Show newer

Pretty striking the difference that a byline makes when asking CLIP+VQGAN to generate images.

"Internet infrastructure"
"Internet infrastructure by James Gurney"

new post at!

Tried to generate a chair made of beetle kill pine.

It's a wood with distinctive blue streaks, salvaged from trees killed by pine bark beetles. Turns out it was easier to get CLIP+VQGAN to generate the texture than the chair.

Asking CLIP+VQGAN to generate

"a car driving down a road in monument valley"

versus "A car driving down a desert road in monument valley | dramatic atmospheric ultra high definition free desktop wallpaper"


What fascinates me about generating images with CLIP is that it CAN generate depth and realistic textures, but only if you know how to ask for them.

"A herd of sheep grazing on a lush green hillside" alone

vs with "amazing awesome and epic" added

The problem with taking startup advice from a text-generating neural network is even the advanced ones know more about what sounds good than about what actually works.

In which a neural net prompted with NOAA's Atlantic Basin Storm Name Pronunciations guide further erodes the already tenuous link between English spelling and pronunciation

Generate buzz with these AI food truck concepts.

Remember, GPT-3 scanned most of the internet during training so it knows what's popular!

Botsplaining: turns out it's much easier to get a neural nets to sound confident than to actually be correct.

I used GPT-3 to generate wiki-style articles about Disney princesses, and uncovered the secret powers and items of Snow White

Being vaccinated does NOT mean it's okay to ask neural nets to participate in meme formats about pandemics that didn't exist when their training data cut off in Oct 2019. You will get memes that make no sense. Be smart.

I used a few GPT-3 variants to generate April Fools pranks suitable for playing on oneself at home.

For the format they had to rely on pranks they'd seen during their internet training. Curie, the 2nd largest variant, sounded the most pranklike.

I assumed that if GPT-3 generated pickup lines, they would be terrible because they would be something like human pickup lines. Instead, they are terrible for more delightful reasons.

Gave the lyrics of this sea shanty to CLIP+FFT and had the AI generate illustrations.

It's a brand new kind of horrifying.

Obsessed with CLIP+FFT, which uses CLIP, an internet-trained image-judging AI, to steer a deep dream-like image generator.

It'll do what you ask, but it will do it excessively.

"Frodo Baggins delivering pizza through the mines of Moria"

Gave a neural net some examples from @mondomascots and it responded with some deeply weird mascots.

The Brando, a guesthouse in Tokyo, is represented by Garapon-kun, a shy yet terrifying Jack-in-the-box that likes to play with balloons.

Makuo, a sea anemone from Kagoshima, can see the future. His antennae are mittens, but he has eight toes on each hand.

Show older
Wandering Shop

The Wandering Shop is a Mastodon instance initially geared for the science fiction and fantasy community but open to anyone. We want our 'local' timeline to have the feel of a coffee shop at a good convention: tables full of friendly conversation on a wide variety of topics. We welcome everyone who wants to participate, so long as you're willing to abide by our code of conduct.