Rivers Have Wings @RiversHaveWings Twitter profile | Pikagi

Pikagi

Rivers Have Wings

@RiversHaveWings

31,152

Followers

225

Following

2,168

Media

5,411

Statuses

AI/generative artist. Writes her own code. Absolute power is a door into dreaming.

Joined June 2016

Don't wanna be here? Send us removal request.

Pinned Tweet

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

1 year

Humanity is going to make all parts of the world touched by humans beautiful. We are going to create beauty too cheap to meter. And not just an enforced-from-above standard of beauty either, everyone will be able to make their own domain beautiful in the manner of their choosing.

24

47

316

Last Seen Profiles

@smokley

@JetsonMade666

@rjprobinson

@nottpadu

@anikidamk

@longingforcolor

@_kairiichan

@OfficialMojflix

@MajinWuu47

@Karivayu

@AwaBDrame

@paige_bylund

@ldstudiomtl

@TeamAspectEU

@thetraehannibal

@mularroya

@PalaboyTrader

@Arabicaofficial

@terjekarlsen

@OJHLRaiders

@dgn_893x

@utopianray

@AnoukLHermitte

@CheetahsRugby

@FOUCHERFlorenc3

@_elmeta__

@JenkinVan

@kellyintulsa

@Arts_Alive

@LoungeDrs

@arts_tig

@_Su_ceren

@the_bones_zone

@Artemis_Chilena

@dudapavao

@mnbspark

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

This is just wild. #StyleGAN3

77

761

5K

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

I'm doing AI NFTs now. :) Our main product is called Control The Soul. It's a 1k piece collection meant to make cool Twitter avatars. The theme is wizards, eldritch artifacts, and weird landscapes. Most items 3-4 XTZ. Check out the gallery at

Tweet media one

Tweet media two

Tweet media three

Tweet media four

62

101

956

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"🌹 fractal" StyleGAN XL + CLIP

14

75

975

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"A vision of paradise, Unreal Engine" StyleGAN XL + CLIP

22

80

882

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"the gateway between dreams, trending on ArtStation" (CLIP guided YFCC diffusion, 968M parameter model, it's not done training yet)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

18

102

797

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"The Rise Of Consciousness" StyleGAN XL + CLIP

12

77

759

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

Stable Diffusion weights are out for academic research purposes upon request: ! Good job, everyone! :)

Tweet card media

GitHub - CompVis/stable-diffusion: A latent text-to-image diffusion model

A latent text-to-image diffusion model. Contribute to CompVis/stable-diffusion development by creating an account on GitHub.

16

133

699

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

d i s s o l v e (config-r edition, not a full dissolve) #StyleGAN3

15

88

642

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

1 year

If Stable Diffusion 2.0 doesn't know your favorite artist, you can search for similar artists on and try prompting with them instead. For instance here are several Rutkowski-like artists: Raphael Lacoste, Andreas Rocha, Titus Lunter, Richard Anderson.

Tweet media one

Tweet media two

Tweet media three

Tweet media four

58

89

566

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

1 year

I've trained a latent diffusion upscaler for the Stable Diffusion autoencoder (and anything you feel like feeding into it if you can tolerate a little artifacts) in collaboration with @stabilityai . Try the Colab written by @nshepperd1 here:

Tweet media one

Tweet media two

17

108

546

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

10 months

I have a new Colab notebook, “Reinforcement Learning from AI Feedback” (), which fine-tunes a language model (Pythia 160M) using natural language criteria for its sampled outputs. Standalone version for larger models below. Have fun, report bugs! :)

Tweet card media

Reinforcement Learning from AI Feedback.ipynb

Colaboratory notebook

colab.research.google.com

4

97

491

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 months

Hourglass + Diffusion = ❤️ We introduce a new transformer backbone for diffusion models that can directly generate megapixel images without the need for multiple stages like latent diffusion. Read here! → Project page →

Tweet media one

19

89

458

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

a propaganda poster for building a space elevator

Tweet media one

10

47

443

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

I scaled up CLIPDraw () a bit... "a beautiful epic wondrous fantasy painting of [the ocean / lightning / wind / a deep valley]":

Tweet media one

Tweet media two

Tweet media three

Tweet media four

10

88

433

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

OMG. CLIP knows how to Deep Dream-ify images by maximizing similarity with the text prompt "Deep Dream".

Tweet media one

Tweet media two

10

65

422

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"the vision of the ophanim by Ross Tran"

Tweet media one

Tweet media two

5

54

403

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "a painting of climate change killing humanity, by Greg Rutkowski" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

14

49

374

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

the oracle of glitch

Tweet media one

Tweet media two

Tweet media three

Tweet media four

6

57

364

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

4 months

I made a flexible new image captioning method based on only a base model LLM and CLIP. It lets you go beyond just describing what is in an image and analyze narrative themes, latent knowledge in CLIP embeddings, and more! Colab:

Tweet card media

Image To Text (LLM + CLIP).ipynb

Colaboratory notebook

colab.research.google.com

8

79

373

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

My 602M parameter CLIP conditioned diffusion model trained on Conceptual 12M is out at ! It can generate images matching the prompt quickly using its CLIP conditioning, but still requires CLIP guidance for best results.

Tweet media one

15

71

370

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

The Egyptian pyramids in a snowstorm

Tweet media one

Tweet media two

Tweet media three

Tweet media four

5

44

359

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "Van Gogh's Starry Night by Ivan Aivazovsky" (CLIP guided diffusion, with Starry Night as an init)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

5

37

357

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "a surreal and organic stone temple of Cthulhu" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

4

29

346

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

From #EleutherAI Discord:

Tweet media one

10

38

349

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

A nuclear explosion

Tweet media one

11

32

339

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

Our VQGAN+CLIP paper (authors: me, @BlancheMinerva , @jbusted1 , @dashstander , @EricHallahan , @lcastricato , @EdwardRaffML ) is finally out!

@_akhaliq

AK

2 years

VQGAN-CLIP: Open Domain Image Generation and Editing with Natural Language Guidance abs: github:

Tweet media one

1

77

315

8

61

345

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

New Colab notebook: CLIP Guided Diffusion HQ 512x512

Tweet card media

CLIP Guided Diffusion HQ 512x512.ipynb

Colaboratory notebook

colab.research.google.com

11

37

330

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

water is a circle

Tweet media one

Tweet media two

7

27

321

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

1 year

Here is a notebook of the method I have been using to make these animation posts: (Stable Diffusion KLMC2 Animation). Have fun, report bugs! :)

Stable Diffusion KLMC2 Animation.ipynb

Colaboratory notebook

colab.research.google.com

24

52

323

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"a nuclear explosion by Dan Mumford"

Tweet media one

Tweet media two

6

23

317

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"the gateway of the sun, trending on ArtStation" (CLIP guided YFCC diffusion, 968M parameter model, it's not done training yet)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

7

35

318

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

New Colab notebook: CLIP Guided Diffusion HQ 512x512 Uncond This notebook contains my 8100 step fine-tune of the 512x512 class-conditional ImageNet model into an unconditional ImageNet model for better controllability with CLIP.

Tweet card media

CLIP Guided Diffusion HQ 512x512 Uncond.ipynb

Colaboratory notebook

colab.research.google.com

15

46

318

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "The mountains of Oregon with wildfire smoke by Ivan Aivazovsky" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

8

32

305

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

I updated my secondary model diffusion method notebooks with a new, bigger but still efficient secondary model: (256x256), (512x512). The original model is still in there and you can select it instead!

Tweet card media

CLIP Guided Diffusion 512x512, Secondary Model Method.ipynb

Colaboratory notebook

colab.research.google.com

12

45

308

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

elegant hell butterfly

Tweet media one

Tweet media two

2

45

301

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "a painting of artificial intelligence, the sky is burning, by Greg Rutkowski" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

5

41

292

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

Prompt: "A cyberpunk cathedral by Greg Rutkowski and Claude Monet, oil on canvas" ( #StableDiffusion )

Tweet media one

Tweet media two

Tweet media three

Tweet media four

8

41

296

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "a painting of humanity surviving artificial intelligence, by Greg Rutkowski" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

10

35

292

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

I added the PRK and PLMS sampling methods (from ) to and the cc12m_1_cfg Colab at . The default for the Colab is now 50 steps PLMS which seems to give pretty good quality. Have fun! :)

Tweet card media

Text to Image (CC12M Diffusion).ipynb

Colaboratory notebook

colab.research.google.com

21

47

298

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

CLIP Guided Diffusion 512x512, Secondary Model Method: This is the faster/lower memory method I promised yesterday. Quality may be lower with this notebook but we can probably train a better secondary diffusion model now that we know they're useful.

Tweet card media

CLIP Guided Diffusion 512x512, Secondary Model Method.ipynb

Colaboratory notebook

colab.research.google.com

14

39

302

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "a cyberpunk church resembling the Death tarot card by Greg Rutkowski" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

2

35

290

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

I made a text to image Colab notebook for my latest model, cc12m_1_cfg! It's more basic right now than the CLI sampling scripts (it only does one text prompt and doesn't have init images yet) but I plan to add more features later.

Tweet card media

Text to Image (CC12M Diffusion).ipynb

Colaboratory notebook

colab.research.google.com

21

39

287

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

oh hey i discovered a way to speed up CLIP guided diffusion by nearly 2x and lower memory requirements by a similar amount... more details later, i need to go to bed

13

8

286

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

It's one year to the day since I got my first batch of good CLIP Guided Diffusion outputs! I wrote the public notebook, "CLIP Guided Diffusion HQ 256x256", the next day. The prompt was "Desert dynamism by Giacomo Balla". (Notebook link: )

Tweet media one

Tweet media two

9

37

284

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

"industrial demon, a matte painting, trending on ArtStation" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

10

23

274

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "Artificial Intelligence is the vehicle for demons to enter our world, by John Harris" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

5

27

262

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

Prompt: "A utopian arcology by H. R. Giger, oil on canvas"

Tweet media one

Tweet media two

Tweet media three

Tweet media four

8

29

266

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

Prompt: "The organic machine by Rene Magritte, oil on canvas" Wow this model...

Tweet media one

Tweet media two

Tweet media three

Tweet media four

14

25

258

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

I finally managed to find hosting for my CLIP conditioned transformer (it's on GitHub as a release right now), so here's the Colab:

Tweet card media

CLIP Decision Transformer.ipynb

Colaboratory notebook

colab.research.google.com

11

34

262

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

CLIP-based style transfer is a thing! (Using the ViT-B/32 image encoder outputs only, no use of intermediate layers.)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

4

36

259

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "a painting of UFOs, the sky is burning, by Greg Rutkowski"

Tweet media one

Tweet media two

Tweet media three

Tweet media four

9

37

254

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

I just realized I forgot to ever announce my diffusion model training and hackable inference codebase, k-diffusion, on Twitter! It only has unconditional diffusion models rn but text conditioned is easy to add.

Tweet card media

GitHub - crowsonkb/k-diffusion: Karras et al. (2022) diffusion models for PyTorch

Karras et al. (2022) diffusion models for PyTorch. Contribute to crowsonkb/k-diffusion development by creating an account on GitHub.

1

25

251

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

Our first batch of four diffusion models, along with super basic CLIP guided inference code, is out! The models are: Danbooru SFW 128x128 ImageNet 128x128 WikiArt 128x128 WikiArt 256x256 No Colab yet but soon!

Tweet media one

Tweet media two

Tweet media three

Tweet media four

10

30

246

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"Resurrection of the dead via multiverse-wide acausal cooperation by Ross Tran"

Tweet media one

Tweet media two

1

32

240

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"a gateway between dreams, pixel art #pixelart ", @midjourney diffusion

Tweet media one

Tweet media two

Tweet media three

Tweet media four

2

46

248

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

Prompt: "Humanity is killed by artificial intelligence, by James Gurney, oil on canvas"

Tweet media one

Tweet media two

Tweet media three

Tweet media four

11

25

245

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

thinking about how CLIP guided diffusion was a pretty good method for its time and its time was like the second half of 2021

18

4

243

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

a cyberpunk city

Tweet media one

5

30

238

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"a Minecraft screenshot, unreal engine" Deep Image Prior + CLIP ViT-L/14 336px

Tweet media one

5

21

242

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "revolution of the souls" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

2

42

240

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

Experimenting with CLIP as a feature extractor for NIMA (neural image assessment) type things... a simple linear regression yielded a direction in CLIP latent space that can increase the "aestheticness" of an output. Here's "a datura witch" and a couple of Starry Night variants:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

11

26

239

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"The Rise Of Consciousness" StyleGAN XL + CLIP (from a different starting seed)

5

13

241

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

First try guiding #StyleGAN3 with CLIP...

5

22

233

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

By a user in EleutherAI Discord, using the VQGAN+CLIP bot there: "recursive recursion of the recursive imagination of a landscape by james gurney" and "landscape of recursion by james gurney". CLIP understands that "recursion" means you put the thing inside itself etc?

Tweet media one

Tweet media two

4

34

236

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

"humanity is killed by climate change, by Marc Simonetti" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

7

30

233

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "The coast of California with wildfire smoke by Ivan Aivazovsky" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

9

22

230

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

Prompt: "a painting of the beginning of the world, by Greg Rutkowski" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

4

25

234

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

CLIP guided diffusion Colab, super slow HQ version (hope you get a V100):

Tweet card media

CLIP Guided Diffusion HQ 256x256.ipynb

Colaboratory notebook

colab.research.google.com

15

24

227

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

New Colab: CLIP guided diffusion

Tweet card media

CLIP Guided Diffusion.ipynb

Colaboratory notebook

colab.research.google.com

8

32

227

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"The Rise Of Consciousness by Max Ernst, oil on canvas" VQGAN+CLIP+prior

Tweet media one

Tweet media two

Tweet media three

Tweet media four

8

26

225

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "the last day of the air, trending on ArtStation" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

8

25

223

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

"technological singularity cult by James Gurney" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

4

28

221

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

Prompt: "Futurism by Jeffrey Smith, oil on canvas" (CLIP guided diffusion) Interestingly this did not give me Italian Futurism, you have to ask for it by artist name I think?

Tweet media one

Tweet media two

Tweet media three

Tweet media four

9

21

222

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

A Christmas tree in outer space

Tweet media one

Tweet media two

Tweet media three

Tweet media four

2

27

216

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

the conscious void

Tweet media one

5

21

219

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"the winds of change by Ross Tran"

Tweet media one

Tweet media two

Tweet media three

Tweet media four

6

28

222

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

CompVis's new latent diffusion model clearly thinks "a gateway between dreams" should be a book title. Look at that generated text though!

Tweet media one

Tweet media two

17

20

220

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "a giant floating eye monster, by James Gurney" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

3

23

215

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "a sea monster by Ivan Aivazovsky" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

3

24

211

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

I released two ViT-B/16 CLOOB checkpoints, along with JAX training code and JAX/PyTorch inference code, trained on the open dataset LAION 400M: . Zero-shot ImageNet accuracy:

Tweet media one

6

28

213

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

"the lighthouse at the end of the universe by James Gurney" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

1

22

209

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Some diffusion Control The Soul...

Tweet media one

Tweet media two

Tweet media three

Tweet media four

13

33

209

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"twilight is a circle by Ross Tran"

Tweet media one

Tweet media two

Tweet media three

1

34

206

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

Prompt: "The alchemist's laboratory by Greg Rutkowski and Claude Monet, oil on canvas" ( #StableDiffusion )

Tweet media one

Tweet media two

Tweet media three

Tweet media four

4

27

211

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

You can apply a similar trick to classifier-free guidance to autoregressive transformers to sample from a synthetic "super-conditioned" distribution. I trained a CIFAR-10 class-conditional ImageGPT to try this, and I got the following grids with cond_scale 1 (default) and then 3:

Tweet media one

Tweet media two

4

24

213

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

Prompt: "A technological singularity cult by James Gurney and Rob Gonsalves, oil on canvas" ( #StableDiffusion )

Tweet media one

Tweet media two

Tweet media three

Tweet media four

4

13

206

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"a peaceful field of tall sunflowers" cc12m_1_cfg

Tweet media one

4

17

200

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "a cold landscape by Thomas Kinkade" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

7

19

205

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

Me, a year ago:

Tweet media one

9

19

203

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

a propaganda poster for transhumanism

Tweet media one

4

16

198

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

~

Tweet media one

Tweet media two

Tweet media three

Tweet media four

5

7

200

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

some prog rock album covers

Tweet media one

5

17

198

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

cc12m_1 with classifier-free guidance (the model I've been using to make all of these grids for the last few days) is out: ! The repo has a new cfg_sample.py script to sample from the new model.

Tweet card media

GitHub - crowsonkb/v-diffusion-pytorch: v objective diffusion inference code for PyTorch.

v objective diffusion inference code for PyTorch. Contribute to crowsonkb/v-diffusion-pytorch development by creating an account on GitHub.

11

38

205

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "unplanned emergent AI behavior causes catastrophe, by James Gurney" (CLIP guided diffusion) I like how you can't really tell what is going on in some of these, just that it is weird/alien and *shouldn't be happening*.

Tweet media one

Tweet media two

Tweet media three

Tweet media four

4

27

190

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

I updated my 256x256 and 512x512 unconditional CLIP guided diffusion notebooks with support for multiple weighted text and image prompts: (256x256) and (512x512)

Tweet card media

CLIP Guided Diffusion HQ 512x512 Uncond.ipynb

Colaboratory notebook

colab.research.google.com

3

23

201

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

"a nuclear explosion by Ross Tran"

Tweet media one

Tweet media two

1

15

191

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

3 years

Prompt: "a beautiful epic wondrous fantasy painting of wind" (CLIP guided diffusion)

Tweet media one

Tweet media two

Tweet media three

Tweet media four

5

28

199

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

2 years

Prompt: "The queen of the sun, by Jeffrey Smith, oil on canvas" (Stable Diffusion)

Tweet media one

11

24

197

@RiversHaveWings

Rivers Have Wings

@RiversHaveWings

1 year

animated cubism #StableDiffusion2 #AIart hq:

11

30

193