Viridian's AI Comic Art

[cerberus]

... there is a certain charm to drawings that escapes the computer generated images.

I think I may have to disagree with you there.
Why? Hint: note the name of the forum section and thread.

On the AI images themselves, I agree they are great!

[PA Jack]

... there is a certain charm to drawings that escapes the computer generated images.

I think I may have to disagree with you there.
Why? Hint: note the name of the forum section and thread.

On the AI images themselves, I agree they are great!

Got me there! But they look like hand drawn comics, which have a certain appeal.

[sixgunzloaded]

Some nice renders from the Bikini set

Fantastic! Love their faces! Is it just me, or does it seem like the AI has an easier time rendering expressions (and billowing, messed up hair, too) in comic-style than in photography-style?

[cerberus]

Some nice renders from the Bikini set

Fantastic! Love their faces! Is it just me, or does it seem like the AI has an easier time rendering expressions (and billowing, messed up hair, too) in comic-style than in photography-style?

Oh this is so tempting, I have a compulsion when people start a sentence with "is in just me" to interject with "yep, it's just you" and that's before they get to the or/but part! Anyway to the main part of the question; here I think the key word is "seem", there are two parts to this:

First, a lot of AI images are done by people trying it out on the free online services and entering prompts of the "girl with big tits sinking in quicksand" level of sophistication.
Second, a lot of the photorealistic models are trained on "red carpet"/PR images from the internet of people looking straight at the camera and smiling inanely.

The net result is lots of generated images of girls with big tits sinking in quicksand while looking at the camera and smiling inanely, which I guess isn't that surprising really.

As I'm sure Viridian would confirm, even for AI "drawings" you don't get good atmospheric images with interesting facial expressions and dramatic hair by using the prompt in "First" above. If you want "Surprised" you have to ask for it "surprised, eyes wide open, brows raised, lips parted", it may not work all the time but sometimes it will, those are the ones you keep. Same with hair, I have a file of 40 hairstyles that my prompts pick from when I generate images (I have to say, it didn't include "billowing" but it does now),they add to the diversity and variety of the images I produce. Nobody has ever said "nice hair", it is taken for granted, but I feel it adds to the mood and atmosphere of the image and, like the facial expressions, gives variety. There is another aspect to the distinction between art and photographic, our minds are much more attuned to divergence in photorealistic images and if things get a bit over exaugurated something in us screams "abomination", whereas we are much more accepting and tolerant of art/drawings etc.

[Viridian]

cerberus is spot on with AI photography. The inherent problem is that most of the images are basically model shoots. I've had my fun with AI photography, but there are huge limitations in two ways: what we look for in a dramatic expressive scene actually isn't conveyed very well through our human expression, and more so, the AI isn't trained on sets that have that kind of thing. You can generate really good images that very photorealistic, but as Acidtester frequently pointed out, the expressions look bored and don't match the scene.

Examples with the same prompts in different styles:

Bikiniv2.jpg

Sweaterv1.jpg

Explorerv2.jpg

cerberus is also right on the gap between how most people use prompts to generate fairly generic poses and images, and the more specific and personalised prompts create more distinct outputs. There's a lot more customisation and knowing the right combination of options and workflow steps to create a consistent product that highlights what we want to feature.

AI drawing styles are usually more expressive because there is more diversity in the training sets, which is why they're often more fun to play with since you get those exaggerated eyes and mouths as they scream for help, wild dynamic hair that captures movement and intensity. Photo models tend to be static and studio-based. You have to feed it the right reference images to even remotely get expression.

[cerberus]

Whilst I agree that a lot of the photorealistic checkpoint models produce expressionally limited images most of the time, they don't always. Also, there are some models that are a lot more expressive, one in particular I use a lot is LeosamsHelloWorld. I've posted a few images from this in my thread (didn't want to pollute this thread with my images). Have a look and see what you think!