|
Post by CharlesX on May 4, 2024 14:37:34 GMT
It is quite saddening that, for example, movie studios are now using AI to create posters (Civil War being the current example that come to mind with its photo realistic beauty but geographic errors) and in-movie still images such as in Late Night with the Devil. I can see the appeal for the studios, there must be huge cost and time savings, getting near instant art in the style you want with hardly any effort. They'll be replacing the actors and actresses one day, too. For the same reasons. What's to stop them? At least it will spare us the 'spectacle' of the Oscars year on year. It is amazing how little time and skill is involved in creating these pieces of βartβ. Sometimes I take modern technology for granted. Is this all a good thing? ... the skilled artists who were previously hired to design these posters (and my walls were adorned with some fantastic ones in my youth) are no longer required and have less potential work and less motivation to become artists in the first place. π I don't want skilled artists to disappear any more than you do, but I don't know what we are to do about it. Having said that, AI art must be very useful for a self-publishing independent creator of gamebooks just starting out in the field, with no money to pay for artwork. I think illustrations are a fundamental part of gamebooks. In fact I wish MORE fiction was illustrated, across the board. Time will tell whether AI art does indeed replace manmade art, and computer-made 'actors' replace human ones. I've a feeling, at least for a while, it will be like VR\Occulus Rift* or Ebooks where they have a strong following but their older opponents continue to easily outsell them. I've no idea how far technology has moved on - it probably has passed this point - but does anyone remember Disney's attempts to recreate the late Princess Leia? That's the problem with this technology, we're at a very basic stage where it can't replicate images or get the numbers of fingers or hands right. So, it has to get this 'common sense' level, then entire layers more, before its close to competitive.
|
|
|
Post by paperexplorer on May 6, 2024 11:39:35 GMT
Okay I think the AI had a big weekend and may be still in recovery mode from the evidence of my attempts today. I thought this would be an easy one. Ref 226, Masks of Mayhem. Moonlit scene of a lake, campfire, tentacles... not to hard, right? What on earth, they are growing like trees in a line in the water. What sort of creature would have them like that... and there's 14 of them. I only ordered 6. Despite multiple attempts, it couldn't get neither the number of tentacles right, nor could I get anything that could look logically like coming from the one beast. Time for a new angle. Let's tell the AI it's a kraken. Ye gods, I've summoned cthulhu! And what's going on down there by the fireside? Do they not see the giant cosmic horror? Heeellloooo, time to start running people. Okay, clearly the AI definition of a kraken is not working. Let's try giant octopus instead. Ah yes, the creature is looking much better now.... but what in the damn hell is it doing in the tree? Nobody warned me about tree krakens. No wonder this thing snuck up on me while I was sleeping. I bet I see the spirits of squirrels after defeating it. Sigh. Okay, maybe where I wrote "coming out of the water" the AI just read "out of the water", so let's try rewriting that a little...
|
|
|
Post by paperexplorer on May 6, 2024 11:46:11 GMT
And now it's levitating. Sigh. At least now I know how it got into the tree. And look, another calm onlooker. And lots of random calamari floating around it. AI clearly went ultra realistic on this one. Okay, one last try... Oh I give up. It's close enough, isn't it? I mean, I got it in the water. A bit. That counts. So much for a simple picture...
|
|
Per
Traveller
AHAHAHA!
Posts: 167
Favourite Gamebook Series: Morris VR, some FF, Chalk LW
|
Post by Per on May 6, 2024 12:26:28 GMT
Did you explicitly tell it to do orange and azure?
|
|
|
Post by paperexplorer on May 6, 2024 21:31:57 GMT
"Night scene"
At least it was consistent in the interpretation I suppose.
|
|
Per
Traveller
AHAHAHA!
Posts: 167
Favourite Gamebook Series: Morris VR, some FF, Chalk LW
|
Post by Per on May 6, 2024 23:16:28 GMT
I guess it just shows the trends in the training material. I thought there might be some kind of forum game here where we make images using the same free model and some additional constraint or two, then try to guess where it's from. It just seems pretty hard to make something recognizable, for instance with Stable Diffusion, which I realize is not state of the art: Prompt: "Inside a tent, a portly bearded merchant is gesturing invitingly towards the left, where his wares are displayed among a clutter of chests. Among the wares are a key, a mirror, a scarab, a flute, an egg and a bracelet. In monochrome ink style."
I had to remove the facemask from the list of items because it kept rendering two figures! I don't know if that was responsible for the time the merchant got merged with a stack of chests, though.
Also if someone finds it noteworthy that the merchant is gesturing to his left, this was just random across all the images rendered. Prompt: "A young man facing the viewer and wearing a tunic is pulling his tunic open with both hands, revealing an amulet hanging around his neck. From the amulet comes a beam of bright light towards the bottom of the image. The young man casts a huge shadow on the stone wall behind him. In black and white ink style."
It may be revealing that the third arm does not stop this from probably being the best I got after at least a hundred attempts and a number of variations of the prompt, some of which had bizarre effects. Thinking at first "tunic" might be too exotic, I tried "robes", resulting in endless images of boys posing strangely in baggy robes. Changing it to "shirt" invariably created people wearing nothing on their upper body. Also "black and white" instead of "monochrome" was my attempt to get rid of the greyscales that seemed to come automatically with this prompt, but it didn't work. Probably I suck at writing prompts though.
|
|
|
Post by bloodbeasthandler on May 8, 2024 17:59:58 GMT
I thought there might be some kind of forum game here where we make images using the same free model and some additional constraint or two, then try to guess where it's from. That is a good idea. And one that it going to result in quite a lot of amusement by the look of it.
|
|
|
Post by paperexplorer on May 10, 2024 7:36:32 GMT
I think making a game of it could be fun. I didn't get either of the above. I learnt a new learning bias today: barbarians all have beards. Apparently. Anyway, beard aside, this should be a familiar character:
|
|
Per
Traveller
AHAHAHA!
Posts: 167
Favourite Gamebook Series: Morris VR, some FF, Chalk LW
|
Post by Per on May 10, 2024 15:15:58 GMT
If anyone wants to give it a shot (here or in a dedidated thread), I used this version of Stable Diffusion. The demo on the Stable Diffusion Online page should use a more recent model but you now need to register to use it, which I just find to be a pain. My preference would be to try to make images in the traditional FF ink style rather than airbrushes and such. When creating images for RPG purposes I've typically appended "in woodcut style" or something similar, since it tends to give me more of the type of image I'm looking for, but of course this also influences what the model's got to work with and it's quite possible there are many better ways, e.g. "black and white fantasy book illustration style" may work. Just don't tell it to do "children's book illustration style" - it'll report you to the FBI.
|
|
|
Post by Wizard Slayer on May 13, 2024 8:34:02 GMT
My preference would be to try to make images in the traditional FF ink style rather than airbrushes and such...but of course this also influences what the model's got to work with... You might find it better to let it do its thing and then convert it to ink after, e.g. to take a couple of examples from here: (Done using Lunapic.)
|
|
|
Post by Pete Byrdie on May 13, 2024 9:24:04 GMT
That makes a huge difference. I could mistake those for FF illustrations. It's long enough ago since I saw the Master of Spiders illustration, if I'd seen that version I'd assume it was the one from the book.
|
|