What is AI Art?
AI art is digital images that were generated by an Artificial Intelligence model.
There are many different types of AI art, but all of them follow a similar pattern:
- A human inputs a request (I.E. a text prompt describing an image)
- The AI interprets that request
- The AI generates an entirely new image that it thinks fulfills that request
Text Driven Image Generation
This is the most popular type of AI art right now due to the recent momentous breakthroughs in the technology.
In 2015 data scientists were able to train an AI model to generate natural language captions for an image. Ever since others have been attempting to turn that process around so that an AI can generate an image from an inputted caption.
Many data scientists attained various different levels of success at creating such a system, but it wasn’t until 2021, when OpenAI came out with DALL-E when a true text drive image generation system was born.
DALL-E is an AI system that can generate images with shocking levels of detail from a simple text prompt.
Unlike GAN AI systems, text-driven systems such as DALL-E are trained on massive data sets with a diverse range of topics. This means you theoretically can prompt the system to create anything you can think of, even if it doesn’t exist in the real world.
Want to create a realistic image of a futuristic city full of plant life? Not a problem! Want to see an evolved cat watching the beginning of the universe? You got it!
Prompt: Hyper realistic futuristic city, covered in green plants, big skyscrapers, cinematic, super detailed, white buildings, octain render, unreal engine, octane render, highly detailed, 8k
Prompt: cat floating in space, cosmic, cinamatic, galaxies, colourful, unreal engine, hyper realistic, 8k
These systems are not yet perfect by any means, most of the companies building text AI products only have beta systems out to the public at the moment. However, the technology is quickly progressing, and it’s easy to see how this type of AI art could quickly take over certain aspects of the design and art industries.
Wanna start playing around with Text Generative AI Art? Click below to skip ahead.
Image Source: NY Times – https://creator.nightcafe.studio/create
Style transfer art is the simplest to understand. It doesn’t require a huge data set to generate a completely new image. It simply takes two images provided by its user and combines them.
This is the easiest to get started with. Many of the text-driven AI systems also offer style transfer so use the text-driven section to learn how to get started.
GAN Generated Images
GAN AI systems are similar to text generative ones in that the AI generates images based on what it learned from its data set. However, the GAN system’s data set is much smaller and is focused on a very specific type the image.
For example, you may have heard of the AI-generated piece of art, “Edmond de Belamy, from La Famille de Belamy”, that sold for over $400,000 in 2018 (pictured below). This painting was trained on 15,000 pieces of art created between the 14th and 19th centuries, and as you can see below, the resulting art is reminiscent of that time period. As such, the model that created Edmond de Belamy, would not be able to create the space cats we created in the text-driven image generation section.
Image Source: NY Times – https://www.nytimes.com/2018/10/25/arts/design/ai-art-sold-christies.html
This type of AI art has the largest barrier to entry since it requires a person to develop advanced machine learning programming knowledge and to have access to a large enough bank of images in the style of the art they are trying to create.
That being said, if you are interested in GAN AI art, I point you in the right direction of where to get started in the next section.
Where to get started
Text Driven Generative Images
Image Source: OpenAI DALL-E 2 – https://openai.com/dall-e-2/
DALL-E 2 is widely considered the best text-driven AI system out there right now.
Luckily for you, Open AI (DALL-E’s creator) removed the waitlist to their beta. Now anyone can create a DALL-E account!
You get 50 credits for free when you first sign up and DALL-E will top up your account with 15 more free credits each month.
If that’s not enough, you can buy more credits in 115 chunks for $15 USD.
Prompt: painting a cyber punk city street, neon signs, raining, synthwave, hong kong
Midjourney is the AI I’d most recommend. While DALL-E is the best, Midjourney is a pretty close second and is getting better by the day.
Best of all, they just opened up their beta program to the public! You can sign up for the program and create 30 free images.
After you use your 30 free credits, you can sign up for one of their subscriptions. There are two plans to chose from:
- Basic Membership: $10/month, 200 images/month
- Standard Membership $30/month, unlimited images
Midjourney is built off a bot on Discord, so you need to create an account there too. No worries, Discord is free. Once you have a Discord account, you can sign up for Midjourney.
NightCafe is the least effective AI system on this list, but it’s the easiest to get started with. It is a web-based program, and you can get started by simply pressing the “create” button at the top of its homepage.
NightCafe runs off a credit system. You spend a credit per action you take on the site. Submitting a prompt, and enhancing an image each cost one credit.
NightCafe gives you 5 free credits to get started and test the platform, and you get 3 extra from making a profile. If you’d to keep creating one you’ve used your free credits, they have subscriptions starting at $10/month.
There isn’t a program I can recommend to you where you can start using GAN, you’d need to create your own model. The following is a list of popular beginner tutorials: