Artificial Intelligence artwork has been all the rage this summer. It's opened up the doors to an enormous amount of people to start expressing themselves in ways that were previously not possible. It's also been the topic of countless debates, but one thing that isn't debatable is that all of this stuff is here to stay.
We've done some exploring ourselves, and today we're giving you a quick rundown of the hottest AI tools people are creating with today. The speed at which these tools are evolving is impressive, but most of the work can be divided into 4 distinct toolsets - Dalle2, Midjourney, Stable Diffusion, and Google Colab notebooks.
Dalle2 has been making waves for months in the AI artwork scene. It's also been the most challenging for regular people to actually use. They had a massive waitlist that you had to join before gaining access, and it was initially unclear on what is and isn't allowed in terms of selling the works generated.
Despite all that, it's pretty clear that Dalle2 is ridiculously powerful and has created some extremely compelling content. The banner/featured image for this article was generated by Dalle2, for example.
This is where all the magic starts, though. All it takes to get started is a few words, but if you've never used a text to image generator before, I'd recommend looking at their help section so you don't end up wasting any credits.
Example Dalle2 prompt: anime dog with human arms lifting weights
It feels like there are a million different ways to take Dalle2 and the amount of things you can output for the cost is totally reasonable.
Midjourney is a strange one when compared to Dalle2 and the various flavors of Stable Diffusion out there. It's deeply integrated into Discord and the process for creating things requries sending your prompt into a channel that resides on their Discord server, whereas with the other options like Dalle2 & Stable Diffusion are a lot more standalone. This can be a good or bad thing depending on how you view things.
This is an example of how images come out in Discord. I used the prompt "a basketbal on fire".
Midjourney also has an aesthetic that many people find recognizable when compared to something like Dalle2 or Stable Diffusion.
It's got a similar pricing model to Dalle2. A subscription of $10 entitles you to ~200 image generations, with additional credits able to be purchased afterwards. It's worth checking out, but I think there's better value to be had in Dalle2 and some of the Stable Diffusion GUIs in terms of pricing. If you have any specific questions, their docs section probably has you covered too.
Stable Diffusion is my favorite tool out of the current bunch that people are using. It's able to generate a wide variety of styles and has a ton of depth. It's also easier to get started with since it's a lot more accessible than Dalle2, and there's a plethora of Google Colab notebooks available for it.
Since Stable Diffusion is more inherently open than the other models in this blog post, there's a lot more choice available to users in generating images. My favorite app or frontend for the Stable Diffusion model is DreamStudio. DreamStudio has similar pricing to Dalle2 and Midjourney, but their GUI is extremely flexible and powerful, presenting you with many options (that are clearly explained) in a much more mangeable way than using Google Colab notebooks.
Here's the DreamStudio UI:
DreamStudio is built and maintained by Stability.ai who have deep roots in the AI scene, making this one of the most polished AI tools around.
Stable Diffusion v1.5 prompt: "anime dog with human arms lifting weights"
To be clear, almost everything that you can do with a paid service like DreamStudio can be done for free using Google Colab notebooks, it's just a difference in ease of use. $10 gets you 1000 credits on DreamStudio, and the default settings come out to 1 image per credit. You can crank the settings up though for more resolution, more steps to create, and it will use up more, but it's all very clear.
The last item to consider is familiarizing yourself with Google Colab, which are basically development environments that can be easily reproduced and ran through your own browser. There's significantly less handholding with Colab notebooks as opposed to the other tools mentioned like Dalle2 and Dreamstudio, but it is a lot more flexible, and Colab Notebooks are where the most cutting edge tech is available.
Google offers free and paid tiers of service - the paid tiers offer up more memory and faster GPUs to run these notebooks on. It's a higher learning curve, but can yield the most interesting results sometimes too.
The AI and Machine Learning fields are moving at a breakneck pace and the popular tools today may not be in vogue tomorrow. I think the ceiling is quite high with any of these tools mentioned, but I would certainly give the edge to Stable Diffusion right now in terms of output and costs to use.