Saw Shadiversity make a video on ai generated swords, many of them are absolute ambomninatioms. Want the link?
2 Likes
My latest pursuit that made me spend over 40 night cafe credits and after running out, I ultimately still had to use hundreds of Dream generations for is Jasper. Jasper is an Incubus demon and heâs my MCâs father.
I wanted to share to show how much effort takes to create these images.
Before I ran out of credits, the best Jasper images I got out of that fiasco were created with Dalle, not the most imaginative but they sort of have his vibe:
Not bad, but not the Jasper I imagined.
So I tried to create Jasper out of a prompt in Dream (no luck), and I tried to create him out of one of the many images Iâve created of my MC. Unfortunately, every time I mentioned âdemonâ in the prompt, Dream kept interpreting it as an angry demon face and that doesnât represent Jasper at all. Heâs supposed to be seductive and ageless. Heâs 300+ years old but should look fairly young because his race doesnât age. And he should look not too different from my MC. I have hundreds of my MCâs images so he doesnât have one definite face but still, thereâs a certain feel I have for what heâd look like and it took a long time to get his father.
Eventually, Iâve used a mashup of my teenage MCâs image plus a demon image derived from that image and then mashed them together with pixlr/e just to get my MC + horns.
and
mashed to create
And after many regenerations in Dream, I finally got these two which could work as Jasper.
Still not sure which one should be the official Jasper. Both could work, depending on how demon-like I want him to be, but overall, Iâm pretty happy with the result.
It was just frustrating how long it took to get there. Fantasy races remain to be really difficult.
Hmm. I just noticed the fake writing over one of the Jasper images. I could overlay them, I guess, to get rid of that. Sigh. No matter how much time I spend on getting AI to do what I want it to do, it always takes manual image manipulation for the finishing touches. AI, you still got a long way to go.
2 Likes
He looks like a satyr
Yeah, I can see that. No hooves though. He has a demon tail instead.
1 Like
Full Jasper.
Hungry Jasper.
2 Likes
It reminds me of Buffy vampire lore (face changes as they get their vamp on).
Itâs certainly something to consider - should Jasperâs appearance change? But he wouldnât become an ugly demon.
The lore I created is basically that Incubi are everyoneâs type and everyone finds them attractive.
I was considering doing that without any facial changes other than them knowing how to adjust their seduction methods based on whoever theyâre trying to seduce.
I wrote one scene like that already and I think it works, but I havenât had anyone read Jasper scenes yet so Iâm not sure if it comes through to anyone other than me. 
1 Like
I usually play it as partially psychic. You see what they project, and while the true face may be pretty, its more noticeable that that ainât kosher.
Which means if you write a person with near perfect imperviousness to being tricked like that, they would see the real face. Then you could decide on adversary or âlove you no matter whatâ.
2 Likes
I just made my incubi and suucubi have a true form that looks like a demon ghost chihuahua.
2 Likes
Apple headed yappers?
1 Like
Blood-sucking apple headed
2 Likes
"Circus beyond the stars - made with @NightCafeStudio
#aiart #nightcafe #digitalart
How are everyoneâs generations doing lately?
Iâve been working on another project and naturally that means I had an excuse for a whole bunch of images.
So I had fun with that.
But then I got into a conversation on discord and I guess anti-AI sentiments run pretty deep. Itâs become one of those sensitive topics like politics that are better left untouched. People get really passionate about their anti-AI stance and youâre painted as the bad guy if you say anything otherwise or make an argument that doesnât support their stance.
Honestly, Iâm in a strange place where my mind is mostly made up about the issue but thereâs some conflict Iâm not sure about yet.
My philosophy is
You canât stop it. Instead of all this outrage to stop AI artwork, we should be spending our energy fighting for the actual issues surrounding the topic, and thatâs if someoneâs work is being used to train AI, then that artist should be compensated.
It really is that âsimple.â
Of course, itâs not that simple because thereâs not one AI model. There are many and theyâre being used for commercial purposes already. The cat has been let out of the bag long before anyone even thought of how this can be done.
So I agree with the approach that Shutterstock took which is to use their own stock to train their own model and give the contributors a cut. How well will their approach work? How fair will they be? Time will tell but at least theyâre doing something.
What about the existing models? No clue how this can be contained. Maybe the future will include the âresponsibly sourcedâ AI models vs bootleg unethical models.
In an ideal world, I wish that there was an initiative to create a model and a platform in one that would allow artists to upload training art and then whenever their art is used, they would get a cut and credit. And then make it so that thereâs a way to prove that art is inspired by an artistâs style for two reasons:
- To credit the artist.
- To ensure that AI art isnât mistaken for the actual artistâs work.
If I was an artist, I would totally be on board because itâs passive income! And being an artist is hard already. Making money from art is hard already.
I do not believe that AI art substitutes or ever will substitute what a human can do. I do not believe it steals customers.
It might steal some stock photo revenue, yes. But if we put safeguards that ensures that artists get paid a cut then that will be solved.
But does all AI art constitute stealing revenue? No.
And my argument for it is this example:
I got myself a Roomba recently. It cleans my house. Itâs a cool tool.
Did this Roomba steal anyoneâs job? No. I wasnât hiring anyone for this job. When I didnât have time to do it myself, my house was simply dirty.
Same with art. I wasnât hiring anyone for my art before. I tried to do it on my own. Results were mediocre because Iâm not that good. So I either had bad art or none.
But by using tools, I now have a slightly cleaner house and art thatâs not only better than my own but more of it.
Before AI, I wouldâve hired someone to do letâs say a commercial book cover.
After AI, I wouldâve hired someone too.
2 Likes
Looks great.
You know whatâs really cool?
That the made up Japanese writing look realistic to me. Itâs like if I (someone who doesnât know Japanese) tried to write in Japanese and made up a bunch of characters.
It really is good at acting like humans when copying and learning.
I used it to create a bunch of fake language letters. It looks so realistic and AI thinks itâs giving me actual letters. Of course itâs gibberish but the point is that it should be gibberish. I donât want it to be real.
1 Like
The people who are using AI generated art for their projects likely donât have the money to spend on comissions in the first place. Itâs like book pirating, you donât lose money because the person pirating it would have never bought it in the first place.
I would love if my art could be used to train AI-I already like to use refernces images to train it to give me what I want.
1 Like
I am beyond excited because I managed to create images of my two mains that go together, are of the same style.
Hereâs Sam
And hereâs Ian
1 Like
Wrow.
Have you ever messed around with using a starting image? I made Nightcafe make hyper-deatiled artistic portraits based on refernece images of people who are decidedly not that.
1 Like
Iâve had no luck with starting images in NightCafe. Most times it just distorts them.
Unless itâs just the type of prompts I use.
However, starting image works great with Dream. The only drawback is that it canât be upscaled. Some of the issues can be fixed with an editor though.
I dealt with that the other day when trying to create Ian in Dream. If the face was okay, then the hair wasnât, or the setting or the position or facial expressions or clothes. Sometimes, it blatantly creates them off-screen. lol.
After many efforts, I created this one in Dream. I thought, okay, weâre getting close. I downloaded it and uploaded it back as a reference image.
Then I used it to generate a bunch of images until I landed on this one. I figured, thatâs as close as Iâll get. (not sure whatâs going on with his sweater, itâs like he was pulling it over his head and didnât straighten it after, lol)
it came out washed out and the eyes are freaky so I put it in an editor.
Better colors, some touchups and realistic eyes. Eyes make so much difference.
Something is still off about the eyes, their position or something, itâs bothering me. I might return to it some other time.
I did that with another of Samâs images from NightCafe.
She came out really cute here but the eyes were too blue and⌠ghostly.
Fixed the eyes and did a couple of minor touchups. Now sheâs perfect.
3 Likes
What are some good (free) AI programs / websites to use for high resolution images (above 1200x1200)?
I experimented with making some character portraits on a few websites, but the results missed the mark by a few postcodesâŚI might just be using the wrong inputsâŚ
Oh, I have also encountered Luminar AI and Aurora HDR AI-based photo editing software. Any experience or opinions with this stuff?
2 Likes
On NightCafe you can upscale up to 8k.
The above image of Sam is upscaled to 3k.
I think Midjourney allows upscaling too (not free though).
In my experience, it takes a good number of generations to get results. First, you tweak the prompt and settings until you start seeing something good. Then continue tweaking, or if you land on a good combination, just keep reusing the same prompt.
Itâs interesting. Sometimes a little keyword change can make a big difference. The trick is to not waste too much time on the failing prompts.
I havenât tried those.
2 Likes