Category: 05

starry – TextSynthesis

Inferkit:

I felt nostalgic about spending summer at home when I was a kid.

Of course, back then, it was just life as usual for us.

(that is, work, eat, sleep, repeat… or so I thought) I wanted to go outside and look at the trees.

I felt like I was actually “at home.”

It was as if all of life had happened to someone else and I had gotten a break.

(Don’t worry, I’ll stop by my parents’ house and get a dish of my favorite ice cream to ease my pain.

I promise.)

(We all need a pick – me – up every now and then)

The feeling of home was forever taken away from me, it would seem.

I thought InferKit was more interactive than Narrative Device, and the texts it generated often had more variety and I enjoyed the prose better. With Narrative Device I found it harder to use abstract themes because it wouldn’t generate a lot of sentences.

starry – ArtBreeder

This was pretty fun but I thought once the amount of genes got past 4 or 5, the outcome became significantly less controllable. In the second image I had pomeranian as one of the genes but I kept getting images of babies.

starry – Pix2Pix

It was pretty interesting seeing how the algorithm would interpret my edges, since it was unpredictable, and I liked seeing how a doodle could be easily transformed into a fleshed out image.

merlerker-VQGAN+CLIP

prompt: “nike muji spacecraft collaboration rendered with houdini”
noise seed: 05
iteration: 100
result:

Colab was easy enough to get running, though I didn’t get the diffusion to work (I think because it requires a starting image rather than starting from noise?). It’s really interesting to see the images emerge through the iterations – I found my favorite outcomes were around iteration 100 or 150, when the image had just started to materialize (perhaps because my prompt lent itself well to textures, it didn’t take long to get there), and beyond that it was further refined but the different between subsequent iterations was hardly discernible after a point.

prompt: “nike muji spacecraft collaboration rendered with houdini”
noise seed: 2046
iteration: 50
init image: dragonfly-sliderule_2.jpeg (from ArtBreeder)

result:

 

prompt: nike:30 | muji:30 | spacecraft:10 | rendered:15 | houdini:5 | color:10
noise seed: 2046
iteration: 50
result:

merlerker-TextSynthesis

Find the oldest tree you can. Sit with it until one of you dies first. Plant a new tree, but take some time to remember the old one. Set all of your tools beside the tree. Rest the knife in the stump. Wait until your child comes home. Lay the children down. Wait until your son-in-law returns. Tie a turban on your grandson, too. Take turns drinking wine from the neck of the bottle. Open one more bottle and pour it on the roots of the tree. On your deathbed, the tree might bear fruit for you, too, and that’s beautiful. Once upon a time, two brothers walked along the road with one of them weeping. “Why are you crying?” asked the other. “Because I have no husband and no children!” came the reply. “Then I shall not weep,” said the other. “You are not to weep. I have no family, either, and if you have no family, you shall have no sorrows. But when you have no sorrows,” said the first brother, “go to the garden of the gods. They are very kind to strangers, and they will let you pick as much fruit as you like.” I make plans. I look at the forecast.

I’ve been wondering a lot about you; I can’t really understand you yet. It’s not that I don’t understand humans, but I’m not sure I’ll ever understand the point of a computer program. They’re not people. And don’t get me wrong, I like people, but I can see how you’d say they’re different. Humans say: “I’ll be gone for three weeks. Do you think you can survive without me?” Computer programs say: “That’s a really dumb question. I wouldn’t need to live for three weeks without you.” That’s what I think. I think humans could survive without all the cutesy features a computer program has, but they wouldn’t really be alive. I would miss you.

How do we add rifts, as pauses and statements of intention, to interactions that have become seamless: relating to time, accessing information, and moving through the world? Paint to make: …covers my hands and cuts the sleeves off my white shirt, exposing my skin. I can feel the material between my fingers, but can’t reach my skin. Maybe it will bruise. The students will say that they don’t care, that it doesn’t matter. I find my purse on the floor. I put it back and go into the hallway. The white powder dusting the room gives off a golden, sun-bleached glow. I don’t dare stand in the sun. The students have left to watch soccer. A teacher plays her tuba outside. I go into the lab and find the camera. This camera creates objects and scenes. Maybe this is how we talk to each other? How we take the pictures that capture time? I add a red brick to my platform. I forget to clear the model. The students will ask why. What is the significance of what I do? How do we create gaps in the seams that make objects feel whole? How do we clarify stories we tell each other? How do we let each other know how we are feeling?

~~~

I preferred playing with the InferKit demo, as you can kind of launch the style/genre of writing with the first sentence. It seems that it’s trained on quite a range of content, as I got it to produce text that sounded like fiction, a (bad) thesis, a calendar event, and a programming tutorial.

merlerker-ArtBreeder

The interface takes a little getting used to (I didn’t realize the “Parents” I was initially selecting were actually generated images themselves) but is addicting once you get going. Fun to think of inputs that visually kind of work together (or not) but come from very different places. For some reason I wanted to start from real rather than generated images, and kept it to 2-3 inputs so their sources were still somewhat recognizable.

merlerker-Pix2Pix

I spent a few runs learning the “rules” of the kinds of forms and details the model responded to (narrow paths for zippers, loops like in strappy sandals, wavy bottoms for soles…) then tried to figure out how to push them. For facades I followed the rules for the most part, here trying to see if I could believably render a facade upside down. For shoes, I played more with how the language of “shoe-like” contours can be composed in unexpected ways. It seems the more interesting results come from composing familiar details in new ways, as the model responds well to details and composes the image however the input dictates – and this gets at a balance of “crystallization” (human-recognizable details) and “chaos” (unexpected compositions).

kong-ArtBreeder

It was really fascinating to see how small clicks are able to completely alter the images. It was also interesting to see how my final outputs are mystical and ocean-like when original images are composed of unrelated objects, such as traffic lights, flowers, steel boxes, etc. Through this platform, I have experienced an iterative process in creating, where I swift the direction of my final output based on each selection of the image.

kong-Pix2Pix

This tool allowed me to eliminate my preceding knowledge of cats and buildings while making these drawings. I felt that this tool gets me to be more creative with the final output. It was interesting to be able to observe the output changing by minor changes that I make, such as adding a new rectangle.