ai-generated-images

[ai-generated-images]

Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 6.5, Seed: 259, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.10.1

Prompt: ""

One of the more terrifying renditions of a human face I've seen come out of these models. The one in the pink outfit is the centerpiece here, with her head melding into the background, a blob of rough skin with a sunken dark spot where her nose and mouth should be. The man behind her is sprawled out on the rubble, and she seems to be on her way down as well.

[ai-generated-images]

Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 6.5, Seed: 242, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.10.1

Prompt: ""

This one looks like a multi-media work that I would take a liking to at a contemporary art gallery, with the medium listed as something like "Canvas, acrylic paint, cardboard, scrap metal, spray-paint". I think the computer was trying to make a cityscape but failed horribly.

[ai-generated-images]

Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 6.5, Seed: 227, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.10.1

Prompt: ""

It's simultaneously well-lit and abandoned. The place is an absolute mess, and I don't think it ever wasn't a mess; the bar stools seem to be fixtures in the ground but are not even close to being aligned. Perspective is only a suggestion here. Is that a structure on top of the island, or is it a view into a non-Euclidean space? The dimensions of the room are impossible to make out. I like the black ooze coming off of the island.

[ai-generated-images]

Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 6.5, Seed: 196, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.10.1

Prompt: ""

This is the kind of incomprehensible nonsense I really like from diffusion models. It's only barely a portrayal of something.

I see a garden with a taxidermized statue of a vulture guarding a gate to a mountain path from a hummingbird statue. But only barely, all the subjects blend into each other in a way that makes things hard to make out. Trying to tag this one is hard because I don't have any confidence in this actually being an "animal" in a "garden" but it's what I'm most happy with, I think.

[ai-generated-images]

Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 6.5, Seed: 173, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.10.1

Prompt: ""

Two children wading through shallow water. Their forms are particularly horrifying to me. The front boy's posture suggests to me that his face is forced upwards, and he appears to be simultaneously shirtless and yet wearing a straitjacket, as if trapped inside his own skin. His gait seems uneven.

The boy in the back exhibits a healthier posture and stride, and yet has these gruesome red growths, one coming off of his right thigh, and his left hand glowing red in motion. Skin does not cover his entire body, and yet he appears to be in control of the situation, perhaps a caretaker or even a captor of the boy in front.

I love the background in how it is an abstract painted scene, adding to the unreality of the image. The driftwood seems to indicate this was taken from the vantage point of a hiding place, adding some paranoia to the scene. Not to mention how each of their faces is creepy in a unique way. This is one of my favorite images generated through this method so far.

[ai-generated-images]

Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 6.5, Seed: 138, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.10.1

Prompt: ""

Here's another painting. Looks like one of those minimal modern paintings you might see for sale in a Crate & Barrel, but darker, and clearly rendering a scene of a craft on the water with landscape a further distance away. I find the watercraft to be interesting in how it doesn't fit the drab scene, with a relatively vibrant blue and an oddly glowing wake. If that is a wake, anyways.

This one really makes me wish I had a greater breadth of knowledge with genres/styles of painting so I could look into artists of a similar style. Running this image back through this thing "BLIP" gives a description using the names Adrian Ghenie and Jeremy Mann. Both very cool artists, but their styles are not nearly as minimalist as this image.

[ai-generated-images]

Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 6.5, Seed: 87, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.10.1

Prompt: ""

This one is haunting, and I see these models make images like this with some regularity. I like this one a lot. The pathetic flower crown, one eye not quite placed on the child's face, the torso blending in with the background. The child looks ghostly.

I initially interpreted our subject to be female, probably due to the flower crown and blurry earring, but upon looking at it closely I can't help but see the child as male. I think this image is interesting in regards to gender, but maybe that's just me.

[ai-generated-images]

Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 6.5, Seed: 76, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.10.1

Prompt: ""

This one, at a glance, is pretty. Shiny, well lit. Gothic architecture is cool and nice to look at. This image is cool because it's a weird take on it, where the general shape and texture of such cathedrals is present but everything is smudged in a way that is difficult to replicate outside of a dream.

[ai-generated-images]

Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 6.5, Seed: 30, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.10.1

Prompt: ""

I'm somewhat conflicted about output like this. The AI is already quite impressionist with the way it tries to create photorealistic stuff, with the way it smudges things like perspective and form, so when it copies a painterly style, I find it does a pretty good job of rendering an impressionist scene. So we get something like this, which I think is pretty, but I realize that I could easily find art in pretty much any art museum that's better than this. While I like this image, it's not really something that is unique to what the computer generates. A person could and would paint something that looks a lot like this, so the computer's role is redundant.

Sometimes however, we do get something that is obviously a mimicry of impressionism and yet has some bizarre element that sets it apart from "human" art, which is where I find value in what the computer generates. This doesn't have any of that, but I did ramble in an earlier post as an example of what I'm talking about.

[ai-generated-images]

Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 6.5, Seed: 25, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.10.1

Prompt: ""

I think this is a bench, but it very well might not be. This is the kind of thing I might find at a gallery and take to. I like the background, how it looks like one of those curved backgrounds at a photo studio to obscure where the wall ends and the floor begins. I like how it looks like we have a view of most of the object, but we can't know for sure. The way it seems to taper off to the right might indicate that this is basically the entire thing, but the asymmetry tells me that we might only be seeing about half of it. But we can never know for sure. It's kind of grotesque. The surface looks to be made of meat, and the arm rest is broken and fixed with, what, a fish? This stands in contrast to the crystalline center and right leg. A baffling object.

[ai-generated-images]

Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 6.5, Seed: 0, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.10.1

Prompt: ""

Back at it again, this time with a better graphics card. Each 512x512 slop square takes only 2 seconds to generate, wow! I just installed automatic1111's webui and am now going through seeds sequentially for empty prompts and will be posting the images I like. Also going to try tagging them with what I think the image is of, and writing a little bit about each image. It's cool that seed 0 for this ended up being an image I like! What stands out to me about this image is how the male figure blends in with the clay wall. The patterns on his lower half particularly are quite ghastly, like a zombie.

The couple has a chemistry that I like, with both of their obscured faces looking in the same direction, as if they saw a small animal and are admiring it together.

[ai-generated-images]

Hey, I haven't posted many images here in a while. It's been almost a year since my last post, and those posts were scheduled so I haven't really updated this in over a year. I haven't been running Stable Diffusion in a long time and I've pretty much posted all the images that I thought worth sharing. Out of 26,932 total 512x512px .pngs generated, I've curated 399 to post here, so I've only shared my favorite 1.5% of images Stable Diffusion would spit out. As image (and now video!) generators have been getting better and better at mimicking human-made work, I wanted to go back to some of these images to write about why I like them so much, and why I think image generation models can create something worthwhile of considering intellectually.

I think it's important to note that I did not make these images. I did not even supply a prompt other than an empty string; I just let Stable Diffusion do its own thing. It was my GPU that crunched the numbers to put them together, but anybody who downloaded the same model and used the same seeds would have gotten the exact same results. These images were made by the combined effort of however many thousands of creators who had their images scraped and however many thousands of mathematicians and programmers who built on each other's work to make a system like Stable Diffusion possible to run on an old 1080 graphics card.

Now let's take a look at the image reblogged above and decide what the image is of. Clearly it's a painting, but what is the painting about? Well, it's about nothing: nobody made this image, and it was generated without any sort of prompt or guide to tell the computer what to create. But when we look at the image, our brain still figures it's depicting something. It's like seeing images in clouds; nobody arranged the cloud to look anything in particular, but sometimes a cloud just looks like a dog. What does this image look like?

Here's what I see: The point of view is that of somebody standing on the main deck of a ship, looking at the bridge, which is mostly obstructed by some sort of growth. I've always seen this image as portraying a view from a ship due to the water visible in the background in the top right, along with the general shape of the bridge seeming to have steps leading up to it and a red mast jutting out from the top.

I call the obscuring element of the painting a growth, as I see it as some sort of organic structure. But what the hell is it? Maybe it's a massive Eldritch creature attacking the ship; but the scene seems too calm to be so violent. I see it as an abandoned ship, overgrown with rigid vines of some sort.

The bottom center almost appears to be a figure of a woman with a blackened out face. I can sort of make out her posture, hunched forward, hands in her pockets, with the wind blowing behind her. Perhaps she is some sort of siren-like structure, acting as a lure for unsuspecting seamen, but she is far too obscured to be an effective lure. Rather, I see her as a victim that has been amalgamated into some grotesque mass of organic matter.

Some critters explore the odd structure, seemingly unharmed. A cat looks down at the figure, a possum looks starboard, perhaps to jump to the other side of the painting. Their size doesn't make sense, not to mention their presence.

It's a bizarre scene out of a dream, like a lot of computer generated imagery. This one in particular sticks out to me with how the representational scene in the background interacts with the abstract foreground, and how it violates our understanding of space and perspective. It wants to make sense—as has always been the goal with these image generation models—but it fails spectacularly and makes something interesting instead.

[ai-generated-images]

[ai-generated-images]

[ai-generated-images]

[ai-generated-images]

[ai-generated-images]