While I currently make my living writing werewolf novels, my creative outlet of choice was visual arts from middle school through college. So when AI art opportunities started popping up, I couldn’t resist diving in with two feet. Here are my ultra-specific experiences asking Midjourney, Nightcafe, and DALL-E to create images that would fit into my fantasy novels.
For most people, Nightcafe is currently my top recommendation. It’s a good middle ground between the ultra-high-powered Midjourney and the light-weight DALL-E, with some bonus elements all its own.
Pros: I like everything about the user side of this AI engine, from the way you can follow other users and explore a feed of the most popular images (seeing what prompts created many of them) to the way you get free credits daily and for jumping through various hoops. It does pretty well with wolves and foxes and people and an excellent job with landscapes and styles. If you click “add modifiers” under the text prompt box, you’re guided through lots and lots of artistic options in a way none of the other engines make very clear. Finally, you can enlarge an image you love without changing it in any way other than making it higher-resolution. Overall, Nightcafe is a great engine to grow with.
Cons: I just started playing with Nightcafe, so perhaps it’s no surprise that I’ve yet to make images as close to what I had in mind as those I made with Midjourney. I’m only starting to learn how to iterate images, a process that doesn’t seem as effective as Midjourney’s (so far).
Licensing: As long as you don’t use an otherwise owned image for style transfer, etc., then you own the output files (as far as anyone owns AI art, which is still up in the air.)
Recommendation: Follow me then give creation a try! You should be able to get some good images without dumping in any cash as long as you check back every day to claim your five free credits. (They don’t expire, but they do have to be clicked on daily if you want to stockpile them.) Then, when you want to tweak an image that had potential but isn’t fully there yet, don’t use “evolve”; use “duplicate” with the image you want to tweak chosen as a starting point and the text subtly changed.
I’ll tell you up front that I’ve played the most with Midjourney, so it’s no surprise I’ve created the images that fit my vision best there.
Pros: You can make some amazing images on Midjourney if you put in the time to iterate your favorite of the four options many times. Seeing what others create on the discord server is also a great way to get ideas for prompts (although the browsing process isn’t as smooth as it is in Nightcafe).
Cons: If you’re not already into discord, you may find the creation process overwhelming and esoteric. The tier I recommend (see below) is too pricey to be worth sticking to long-term for most of us. When you upscale small images to larger images, they change…sometimes in awesome ways and sometimes in awful ways. Also, when I was playing with Midjourney this summer, it was terrible at wolves and had a real problem with giving all women skirts unless you specifically told it to give them pants then adding in huge butts — I hope these issues have improved since then or do soon.
Licensing: If you have a free account, your images are licensed under Creative Commons. If you have a paid account, you own your images (with relevant caveats about whether or not anyone owns AI art).
Recommendation: Check out the hardcover I currently have up on kickstarter, built around Midjourney imagery and funded in 26 hours! Then, if you’re diving into a project like this with two feet, go ahead and pay for the $30/month tier. This lets you make unlimited images as long as you’re willing to use the “relax” mode and wait a bit longer for each to be produced. Definitely use an iterative process to choose one of the four options and settle in for a long creative afternoon!
I’ve only dabbled in DALL-E, but wasn’t very impressed.
Pros: Free credits top up once a month. It seems to be better at making wolves and foxes that look real than Midjourney (although not as good as Nightcafe). And its iterative process seems to be better than Nightcafe’s. Plus, DALL-E is very simple to use.
Cons: I only used my first-month free credits, but I didn’t come up with any images that wowed me. It didn’t seem to respond to style suggestions very well, which meant my images came out okay but hum-drum. You’re on your own figuring out how to get good images unless you join their discord channel or one of the facebook groups. And there’s a rainbow “signature” at the lower right that might be annoying to some.
Licensing: DALL-E doesn’t actually say you own the output, but they say you can use it commercially.
Recommendation: This might be a good one to dip into as a toy…or I might be totally off-base and it might be very powerful once figured out.
As a final note, I feel like I’d be remiss if I didn’t add — I wrote this in mid October 2022 and all of the engines are changing and growing daily. Who knows which one will be the best in a month or a year?
P.S. Kicktarter update — we’re working toward our first stretch goal! I hope you’ll join in the fun.