The other day I stumbled on Midjourney, an AI generating images from descriptions.
Out of curiosity, I decided to see what happens if I input one of the seeds from the other day.
I picked the first one of the most recent ’12 seeds’ posts (Eldritch Fairy Tale):
An iron elemental wants to find the path home and their only lead is an occult amulet, but they will live the rest of their life with false memories.
The first three images are all keyed off the same input. I generated one set, then told it to generate new variations twice.
I expanded each of the images in the first quad, to get a better look, and told it to show me variations of that image.
The interface I’ve seen so far is all through discord. I haven’t experimented much with the tool, but I like the results I’m getting. Even if the images don’t cover everything in the input (I can’t imagine how one might represent “but they will live the rest of their life with false memories”) I’m still feeling pretty good about what I see. I’ve really only covered a single input string, and the results are simultaneously varied enough to interest me, but consistent enough in style I could see using several.
Well, I did try one other string, without the conflict line: “An iron elemental wants to find the path home and their only lead is an occult amulet”, and it did give me a different result.
I definitely need to spend some more time with this tool, especially if I can use the results commercially.
… and having now reviewed the licensing, FAQ, and subscription pages, the above images are usable under the Creative Commons Noncommercial 4.0 Attribution International License (but they don’t provide the attribution text, so I’ll just note it as being so licensed).
Paid subscribers can use the generated images commercially. The usage caps are based on GPU usage, each new image is about 1 GPU minute (upscales are more, variants are less). Basic plan is currently $10/month and good for 200 minutes/month. Standard is $30/month and good for 15 hours/month (so ~900 images) ‘fast’ and unlimited ‘relax’ GPU time (i.e. “when we get around to it”, and contingent on how much volume there is otherwise). Normally the prompts and responses are presented in DMs with the bot, and results are accessible in the gallery. Both subscription models allow paying for additional capacity ($4/GPU hour), and both have ‘private’ options that keep your images and prompts out of the gallery.
There also is a ‘corporate’ subscription, which is $600/year for 120 hours/your, unlimited ‘relax’ GPU time, and comes with private visibility automatically. Users of companies with more than a million dollars revenue annually are required to use the corporate plan. I don’t see myself needing this.
Huh. The math on this looks a bit weird. If you’re doing more than a couple hundred images per month it quickly makes sense to go to the standard plan (700 minutes is about 12 hours, 12 hours at $4/hour is $48… it makes sense to shell out $20 more per month for the extra capacity and get the unlimited slow mode). The corporate only makes sense if you’re required to use it, because the annual cost is the same as annual standard + private visibility ($30 * 12 = $360, $20 * 12 = $240, $360+$240 = $600) but you get only 120 hours/year — 10 hours/month on average.
Still, I can see myself subscribing to this service, if I continue to like what I see as much as this.
A few more…
I ran a few more seeds, just to see what I’d get (and I ran out the trial limit, I need to subscribe to get more… fair enough).
A monarch wants to obtain a rare spell component from a seaside grove, but one way or another, it will be their last adventure.
I suppose I could have been more clear regarding what ‘monarch’ means.
A merfolk cultist wants to navigate a maze, at the heart of which sites an eldritch shipwreck, but they must forsake that which they worship.
Yes, okay. I can see this working.
A mage wants to reclaim their home from an infernal wonderland, but they will suffer an injury that does not heal.
Very yes please.
A dragon wants to sabotage a fairy-touched enemy’s castle, but an ancient evil will be released.
This one I might want to explore a bit, adjust the color palettes — so dark! — but it’s certainly something I could start with.
And Something Completely Different
This was one of the first images I tried, just to see what would happen.
Cavern with waterfall and giant mushrooms
Absolutely I can dig on this. Perhaps literally, even.