Is the open source company, Stability AI, really turning the world of AI art on its head?

In short, yes.

A lot of us are having fun using Midjourney to create AI art, and while some have waited months to get access to DALL·E 2, hoping for a less distorted experience perhaps, or wishing that Google would make their coolest versions of an AI art creation tool open to the public for free, something else happened. Something HUGE.

One day, not so long ago, in this very galaxy, and from the back of the field, or the sidelines if you like, a free AI called DreamStudio (formerly known as Stable Diffusion, which has been in a Discord beta release for some time), decided to sprint ahead and join the group at the head of the race. And now, it’s turning the whole AI art scene on its head.

Why is DreamStudio, which is made by Stability AI, disrupting things precisely? Well, imagine yourself in a field of contenders, like a bicycle race. One day the best and most popular riders decide to charge for their services, and as they are great at what they do, they get paid. Now imagine, that one of the says, “I’ll race to win for free.”

Not only is DreamStudio good, faster than most AIs, and easier to use, but it's also cheaper to use than the competition. And let's not stop there the open source version of their AI it’s now available in a free version for you to install on your own computer (be warned though that it’s still rather nerdy, but it will get easier, they say.)

The classic Teddy Bear on a skateboard in Times Square and Chris Rock as a Greek god. I created all the images but borrowed ideas from other users. Thank you to the Stable Diffusion community.

I’ll cover the paid version first:
DreamStudio is easy to use but hard to master (like most really fun things in life.) To begin, just write a brief description of your idea and click on the Dream button. The AI does the rest, creating the image within seconds or dreaming them up if you like.

You’ll quickly learn what works and what doesn’t by messing around with the sliders. For instance, chose the width and height or how many images to generate each time you click Dream.

A quick way to get to understand how this work is to read through the Prompt Guide and the FAQ found in the left-hand menu on the site. Also, check out the videos below, especially the first one.

What does it cost to use?
You get £2 worth of free images, to begin with, but those will soon be gone, especially if you chose to make higher-quality images. Do be aware that the more images you generate each time (you may choose 1-9) or the higher quality settings you select, the more each image will cost. Prices go from 2 pence at the lowest quality and up to nearly 30 pence per image (from UK £0.02 / US $0.02 / DDK 0,18 to UK £0.282 / US $0.33 / DDK 2,46.) Read more in the FAQ once you get access.

How does that compare to the competition?
While the competition like Midjourney costs $30 for a somewhat decent monthly subscription, AIs like Google’s Deep Dream and DALL·E 2 are even more expensive. However, if you compared DALL·E 2 with DreamStudio the last costs roughly a third, but only if you make images of around the same quality. This complicates things as you inside DreamStudio have a lot more features to use than the spartan DALL·E 2 interface. If you use these features to up the scope or quality you will make each image cost more and you will end up paying more each month with DreamStudio. If you’re the sort of person who creates a ton of stuff, and/or like to do quality image, you will find it much cheaper to go with Midjourney’s flat monthly fee of £30, but if you do few images, or a lot of small scale and/or low-quality images DreamStudio is as good or better. As a comparison, the £30 will get you between 91-1.500 images in DreamStudio depending on the sizes and image quality. 91 is like 3 good images a day for a month. Most people I think will find it cheaper to go with Midjourney.

Still, as DreamStudio’s interface is super easy to work with (better than Midjourney’s Discord interface or DALL·E 2’s simple interface,) that will put it ahead of the competition for some. Also for the free version, we are talking open source which potentially means you can change the code, making it better, or different, or, if you’re that sort of person, remove any safeguards against the nastier things the AI can create (yes, any taboo image you can think of is up for grabs.)

What’s this about no cost and open source?
Many of the AIs are harder to get access to. For instance, I’ve been waiting since April to get access to DALL·E 2. Only a few text-to-image AIs are open source. This is ironic if you know that the Elon Musk-affiliated company Open AI that created DALL·E 2 is not actually doing open source (perhaps a name change is in order?) If you think open source is some old-fashioned hippy wannabe crap, or that nobody is really going to use this you’d be wrong. Quite recently one of the top players in the AI field announced they are going to use the same AI code (DreamStudio / Stable Diffusion) in their Artbreeder AI.

So how do I get access to Stable Diffusion for free or their DreamStudio version?
Wait, Midjourney has just gotten a sweet upgrade to their engine making it able to do faces in a much more realistic manner than before, so if you are already paying for that AI I see little reason to shift to Dream Studio. But the AI code is very close in quality to Midjourney and DALL·E 2, it’s easier to use, and all of these AIs are constantly getting better, so it can be difficult to decide.

So how do I download the paid or the free API version of this AI and generate stuff like this?

I was promised access
And you shall have it!

First, a quick word from our sponsor (namely me who wrote this article, Rune S. Nielsen, the fantasy author.)

If you want to see more articles like this, please spend a few dollars supporting me. Buy one of my great fantasy books, I very much recommend this one Doomsayer Prince.

Here’s the paid-for version, the open DreamStudio Beta. Enjoy :-)

(Two fun facts: Most of the popular text-to-image AIs such as Stable Diffusion use diffusion models to generate images. Diffusion is one of the biggest developments in machine learning in recent years.)

The free version you find at Gitbub. This message was posted by the company:

a few tutorial videos to get you started with the paid version

A great first look guide
(by Matt)

A quick tour of DreamStudio beta (by DreamStudio / Chris Allen)

Two simple video tutorials (by DreamStudio / Kali Yuga)

Other good stuff happening
around the AI art scene
A lot of great new stuff out there like:

• This beautiful graphic novel created by Kris Kashtanova.

• Tips for using prompts in Midjourney by Moscangalo.

• Tips for getting good at using Midjourney by Scott Detweiler.

• More than 20 text-to-image AIs ranked by Matt.

  You should definitely check all of these out, they are worth using time on!

  And then there’s the darker side of AI…

The bad stuff. Who’s asking the uncomfortable questions?
What are the legal and moral implications of using AI art? Well, it will lead to a lot of bad stuff, too. As an example: in the two galleries above, I made a couple of images of the celebrities Chris Rock and Bob Dylan. While I was careful to try and show them in a respectful light (I even asked the AI to portray Chris Rock as a Greek God, what could be more flattering?) But what if Chris Rock hates my image? And what happens when people by choice abuse AI images to cyberbully celebrities or people they know? And what happens on the day this causes someone to end their life?

How do the artists feel when AIs take all their business or outright copy their artistic styles? On the Midjourney and Stable Diffusion Discord Servers, thousands of users have been instructing the AIs to make art that resembles the works of specific artists or studios. Will this perhaps make some artists even more famous, lead to others losing their livelihood or both?

And if we look at it from a larger perspective, like our whole society, how many governments or companies will use AI tech to create smear campaigns against their rivals? Just think about how big a budget the next big election will use on fake news powered by fake AI images.

Since some of the best AI art tech is now open source (or soon will be) anyone may download and modify it from the school bullies going after other kids to perhaps criminals using it to create for instance fraud or fake official-looking documents.

And since so many big and small companies/governments/education centers are focusing on AI these issues will only become bigger and more complex over time. Some political parties and governments have for years been using information warfare targeting for instance their own population to keep them voting for the, or making them more docile or afraid. How will AI images/tech change this?

In my next article, I’ll be looking into these very issues.