Star Trek AI art thread

Well, the arrowhead on the left side of the chest is far and away the most consistent element of Star Trek costuming. It’s probably the only thing that’s obligatory for human designers, too. Colors, rank devices, cut and style all change, but the badge stays constant.
 
I think AI art is great as another tool in the toolbox. I use Poser mostly because I built up the assets but I also use Faceapp, Toonart, Toonme, Picsart, Thispersondoesntexist.com and lots of other AI stuff. I used the AI voice generator for James Earl Jones to narrate a trailer for a Star Wars Luke Skywalker/Mara Jade fanfiction text aloud. This shit is wild! I love the ability to use all these tools. On Deviantart there are tools whining about it and talking about stealing art and then in their or their mates galleries/favourites are stuff like Star Trek novel covers or screenshots or Photoshopped art that isn't their own and I find the hypocrisy nauseating. The only AI Trek art I tried on Picsart was "Kirk" and "Banana" but I just got a really tanned Kirk-ish action figure-looking motherfucker so maybe I need to try something else.
 
yea i cant stop playing with it after all....:)

DATA_32nd.jpg

airiam_32nd.jpg
 
Dream.ai is fascinating. I've made some random stuff and some has been hilarious. I like using a film poster and then giving a description of a completely different film, like I did The Empire Strikes back and The Wrath of Khan and vice versa. I did Tuvok and he had little furry ears. James Earl Jones as Picard came out good. It's a good tool to make new backgrounds, because for my own stuff in Poser that's more toonish there's already an unreality to it anyway so having weird backgrounds really fits.
 
I saw this on phys.org recently---and it may wind up being a help to those in CGI
https://techxplore.com/news/2023-02-classic-math-formula-looping.html

In an article in the Proceedings of the National Academy of Sciences, Adriaenssens's team presents a method that describes creases as part of a continuous system rather than a separate section of calculations. The new method not only allows engineers to conduct calculations in one effort, it also accounts for different degrees of folding in a structure, from sharp tent-like creases to more gradual curves.

I wonder if that might be a hold up for AI Trek-ship art---note that it doesn't do sharp nacelle bends very well.

GPT from scratch
https://www.nextbigfuture.com/2023/04/karpathy-makes-gpt-from-scratch.html
 
Last edited:
That's not really how these images are generated. When a human looks at something, we extrapolate out an object from it. The computer sees like a baby; it's all just blotches of color, it doesn't understand that those colors indicate contour and structure. The computer lives in a world of abstract art, and it's faking an understand of what images represent by having had them tied to words when it was trained. That's why corners of nacelle pylons, or fingers and teeth, or windows in buildings look sort-of correct from a distance, but if you examine them closely, you'll see lines of perspective don't meet the right way, or objects split and join in surreal ways like the impossible trident. It's not actually calculating any kind of bend, it's just an arrangement of colors it associates with certain arrangements of letters.

The computer isn't starting with an idea of what a starship looks like, or a 3D model, it starts from a bunch of random static, and then shifts and twists it over and over and over again until it looks more like the images it associates with the prompt words you've provided from it's training data, a tiny bit at a time, and ends up with something a human would recognize. Or a weird monstrosity.
 
civitai-model: RealisticVision.1.3
sinkin2Realistic-Vision-V1.3_deanna.jpg


and civitai-model: deliberate
sinin_deliberate_delaneysisters.jpg


so i'm done playing with different models of stable diffusion for now,
it is very interesting. i know that both musicAI and animation/videoAI are coming full speed.
we have the chatGPT for stories and with picture/graphic making AI's like different stable diffusion models...
where will this lead?....

i am almost certain, within a year, i can push a handfull of buttons and "make" a new star trek movie.

What words or phrases did you use for the appearance of the faces? AI-generated human faces usually induce a gag reflex from me as there is always something just "off" about them, but these don't. They look very nice.

Kor
 
What words or phrases did you use for the appearance of the faces? AI-generated human faces usually induce a gag reflex from me as there is always something just "off" about them, but these don't. They look very nice.

These was made before i downloaded the Automatic1111 on my harddisk, i used the online "sinkin" site.
Unfortunately i dont remember the prompts. i tried run them through the png info in stable diffusion now
but i realise i only have them as jpg files. i need to check my backupdisk, i might have them there as png's.
i'll check tomorow, it's midnight here now i need some sleep.

but i checked the date on the jpg's and these are some prompts made right after, it is most likely very similar:

----
analog fujifilm photo of one 29 year old woman IN STAR TREK UNIFORM, background is a starship, pale skin, wild hair, (upper body), dark hair, detailed skin, detailed eyes, realistic eyes, 20 megapixel, detailed skin, detailed face

Negative prompt:
(deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime:1.4), text, close up, cropped, out of frame, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck


Steps: 27, Sampler: Euler a, CFG scale: 9, Seed: 110433842, Size: 512x512, Model hash: 10642fd1d2, Model: realisticVisionV13_v13Inpainting, Conditional mask weight: 1.0

this is the above prompt in deliberate
00000-110433842_deliberate.png


and the excact same seednumber and settings in realistic vision
00001-110433842_realisticvision.png


this makes it a possibility that sinkin are using a checkpoint merger, possibly.

---

i will check my backupdisk, but if i dont find them there it's pretty much impossible to reproduce them without a seed-number from the png's.
 
@Kor im sorry. was looking for it now but i dont have the png's, however... i found a strange one.
delaniee00128-1814805366.png


after looking at the pnginfo the prompt here doesnt make any sense, i tried it but only got robots.
but you can go at the sinkin site and try the seednumber there. I used up all my free credits there and i dont
want to pay for more. Here are the seednumber: 1814805366

and here are the prompts and settings extracted:
----
hyper realistic portrait of a robot astronaut, star trek insignia, fantastic concept art, portrait, well lit, intricate abstract. cyberpunk, intricate artwork, by Tooth Wu, wlop, beeple. octane render,in the style of Jin Kagetsu, James Jean and wlop, highly detailed, sharp focus, intricate concept art, digital painting, (looking in the camera:1.2)

Steps: 39, Sampler: Euler a, CFG scale: 14.5, Seed: 1814805366, Size: 768x768, Model hash: 10642fd1d2, Model: realisticVisionV13_v13Inpainting, Denoising strength: 0.35, Conditional mask weight: 1.0, Mask blur: 4
----
i couldnt reproduce this locally with realisticVision or with deliberate so i must have done this at the sinkin site.
you can try run it through there but change the prompts to the first one i gave you, you might get lucky.
i got this locally with the deliberate-model when i used the seednumber and that first prompts:
00005-1814805366.png
 
Back
Top