• Welcome to Autism Forums, a friendly forum to discuss Aspergers Syndrome, Autism, High Functioning Autism and related conditions.

    Your voice is missing! You will need to register to get access to the following site features:
    • Reply to discussions and create your own threads.
    • Our modern chat room. No add-ons or extensions required, just login and start chatting!
    • Private Member only forums for more serious discussions that you may wish to not have guests or search engines access to.
    • Your very own blog. Write about anything you like on your own individual blog.

    We hope to see you as a part of our community soon! Please also check us out @ https://www.twitter.com/aspiescentral

Learned more stuff, made more stuff

Misery

Amalga Heart
V.I.P Member
Okay, so this is sort of a sequel to my previous big topic. In that one, I was doing some sort of beginner-level art using AI... nothing hugely complicated, and definitely flawed. Since then, I've learned a lot about all of this.

I know there's this idea that with this type of thing you just feed the thing some words and BAM you get a masterpiece, but... no. No sir, that aint how it works. Unless of course you LIKE getting messed up results that really arent quite what you asked for.

Currently, I have access to three separate AIs: Midjourney (what I used to make the previous things), DALL-E, and Stable Diffusion. Each works differently, is good/bad at different things, and so on.

But those arent enough, not if I want to REALLY make something. So, in comes Photoshop... which I'd never used until recently. That's bloody complicated and required a lot of tutorials and such to even get started to do what I wanted. But in the end, I could get MUCH better results... but also results that were what I wanted, rather than what the machine THOUGHT I wanted. There's little need to settle for the crude stuff when you can take control yourself.

First of all I want to show the biggest thing I made. Before that, here's the main image I'd made last time... the one that really stuck with me and took the most work to create (took about 3 hours of arguing with it to create this):

Misery_Scary_young_marionette_girl_hanging_from_the_ceiling_lon_55bd4aa4-8154-4f0d-8b92-dd8de2...jpg


This was my favorite of all of them, and the one that the most effort went into. Basically, I wanted to have a character that was all my own... but I'd never tried something like that before. So, this new AI stuff was the perfect way to try it out for the first time.

Now, it *mostly* got the details... decently right. But, there were some artifacts of the process. Firstly, her forehead is all messed up... I couldnt figure out why. And second, what the heck is that bright puffball at the bottom? I dont know what that is... but it sure aint supposed to be there.

But on top of that, I wanted to take this further. I wanted to get an expanded image of her. What would a full-body image look like? What would I like the rest of her outfit to be? And I wanted a better background to it too. I already had an idea as to what sort of personality and such I wanted for her, so standing in the weird spooky void just super close to the "camera" wasnt doing it.

So... enter Photoshop. And way too much time learning some of the basics.

The final result was this:

SophieInLibraryFinal.jpg


I want to stress that this took all. Freaking. Day. A long, complicated, difficult process that involved numerous things.

Firstly, I needed to get her "expanded". For this, I brought her into DALL-E, which has that feature. And that means ARGUING with it. The thing is... weird. And not particularly bright. I went through about a bazillion wonky iterations of Sophie (which is her name) in a wild variety of screwy backdrops. At first it put out a series of images where her skirt was all stretched out to the sides for some reason, her hands were... not hands, and she had this weird white aura around her, while standing in a place that definitely wasnt quite what I was after. About a bazillion very specific alterations later, and I had her.

So, she needed a background. I wanted a library... but a just slightly surreal one. The result of that was actually the right half of that background. The LEFT half was an expansion of that, which went through many iterations of its own. You cant see the seam because she is standing right in front of it, but the seam is flawless.

However, there's still issues. the front of her shirt is a bit odd. There were a lot of artifacts there, MOST of which I got rid of using photoshop. But I couldnt quite get it the way I wanted... there's a lot more to learn, you see. I did fix her forehead though.

Again, this took all day. I needed multiple tools including the mighty Photoshop itself, a long series of confusing tutorials, and so on... but I did it! I was tired of the entire bloody computer by the end of it.

With this, I've learned that this process... if I want to REALLY get results... takes a lot more work than I thought. But the more I learn, the more I can do with it, and the more creativity I can inject into it.

So, let's now get to some of the other wonky things I've made recently.

Finished_GirlJ.jpg


This one, again, took forever. A variety of things came out wrong and I had to learn even more stuff. Had to fix plenty of things on her face and so on. But I managed it! Eventually.

Let's see, what else...

FoyerFinished.jpg


This is actually a photo from my house. Sort of. Most of it really is my actual house. That open door there should lead to my father's office. But not in this image it doesnt. Instead, it now leads to... er... wherever that is. Took some doing, this one. Not as hard as the first ones, but still, not easy.

Here's something funny though:

eeeeeee.jpg


So, this happened when I was trying to fill in that door. I... I'm not sure what went wrong here. Sometimes when one of these iterations goes bad, it's small bits of weirdness. But other times, it's a giant cursed mess. Out of curiosity I went on and pushed forward with that one, to see what'd happen. I aint gonna fill this topic up with random nonsense, but... boy did it get weird.

Next up: Dogs.

Misery_close-up_pencil_line_drawing_of_a_well_groomed_Yorkshire_ca7e102e-6ff6-4914-ac53-2906af...jpg

DALL·E 2022-10-01 18.06.16 - A vibrant closeup photograph of a well groomed Wheaten Terrier si...png


The first one is my grandma's dog, a Yorkshire terrier named Daisy. That one took a lot of work and a million messed up, cursed iterations. After quite awhile, I got something that looked exactly like her. This image actually had a purpose: We put it on canvas and gave it to my grandma for her birthday. Unfortunately, the dog is quite old and her health hasnt been doing so good... we know there's going to come a day when she's gone, so I wanted something special for grandma in honor of Daisy. So, line drawing of her face.

The second dog was an attempt at getting one that looked like one of my dogs, Dingle, a wheaten terrier. It looks MOSTLY like her... but it's kinda fuzzy. Again, lots of messed up things leading to that.

Next, whatever this horrid thing is:

Misery_A_screenshot_of_a_horror_game_on_the_original_Playstatio_26270115-0ed5-406d-8215-9bd91a...jpg


It's October. That meant spooky stuff. I went through a variety of images in an attempt to get something inspired by old PSX horror games. Many of the results were... strange. This one was the best one, upgraded and touched up to get this result. For other variants of this, I tried to get it to look like an actual videogame screenshot, complete with a UI and everything. It did not like that idea. What cursed screenshots those were. I also tried to get it to do one like an NES game, but... hoboy does it not like that idea.

One more for now because it's late and I need to get out of this chair:

DALL·E 2022-10-01 17.23.51 - green red huge room in a labyrinthine old house, dark lighting, m...jpg


Nothing super special here. I wanted a sorta weird indoor image of a labyrinthine mansion with a red & green theme. There's actually a long series of these, which I guess are all part of the same place.

So, that's that for now. I'd show more but again, it's late, and I have to resize and convert the blasted things to get them into a state where the forum here will let me upload them.

Maybe tomorrow I'll show some of the wonky failures that were left on the cutting room floor.
 
These are incredible!!! Holy crap! They are so beautiful.
What program are you using to make these?

Side tangent, your house looks weirdly similar to my previous house, lol... we had the same exact type of railing and stairs in a similar position. Was your house also built in the 80s, perhaps?
The house I lived in before that was 100-something years old and didn't look like that at all. But I have lived in a lot of houses in a lot of different places.
 
Wow some fascinating images. Why is A I so slow to work with I wonder? Just very complex I guess. You got some great results now you are starting to understand it.
 
Wow some fascinating images. Why is A I so slow to work with I wonder? Just very complex I guess. You got some great results now you are starting to understand it.

It depends on how you're using it. Individual generations are not slow. Like, if I want just a super quick image for fun, I might say.... uh.... "a bear trying to eat a sandwich while surfing" and after like 10 seconds, I get this:

DALL·E 2022-10-06 07.51.57 - a photo of a bear trying to eat a sandwich while surfing.png


It's kinda crude though. The prompt is *very* simple, there's no special commands embedded in it (used to take control of special features that a given AI might have, usually with Midjourney, not so much with DALL-E), there's no other segments to the prompt, and I thought of the prompt in like 3 seconds. No effort.

If I'm *really* trying to get something good, it takes forever for a few reasons. Firstly, just making the bloody prompt takes ages. You wouldnt think it would, but it does. For instance, I saw a really amazing image of a spaceship done on Midjourney recently (I cant find the bloody image though or I'd show it here), it was *really* good, and when I looked at the prompt, what I saw was, and I quote:

"photo of a futuristic spacecraft, picture is taken from the font and above with a distance so all ship fits into the view, Epic light, black reflective glass hull, aggressive look, advanced Red shining led stripes on the wings, round futuristic blue led war logo, four massive engines in the back of the ship with blue and yellow light coming out from the exhaustpipes behind the ship, advanced weapon systems in metalic colour, lots of details all over the ship, picture with blueprints, infographic, technical document, high details, hyperrealistic, realism, ultrarealistic, photorealistoc, photorealism, super realism, insanely detailed, highly detailed, super detailed, finely detailed, glitch art style, slender young teen girl, Tom of Finland, Saudek, Francesca Woodman, Carolee Schneemannphotography for SavageXFenty, hyperrealism, hyper detailed, by ayami kojima, tom bagshaw, yanjun cheng, artgerm, wlop, krenz cushart, gweiz, Thomas Kinkade, Gerald Brom, cinematic lighting, visual clarity, 200mm, UHD, 32k, 16k, 8k, 3D shading, Tone Mapping, Ray Tracing Global Illumination, Diffraction Grating, Crystalline, Lumen Reflections, Super-Resolution, gigapixel, color grading, retouch, enhanced, PBR, Blender, V-ray, Procreate, zBrush, Unreal Engine 5, Cinema 4D, ROMM RGB, Adobe After Effects, 3DCG, VFX, SFX, FXAA, SSAO,Temple, --ar 16:9 --test --creative"

You dont come up with a prompt like that in 3 seconds. Chances are, whoever was making that ship took hours to get to that point, as you have to constantly experiment and be ready to entirely redo things. There's actually an entire site dedicated JUST to helping you build prompts due to how complex they get. Now, none of mine were as complex as THAT one since I'm just starting out, but even mine still take awhile to really get going.

And lastly, there's the fact that it's an iterative process. That first image of the violet haired girl for instance, well, this is some of the images in the archive that led up to her:

soagidh.jpg


There's a couple pages of these, which are varying levels of crude, insane, or both. They get grouped into 4 during initial generation and you work with them from there, that's why some are clumped up. It makes sense when you use the thing, they've done a good job with making these easy to work with (mostly).

The dogs in particular, I have *many* pages of those. Too many.

And lastly, it's because after the first few, I wasnt using just the AI anymore. After bringing photoshop and other things into the mix, the process slowed down a whole lot.

Also yes I have a lot of browser tabs.
 
These are incredible!!! Holy crap! They are so beautiful.
What program are you using to make these?

Side tangent, your house looks weirdly similar to my previous house, lol... we had the same exact type of railing and stairs in a similar position. Was your house also built in the 80s, perhaps?
The house I lived in before that was 100-something years old and didn't look like that at all. But I have lived in a lot of houses in a lot of different places.

The AIs used, are from these sites...

DALL-E: DALL·E

Midjourney: Midjourney

And then of course photoshop... I doubt I need to explain where that comes from... and other programs like Pixage (can be found on Steam).

Anyone can use the AIs to make stuff... try it out, it's not at all hard, and you get some free uses of them (but you have to make an account for DALL-E, and you actually need Discord to use Midjourney for... some reason, I'm not sure why they went that route). After the free uses run out... these get pricey.

However, the third AI, Stable Diffusion, can actually be run locally on your machine (so, no need to pay), if your machine is a beefy monster. It is impressive in its own right but installation is a giant pain (I havent gone and done that yet).


As for the house, it was built around 2002 actually. It's a big weird place.
 
That's cool seeing your transformation of Sophie using Photoshop. By using the different tools and procedures, first showing less of a closeup of her allowing you to see her entire attire, and by creating a background, this helped first create a bit more of her personality there before really perfecting the other details in the closeup. Then things became clearer there, and the creation seemed more picturesque and modern and less surreal there. You created a masterpiece there! Loved the other creations too, and the background info there too! Nicely done.
 
Sophie looks to be real, but then l think well, maybe she isn't real. Photoshop can be frustrating for me because you need to assign things to back or front and as layers, and as effects, the choices felt overwhelming and therein is what makes it a great software to dabble in. However you can spend an entire day just refining your image. I took a photoshop class in community college. Putting actual words into the image is another layer that you add, so l just centered my type because it was so detailed and could end up anywhere if you didn't assign it a window.

Misery, you have a lot of patience to refine your image to success. She looks like she has a doll head on her lap, but l understand you don't want it there. Her forehead just made me pinch myself and think, maybe she is real?
 
Last edited:
Seriously impressive work, well done. And yes, I'm jealous, I don't have an artistic bone in my body. I played around with image manipulation a bit a few years back but only enough to get an idea of how things work.
 

New Threads

Top Bottom