The One Ring Forums: Tolkien Topics: Fan Art:
AI rescues non-artists liike me: an attempt at Gandalf vs the Balrog



CuriousG
Half-elven


Dec 31 2023, 9:32pm


Views: 6551
AI rescues non-artists liike me: an attempt at Gandalf vs the Balrog

I've tried a number of free AI art generators lately such as Microsoft Edge/Bing, and while I don't get exactly what I want, the results are far better than my freehand stick figure doodles.

This was supposed to be Gandalf fighting the Balrog, and I wanted something to closely approximate the movie FOTR Balrog--alas, no. But I'll keep experimenting and learning the best way to enter the words that prompt the AI.

For anyone else interested in trying, this is what you can generate by entering about a dozen words like "wizard white lightning sword duels giant fire demon cave dark mood." And, of course, you can generate anything else non-fantasy: I've had better results creating pastoral landscapes and sunset beaches.




Ataahua
Forum Admin / Moderator


Dec 31 2023, 10:02pm


Views: 6531
Crikey, how many abs does a fire-demon need?

I saw the results of an AI request for a minute-long beer commercial with young adults at an outdoor nighttime party with a fire pit, and the video was nightmarish - what I'd expect an alien to draw if they'd never seen a human drinking or laughing. Or holding a beer bottle, for that matter.

Celebrimbor: "Pretty rings..."
Dwarves: "Pretty rings..."
Men: "Pretty rings..."
Sauron: "Mine's better."

"Ah, how ironic, the addictive qualities of Sauron’s master weapon led to its own destruction. Which just goes to show, kids - if you want two small and noble souls to succeed on a mission of dire importance... send an evil-minded beggar with them too." - Gandalf's Diaries, final par, by Ufthak.


Fantasy novel - The Arcanist's Tattoo

My LOTR fan-fiction


CuriousG
Half-elven


Dec 31 2023, 10:16pm


Views: 6530
When you live alone in Moria for centuries, ab crushes are your only hobby

The results are always a gamble and they can be ghoulish or brilliant. I'm sure they'll only get more realistic, but right now AI art is like the internet before web browsers and quite primitive and clunky. I am only a dabbler, but I've seen all kinds of funny permutations, such as couples hugging and someone's arm magically reaching through the other person's torso and a spare hand emerging on the other side--ok, hugging lessons for AI, please!

But one genre with few disappointments is fantasy cities with lots of waterfall cascades sparkling from brilliant sunlight and adorned with many magnificent temples and gardens, the sort of stunning city that can make Minas Tirith look like a cheap slum. OK, that's hyperbole, but still, it's fun to play around with. I'm told that the real trick is to tidy things up with an art editor like Photoshop, but that sounds like too much work to me.


(This post was edited by CuriousG on Dec 31 2023, 10:16pm)


SirDennisC
Half-elven


Jan 6, 5:18am


Views: 6481
What about someone’s goat getting got?

I’d be afraid to make that prompt.

Questions: if you use ‘Gandalf’ instead of ‘wizard’ as a prompt, do you get something closer to an Alan Lee illustration? Do you have to tell it the figures’ positions relative to each other or describe the blocking you want; similarly do you tell it their relative size, image zoom level, frame, perspective or POV, how much negative space to leave, what other compositional elements to build in, etc? In other words, how many of these things are left up to the AI to decide?

3 horns is a nice touch. Is one missing or were there always only 3?


(This post was edited by SirDennisC on Jan 6, 5:22am)


CuriousG
Half-elven


Jan 6, 11:25pm


Views: 6442
Answers

The more I read on tip sites for AI art, it's very dependent on the words you enter and the word order and punctuation, and some generators also have a "prompt weight" like a scale from 1 to 20 which also matters--I think the weight means how hard the AI tries to match all the words? Probably over my head.

But I think the more precise terms and conditions that a person inputs, like those you listed, the closer you'll get to the desired outcome. The generators all seem to produce 3-4 images each time you hit Enter, and those can vary widely from each other. In one iteration, for example, Gandalf and the Balrog were a single creature, holding Glamdring while sprouting horns and fire--it felt like sacrilege! But the other two images showed wizard vs demon. So even the exact same set of prompts can lead to significant variations in output.

But specifically using "Gandalf": I didn't try his name at all, just permutations of "old white wizard." When I tried the specific term "Balrog" (upper- or lower-case), I got much worse results that looked more like a generic monster (just google "monster" and see how many variations you get; that's what I got). So switching to "fire demon" at least got me more in the right direction.

The basic advice is start with a noun, add adjectives, describe the scene in detail, then also add mood ("scary" vs "calm"), then add type of output (cartoon vs anime vs photo-realistic, etc). But I'm still very much a newbie.


(This post was edited by CuriousG on Jan 6, 11:26pm)


Wainrider
Rivendell

Feb 29, 6:56pm


Views: 5393
Good

It may not match exactly with our expectations, but it's still pretty cool!


CuriousG
Half-elven


Apr 15, 9:51pm


Views: 3172
Update on AI art: Luthien, Huan, and Sauron's cursed island

So AI has come a long way in a few months, with Bing Image Creator renamed Copilot Designer and using DALL-3, which is much better at interpreting language and giving realistic output. And if you're just a dabbler like me, you get lost quickly in all the jargon and decision settings of seeds, refiner weights, and sampling methods.


But I used the prompt below on Copilot a few times and got 2 images now posted here on Imgur that give the spirit, if not exact details, of Luthien and Huan going to rescue Beren from the dungeon of Sauron on Minas Tirith (there's no Sauron anywhere in the image, and Huan was disappointingly doglike and not a big hound she can ride). So anyway, not for literalists, but fun for dabblers who like to take stabs at things. Don't be afraid to try!

prompt:
fantasy scene. barren landscape and dark, cloudy skies. guarded by fierce dog, A mature Elf princess stands at a moat bridge and sings magical song to cause rays of sunlight to fall upon a dark castle on an island, breaking evil spell.

Speaking of stabs, I also had a decent AI image of Frodo vs Shelob, but giant spiders aren't too tough to create.


(This post was edited by CuriousG on Apr 15, 9:52pm)


guitarzankansasfan
Lorien


May 19, 4:50am


Views: 936
I've had some fun with AI mash-ups recently

Anyone who knows me knows I love mash-ups, and AI has made some very cool mash-ups possible. I recently got a couple of AI sites to make me some pictures of LOTR film characters as members of a Star Trek crew. I posted them on Facebook here: https://www.facebook.com/michaelgoble/posts/10231996308786661

(This post was edited by guitarzankansasfan on May 19, 4:50am)