August 24, 2021 -- VQGAN+CLIP -- Abraham AI

The Tuesday Night drawing group of Santa Fe drew together online using Zoom, because of the Covid 19 quarantine.

Krrrl Drawings Playlist

Krrrl Drawings from August 21, 2021

witih VQGAN+CLIP

SEATED LEANING FIGURE

Note that the first three images have been enlarged for printing, and are found on the "Enlarged Images" page.

YANDEX and Google visually similar images

Enlarged (5041 x 6142, 9.83 MB):  GIGAPIXEL__002.jpg

I altered my pencil drawings using AI (artificial intelligence) in the VQGAN+CLIP programs (using the then default database "imagenet_16384" I believe). The 14 images that follow are based on my seated leaning figure drawing above:

AI variations of my drawing

YANDEX and Google visually similar images

Enlarged (3176 x 3144, 11.38 MB):  GIGAPIXEL__003.png

Fake EGON SCHIELE

text:  "portrait of crypto degen | Egon Schiele"

YANDEX and Google visually similar images

Enlarged (1400 x 1400, 3.55 MB):  Hollyupscale.png

Enlarged (3128 x 3167, 11.95 MB):  GIGAPIXEL__001.png

Fake MOEBIUS

text:  "portrait of crypto degen | Moebius"

YANDEX and Google visually similar images

Fake EGON SCHIELE AND MOEBIUS

text: "Human figure in the style of Egon Schiele and Moebius"

YANDEX and Google visually similar images

Fake AUBREY BEARDSLEY

text:  "Thicken lines in the style of Beardsley"

YANDEX and Google visually similar images

Fake AUBREY BEARDSLEY

text:  "portrait of crypto degen | Thicken lines in the style of Beardsley"

YANDEX and Google visually similar images

Fake ART NOUVEAU

text:  "In the style of art nouveau"

YANDEX and Google visually similar images

Fake ART DECO

text:  "In the style of art deco"

YANDEX and Google visually similar images

Fake TAMARA de LEMPICKA

text:  "In the style of Tamara de Lempicka"

YANDEX and Google visually similar images

Fake DIEGO RIVERA

text:  "In the style of Diego Rivera"

YANDEX and Google visually similar images

Fake OTTO DIX

text:  "In the style of Otto Dix"

YANDEX and Google visually similar images

Fake ROBERT WILLIAMS

text:  "In the style of Robert Williams"

YANDEX and Google visually similar images

DEEP DREAM

text:  "Deep dream"

YANDEX and Google visually similar images

MOST UNUSUAL DRAWINGS STYLE

text:  "Most unusual drawing style"

YANDEX and Google visually similar images

IN THE STYLE OF PRINTMAKING

text:  "In the style of printmaking"

SITTING FIGURE

The sitting figure drawing below lead to three variations in AI (the first shown at two different levels of development):

I first converted this figure into a square image so that the AI program does not distort the proportions.  The program first converts the image into a square before it continues with the AI treatments.

First converted into a square image

text:  "In the style of Sir John Everett Millais"

YANDEX and Google visually similar images

Fake SIR JOHN EVERETT MILLAIS

text:  "In the style of Sir John Everett Millais"

(first generation)

YANDEX and Google visually similar images

Fake SIR JOHN EVERETT MILLAIS

text:  "In the style of Sir John Everett Millais"

(second generation)

YANDEX and Google visually similar images

Fake ROGER DEAN

text:  "In the style of Roger Dean"

YANDEX and Google visually similar images

ARCHITECTURE

text:  "In the style of architecture"

text:  "In the style of architecture"

RECLINING FIGURE

The reclining figure drawing below was also altered in the VQGAN+CLIP AI program, and in Infinite Patterns and Deep Dream Generator:

Photoshop Element 2021

This drawing was distorted first

in Photoshop Element 2021

Filter/Distort/Liquify...

YANDEX and Google visually similar images

BAD ART

text: "bad art"

INFINITE PATTERNS

I also altered the high resolution version of the reclining figure using a Google AI experiment -- Infinite Patterns Colab notebook -- by the guy who designed Deep Dream Generator, Alexander Mordvintsev:

This infinite pattern

seeded with my reclining figure drawing

Then I applied the style  of that pattern to the reclining figure drawing in Deep Dream Generator, and further altered it in XnView:

Deep Dream Generator

XnView

Still needs work,

after using two AI processes on my drawing

Pen Drawings

colored by Petalica Paint

ABRAHAM.AI

I created the following images in Abraham.ai using only text-to-image:

YANDEX and Google visually similar images

 text: "Cafe scene in the style of Tamara de Lempicka"

(second try)

YANDEX and Google visually similar images

 text: "Cafe scene in the style of Tamara de Lempicka"

(first try)

I entered the same text twice (because I thought the system didn't work the first time), and was surprised to get the two different images above.

The first image I created in Abraham.ai was horrible:

text:  "Drawing the human figure on Tuesday Night"

Abraham.ai is a strange collective approach to creating text-to-image art, created by Gene Kogan:

 

Abraham is an open project to make an autonomous artificial artist, a crowd-sourced AI that generates art.

I needed a Metamask Ethereum wallet to participate in Abraham.ai:

• 0xB06614BD2B97F14a47112D0673f7FCE35A998AB8

• Abraham Discord

• Gene Kogan also authored this Google Experiment -- What Neural Networks See

• The VQGAN+CLIP programs seem to download images from ImageNet.  That is probably why the AI program works best when referencing better known artists.  Perhaps downloading the WikiArt database would tap into more diverse artistic styles. 

• Screen Saver World

• Deep Dream Generator tutorial video