@judeswae@toot.thoughtworks.com
@judeswae@toot.thoughtworks.com avatar

judeswae

@judeswae@toot.thoughtworks.com

Europe Head of Social Change and #MastoAdmin at https://toot.thoughtworks.com/@thoughtworks

Passionate about computational art, responsible tech and hacktivism.

Opinions, my own.

This profile is from a federated server and may be incomplete. Browse more on the original instance.

judeswae, to ai
@judeswae@toot.thoughtworks.com avatar

#Ai image generators will be great at making #ikea carpet motifs and waiting room decors.

judeswae, to StableDiffusion
@judeswae@toot.thoughtworks.com avatar

#StableDiffusion
Prompt: nothing
Negative prompt: everything
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 0, Size: 512x512, Model hash: 88ecb78256, Model: v2-1_512-ema-pruned, Version: v1.7.0

judeswae,
@judeswae@toot.thoughtworks.com avatar

@ralen what do you suggest?

judeswae,
@judeswae@toot.thoughtworks.com avatar

@ralen Here it is at 50 steps. Same model, same prompts, etc.

I don’t find this significantly better than the 20 step one. Sure, some things went slightly in a different direction. But in terms of quality, not a big bump.

judeswae,
@judeswae@toot.thoughtworks.com avatar

@ralen It's the official SD2.1
The prompt and negative prompt are intentionaly limited.

judeswae, to llm
@judeswae@toot.thoughtworks.com avatar

Prompting -chat-7B: What is your context window size?

Response: As a responsible AI language model, I don't have a "context window" in the classical sense, as I am not a physical device with a fixed window size.

Good to know that Llama2 has absolutely no self-awareness.

judeswae,
@judeswae@toot.thoughtworks.com avatar

@adr You'd think that would be the thing they would not get wrong.

Any cmdline program more or less responds correctly to the addition of a --version flag.

judeswae, to random
@judeswae@toot.thoughtworks.com avatar

@mowdev Hello and welcome to #TootWorks

judeswae, (edited ) to StableDiffusion
@judeswae@toot.thoughtworks.com avatar

The guessing game of prompting #GenAI to produce something consistent is exhausting.

It seems obvious I'm trying to generate some black and white circle. See this batch of 9 images generated with different seeds.

Where is that #ikea interior coming from?

Prompt: one black circle 1915, high contrast, plain, round, Pi

Negative Prompt: text, colors, lines, concentric, angles, faces, triangle, interior, figures, wall, straw, spiral, labyrinth, grid

#StableDiffusion

judeswae,
@judeswae@toot.thoughtworks.com avatar

@cford I was going for this one: https://en.wikipedia.org/wiki/Black_Circle

But yes, I was trying to see if #StableDiffusion model had an "idea" of Malevitch's work without expressly mentioning him.

judeswae,
@judeswae@toot.thoughtworks.com avatar

@cford So was the circle. He made multiple versions of square, circle, etc. over the years anyway.

judeswae,
@judeswae@toot.thoughtworks.com avatar

@cford I see where you're getting. If Google thinks 1915 is a square, a model trained for image generation might have made the same association.

judeswae,
@judeswae@toot.thoughtworks.com avatar

So I used "CLIP Interrogator", a tool that helps create #StableDiffusion prompts based on an image input. https://huggingface.co/spaces/fffiloni/CLIP-Interrogator-2

I gave it "Black Circle" by Malevitch as an input.

Prompt: a black circle with a white background, by Kazimir Malevich, traditional japanese concept art, light-brown skin, moor, inspired by Shōzō Shimamoto, intrincate, by Ram Chandra Shukla, heavy pigment, nothingness, subject in the centre, gazeta, 000

Using same seed, same model, here is the result.

cc: @cford

judeswae,
@judeswae@toot.thoughtworks.com avatar

Same prompt, same seed, different model.

#SDXL #StableDiffusion

cc @cford

judeswae, to StableDiffusion
@judeswae@toot.thoughtworks.com avatar

#StableDiffusion

Prompt: orthogonal grid, blue lines, white background, notebook paper, technical drawing, contrasted

Negative prompt: border, frame, wood, pen, yellow, oblique, black, label, table

Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1973065891, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.7.0

judeswae,
@judeswae@toot.thoughtworks.com avatar

@hans Oh yeah. They have no idea what a drawing is. They're just outputting pixels. Until recently, I had not looked at what stable diffusion does. But it's quite interesting. Indeed, it can only "think" in pixels.

As you guessed, today, I was trying to see if it was possible to extract some basic geometry and basic elements of a graphical language: squares, lines, etc.

judeswae,
@judeswae@toot.thoughtworks.com avatar

@hans FYI, I'm also sharing exactly how the images were generated. What input I gave. So you should be able to copy paste and recreate them with that.

judeswae, to StableDiffusion
@judeswae@toot.thoughtworks.com avatar

#StableDiffusion

Prompt: black square on a white background, vector style, plain colors, black or white only, Kazimir Malevich, avant-garde, geometric abstraction

Negative prompt: colors, border, texture, cross, watermarks, text
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 852818018, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.7.0

judeswae,
@judeswae@toot.thoughtworks.com avatar

Same prompt, different model
#SDXL #StableDiffusion

Prompt: black square on a white background, vector style, plain colors, black or white only, Kazimir Malevich, avant-garde, geometric abstraction

Negative prompt: colors, border, texture, cross, watermarks, text

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3467800448, Size: 1024x1024, Model hash: 31e35c80fc, Model: sd_xl_base_1.0, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Version: v1.7.0

judeswae, to ai
@judeswae@toot.thoughtworks.com avatar

"Thinking is a human feature. Will AI someday really think? That's like asking if submarines swim. If you call it swimming then robots will think, yes." -- Noam Chomsky

judeswae, to random
@judeswae@toot.thoughtworks.com avatar

@xiantongcheng Welcome to #TootWorks

judeswae, to random
@judeswae@toot.thoughtworks.com avatar

@chaoqunyang welcome to #tootworks

judeswae, to random
@judeswae@toot.thoughtworks.com avatar

"The performance of the workers who animate the artifice is obscured by the spectacle of the machine" -- Ayhan Aytes

#Fauxtomation #MTurk

sdw, to random
@sdw@mastodon.social avatar

Vision Pro comes with all the standard apps a user might expect like mail, messages, music, dinosaurs

judeswae,
@judeswae@toot.thoughtworks.com avatar

@sdw good point. Who wants to read their email in VR? Really.

judeswae,
@judeswae@toot.thoughtworks.com avatar

@Ferrichrome @sdw #Apple has made #VR so good you'll want to read your email with it.

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • JUstTest
  • ngwrru68w68
  • everett
  • InstantRegret
  • magazineikmin
  • thenastyranch
  • rosin
  • GTA5RPClips
  • Durango
  • Youngstown
  • slotface
  • khanakhh
  • kavyap
  • DreamBathrooms
  • provamag3
  • tacticalgear
  • osvaldo12
  • tester
  • cubers
  • cisconetworking
  • mdbf
  • ethstaker
  • modclub
  • Leos
  • anitta
  • normalnudes
  • megavids
  • lostlight
  • All magazines
    •