September 7, 2021 -- Spatial

The Tuesday Night drawing group of Santa Fe drew together online using Zoom, because of the Covid 19 quarantine. The model was from the Washington DC area.

Krrrl Drawings Playlist

Krrrl Drawings with AI

September 7, 2021

I used the following programs to make the video above:

• VQGAN+CLIP
• Abraham.ai
• Artbreeder
• Deep Dream Generator
• Pelalica Paint
• Paint 3D
• Windows Video Editor

Plus I used other online art programs to generate images in this blog post:

• AI Art Machine
• CLIP Guided Diffusion HQ
• Paint by Numbers Generator  (GitHub)
• PBNify
• Color-Dragon (paid)
• YANDEX -- I prefer the reverse image results from this Russian search engine over Google's results

SPATIAL

I referenced famous sculptors with AI in order to ultimately create spaces that I could imagine walking into.

Base pencil drawing "initial_image" input

to generate the shallow spaces pictured below

I altered the above drawing several time in the "text-to-image" program VQGAN+CLIP, changing the input prompts:

HENRY MOORE

YANDEX and Google similar visual images

prompts:  "Zbrush | In the style of Henry Moore"

GIACOMETTI

YANDEX and Google visually similar images

prompts:  "A Giacometti figure walking into infinite space"

Different programs, very different results

I used the SAME TEXT -- "A Giacometti figure walking into infinite space" -- in Abraham.ai, a kind of crowdsourcing "text-to-image" generator, but the result came out very differently:

GIACOMETTI

YANDEX and Google visually similar images

Same text in Abraham.ai --

"A Giacometti figure walking into infinite space"

I continued to use the SAME TEXT in a different  "text-to-image" VQGAN program -- AI Art Machine:

GIACOMETTI

YANDEX and Google visually similar images

Same prompt "p":  "A Giacometti figure walking into infinite space"

Note:  Neither of the above two different VQGAN programs allowed me to input an initial image -- both of them were completely "text-to-image."

I then combined TWO of my drawings and the final result was better (note that this source image is a square, the dimension that the VQGAN program uses):

Base "initial_image" input

to generate the spaces pictured below

GIACOMETTI

YANDEX and Google visually similar images

Nvidia Inpainting Demo

Englarge (1792 x 1792, 4.33 MB):  Giacometti_less.png

prompts:  "People occupying a Giacometti space"

HENRY MOORE

YANDEX and Google visually similar spaces

prompts:  "People occupying a Henry Moore space"

PIRANESI

YANDEX and Google visually similar images

prompts:  "People occupying a Piranesi space"

STYLE TRANSFER

Deep Dream Generator

I transferred the style of an image I created earlier this year in Paint 3D to some of my figure drawings using DEEP DREAM GENERATOR:

January 26, 2021

The style of this image (made in Paint 3D)

transferred well to the

six different figures below

Deep Dream Generator

YANDEX and Google visually similar images

Deep Dream Generator

I took some advice from Steve's Makerspace YouTube, and made sure that both the style image and the target image had black backgrounds.

I inverted the drawing 

to make the background dark 

before importing it into Deep Dream Generator

YANDEX and Google visually similar images

Deep Dream Generator

Petalica Paint

YANDEX and Google visually similar images

Deep Dream Generator

Deep Dream Generator

Petalica Paint

YANDEX and Google visually similar images

Deep Dream Generator

Petalica Paint

Paint 3D

YANDEX and Google visually similar images

Deep Dream Generator

Petalica Paint

Paint 3D

YANDEX and Google visually similar images

Dancing

Petalica Paint

Paint 3D

Dancing

Petalica Paint

PORTRAIT ALCHEMY

I used AI (artificial intelligence) again to alter my portrait drawing:

Scanned raw pencil drawing

and base "initial_image" input

for the images below

Altered in VQGAN+CLIP:

YANDEX and Google visually similar images

prompts:  "Renaissance portrait"

Then I imported the above portrait into Artbreeder:

Artbreeder

Variations:

Artbreeder

Artbreeder interface

for making variations

Artbreeder

combining this portrait

with the portrait from the last drawing session

Artbreeder

AnimeGANv2

Update (Nov 18) -- Cartoonified

AnimeGANv2

Artbreeder summary

Paint 3D

PNG

I used VQGAN+CLIP to generate another variation on my portrait:

BOTERO

YANDEX and Google visually similar images

prompts:  "Portrait in the style of Botero"

Paint by Numbers

If I simplify this result above, I might be able to make a serigraphy edition of the AI artwork.  I used three Paint by Numbers programs to reduce the AI portrait into 4 colors:

• GitHub:  Paint by Numbers Generator
• PNGify
• Color-Dragon (paid)

YANDEX and Google visually similar images

Github:  Paint by Number Generator

PBNify

Color-Dragon

COMBO

YANDEX and Google visually similar images

Paint by Numbers Generator

16 colors

Paint by Numbers Generator

4 colors

 RIOT

2000 x 2000 pixels

733.77 KB

reduced by RIOT

CLIP Guided Diffusion HQ

There seems to be many  "text-to-image" variations on the "VQGAN approach.  The CLIP Guided Diffusion HQ is posted on Hugging Face.

I uploaded both the "initial image" and "image prompt" -- the top two images in the collage below -- to the program on Hugging Face, and typed in the text "Walking through the street aimlessly," to get the resulting ambiguous image on the bottom:

text:  "Walking through the street aimlessly"

Video --

text:  "Walking through the street aimlessly"

The Hugging Face interface

RELIEF PRINT

I drew the last post on a 7 x 7 inch square of expanded PVC, and cut and printed the plate immediately after the drawing session, on the Tortilla Press.

Deep Dream Generator

AI variation

PHOTO BOX

Ethan Moses brought his homemade camera/dark room to Quelab on September 8th, to take and develop photographs.  The ingenious system worked quite well, and Ethan took flattering photos of everyone.

Ethan Moses taking my photo

I'm developing the  photo

inside the box