site banner

Friday Fun Thread for September 29, 2023

Be advised: this thread is not for serious in-depth discussion of weighty topics (we have a link for that), this thread is not for anything Culture War related. This thread is for Fun. You got jokes? Share 'em. You got silly questions? Ask 'em.

1
Jump in the discussion.

No email address required.

Decided to redo the cover art for my novel:

Any strong opinions on which one looks best?

/images/1696055832712833.webp

/images/16960558334619613.webp

/images/16960558340576553.webp

I think the font and colour choices would be better suited to some girly Bridget Jones-y chicklit novel than to a sci-fi novel.

#1 makes me want to read the book more than the other two

Well, it's been updated already, not that I've noticed anyone complain in the past.

#3 because I can take it more seriously

seconded

Scantily clad women still catch eyeballs, though these examples are just the AI being horny af rather than an intentional attempt at my part!

In hindsight, I guess it's not surprising that a machine learning model trained on data largely from the Internet has seen an overwhelming amount of porn, or at least porn-adjacent content.

I wonder if coming up with a less-biased dataset will eventually require something silly like, say, street view to make observations more like reality. On the other hand, humans watching Hollywood movies already see a biased dataset compared to everyday life.

Porn not needed.

An AI trained on female social media accounts and/or street view would also likely reflect the revealed preference of (young) women to be scantily clad and to be sex objects.

#3. Rockets FTW

I want cyborgs, rockets and superheros, which is a tall order for most image models, but DALL-E 3 can handle it most of the time.

There are a few that evoke similar moods, which I could use too

/images/16960980249845638.webp

/images/16960980252711985.webp

/images/1696098025535511.webp

/images/1696098025836889.webp

The second one is pretty cool

Why don't you use several prompts and do the compositing yourself?

I mean, I could, I know my way around SD, and photoshop, but it's a pain, and DALLE-3 is so good it's easier to try multiple runs and pick the best.

Wow, the ladies in that first one are certainly scantily clad. And yeah Dall-E 3 is frickin good. Man can I just go into cryo sleep for 5 years that would be dope.

The man in the first one is scantily clad too ;)

I think the first one has the best composition. Third could work, but the text would need to be placed a bit differently so as not to cover the kid.

The first two are nearly on par, but unfortunately the second one is slightly spoiled by the child’s hand obscuring the superhero’s hand in the picture rather than interacting with it in some interesting way. One of those ways AI can still mess up in a way no decent artist would.

Nice proompt-fu tho!

Thanks! It's going to be a relatively small thumbnail/cover, so it's unlikely that anyone is going to spot the small details.

Bing Image Creator with DALLE 3 is a godsend, it's so much better than MJ or SD at understanding semantics and interactions between objects. In the older models, having a woman holding a child while looking at a painting of a superhero would result in woman painting a child superhero half the time, leaving aside the woman is also supposed to have cyborg legs.

We're so close to the point where we don't even need prompt-fu anymore that I can't take too much credit, but I know that making something like would be a PITA even using advanced tools like Controlnet for SD.

At any rate, I'm no longer sorely tempted to get an MJ subscription. This is at least on par aesthetically, and maybe better.

The subtitle font looks like an Instagram font, I don’t like it.

Understandable, I'm no graphic designer, and that's what I managed to wrangle out of the free tier of Canva. I'm already torturing the crap out of DALLE 3 with the prompt, I don't think it's going to handle extra text on top. If you have any concrete ideas on how to make it better, I'd love to hear them, I'm not fully content either.

You can create any image, drop it into Keynote as a slide, then superimpose any font(s) you like. Then simply export the slide as an image file. PowerPoint can also probably do this. I'm sure Photoshop, etc. can do this as well but I've no experience.

I’d try a few different font varieties.

I don't like the third because the scene makes no intuitive sense. Space suit guardian so far from the space shuttle. Is he the equivalent of a full kit wanker? So for me the two top ones remain and I have a slight preference for the first one.

I'll forgive you for insinuating that the dude is some kind of try-hard LARPER haha, he just has a whole bunch of cybernetics. Maybe he stood too close to the launch last time?

I do prefer the first two myself, I was particularly torn between them.