πŸ‘οΈ

This is an interesting paper that shows how fake images of faces can be easily detected by looking at the shape of the pupil.

The pupils in GAN-generated images are usually not round - see the image!

πŸ‘‡
Here is the actual paper. The authors propose a way to automatically identify fake images by analyzing the pupil's shape.

arxiv.org/abs/2109.00162
The bad thing is, GANs will probably quickly catch up and include an additional constraint for pupils to be round...
Read 5 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Thank you for your support!

Follow Us on Twitter!

:(