Tuesday, 28 February 2023

February 2023 Review -- AI Revolution

I want AI (Artificial Intelligence) to generate images in my style, both for me and anyone else. And this might be art too -- according to Essentialism.

Leonardo.AI  (YouTube) generates images in MY DRAWING STYLE after it trained on 30 drawings from my book :

KRRRL AI Art -- February 2023

The images from the video above are posted on these two web pages:


Only Midjourney allows me to upload two images and "Blend" them, which is a great feature.  Yet Midjouney's NSFW filter is way too sensitive, and will NOT usually allow me to upload and blend my Figure Drawings. So there is no good way to generate and blend images in Midjourney with my drawing style.


Midjourney does a great job
blending an artwork with a photograph


OpenArt however would accept my figure drawings. Therefore last December 2022 I trained a model on my drawings, using their Photo Booth feature.  While OpenArt then generated images in my drawing style, it was limited to just figures of women (like the source images it trained on). For instance note the single figure below -- generated with the prompt: "woman riding a rhinoceros" --  but sans rhinoceros:

prompt: "woman riding a rhinoceros" --
but no rhinoceros --
Generated in OpenArt AI
using their Photo Booth feature


Albeit limited, the OpenArt results worked to make a laser engraved print:

Laser engraved small linocut
for the Open Print Exchange #2


However OpenArt allowed me to download the "ckpt" AI model I created with their Photo Booth feature:
I was inspired to share that model, which one can do by uploading to CIVITAI (though I did not).

Back in 2020 I did share an AI model that I made in made in Runway ML based on my drawing style.  This month I discovered that my model was used as part of a large AI projection project -- AUTOLUME ACEDIA -- in Kelowna, British Columbia, Canada, last February 2022:
I had uploaded that previous 2020 model to Justin Pinkney's Github:  


Ultimately I chose share my drawing style inside Leonardo.AI.  That website allows me to make AI models based on my figure drawings, and share my model with the rest of their users.  My "Krrrl Lines 06" model was trained on PG images -- which I made it "public," so that everybody can generate images in my KRRRL drawing style.  I can't download that AI model and post it on CIVITAI however.


LEONARDO.AI Images

Naturally I went crazy this month generating images in my drawing style while the service is still free (after getting access on February 18th):

John and Yoko
altered with AI model: "Krrrl Drawings 01"


AI KRRRL Paintings
Based on the AI model
of my painting style --
model: "Krrrl Paintings"


Also "outpainted" in 
Leonardo "AI Canvas"




Other people on Leonardo.AI also generated images from the AI model based on my KRRRL drawing style:

Generated by user "Adenda"
using my AI style model
in Leonardo.AI


I even generated a linocut image with Leonardo.AI, and laser engraved it into linoleum at Quelab, and printed it at Henry Morales' studio:

Leonardo.AI Linocut
Printed at Henry Morales' studio
in Maya Blue


OTHER AI Programs

I also paid for a Pro subscription to Mage.Space.  I can upload my figure drawings and Mage.Space will alter them. While this online program will generate AI images for free, one needs a subscription to use all 36 models (plus with a subscription there is no problem with NSFW).   I get the most out of their "Pix2Pix" and "ControlNet" -- which they added on February 26th -- especially "ControlNet Canny,"


February 27, 2023


February 27, 2023
Created on Mage.Space with "ControlNet Canny" 
after I uploaded one of my drawings
prompt: "3D Sculpture of a sitting woman made of blue ice"


Blue Willow is free and works inside Discord, like Midjourney.  While it's quicker than Midjourney, the results are not always as pretty:

Blue Willow
Blue Willow image with
prompt: "3D Sculpture of a sitting woman made of blue ice"


DEZGO is free, and no sign up is required.  It allows one to upload and alter an image (and does not have an NSFW filter):

The Dezgo generation on the right
smoothed out the input Midjourney generation on the left



HOLOGRAMS

After making a few YouTube slideshows of my AI Art creations I realized that I could also convert those same slideshows into 3D HOLOGRAMS.

RGBD PHOTO will convert a 2D image into a diptych-with-a-depth-map -- which turns into a hologram when uploaded to the Looking Glass Portrait display.  I converted and uploaded the images into HOP files, and then posted the HOLOGRAM slideshows on my GDrive, for anyone to download:



Leonardo.AI -- KRRRL Drawings

KRRRL Paintings by AI

Hologram files can be downloaded from the following GDrive folders:



My Leonardo.AI Models

In February I trained six models in Leonardo.AI on my drawing/painting style:

Leonardo list of all my AI models
trained in February 2023


1)  Krrrl Drawings 01 -- I trained the first model on 30 Drawings from my book "Finish My Figure Drawings": 
Figures: 001, 002, 005, 006, 007, 013, 014, 015, 018, 019; 021, 025, 026, 030, 033, 034, 039; 043, 044, 045, 046, 049, 050, 052, 053, 054, 056, 057, 058, 059

Krrrl Drawings 01
instance prompt: "Krrrl01"

Generated with 
"Krrrrl Drawings 01"


2)  Krrrl Drawings 02 -- I trained the second model on 30 of the best drawings from 2022, taken from the following blog posts:
Krrrl Drawings 02
instance prompt: "Krrrl02"

Generated with 
"Krrrrl Drawings 02"


3)  Krrrl Paintings -- I trained the third model from 30 photos of my paintings, taken from the following two blog posts:
Krrrl Paintings
instance prompt: "Krrrlpaint"

Generated with 
"Krrrrl Paintings"


4)  Krrrl Drawings 04 -- I trained the fourth model on 30 Drawings from my book "Finish My Figure Drawings":  
Figures: 061, 062, 065, 068, 072, 073, 074, 080; 082, 084, 085, 088, 089, 90, 94, 95, 98, 100; 101, 103, 105, 109, 111, 113, 114, 115, 119

Krrrl Drawings 04
instance prompt: "Krrrl04"

Generated with 
"Krrrrl Drawings 04"


5) Krrrl Pencil and Ink 05 -- I trained the fifth model on 30 of the best of my pencil and ink drawings of January and February 2023:

Krrrl Pencil and Ink 05
instance prompt: "Krrrl5"

Generated with 
"Krrrl Pencil and Ink 05"


6)  Krrrl Lines 06 -- I trained the sixth model on 30 PG generations made from the first and fourth models, based on the drawings of my book.  Since all the data images were safe-for-work, I made this model public, so that other users on Leonardo.AI can generate images based on my drawings:

Krrrl Lines 06
instance prompt: "KrrrlLines"
*The PG model
that I offered to the public*

Generated with 
"KrrrlLines 06"

  • The first model -- Krrrl Drawings 01 -- generates the best images.
UPDATE (March 9th): My public model -- "KrrrlLines 06" -- showed up during a scroll down on a new Leonardo tutorial YouTube video -- LEONARDO.AI Tutorial - How to use this fantastic Ai Art Platform!  So I know that other people are seeing it, and perhaps even using it.

YouTube: LEONARDO.AI Tutorial - 
How to use this fantastic Ai Art Platform!
My model "KrrrlLines 006"
 -- the bear in the upper left --
shows up as the YouTubers scrolls down



VECTORIZING

Since I was drawing in pencil and ink I thought those drawings might translate well into vector files:

Individual SVG vector files
can be downloaded from the original blog post

February 27, 2023
Download all the SVG vector files from this post:


NOTABLE

The AI Art controversy flared up in the Netherlands, when the original "Girl with a Pearl Earring" was replaced with an AI version.  The original was loaned out for a big Vermeer exhibition at the Rijksmuseum in Amsterdam, so the Mauritshuis Museum had an Open Call to artists for variations of the painting -- five of the 3,482 entries were chosen to hang in the museum.  While the AI version apparently was the grand winner, a photo by Vallarie Enriquez of El Paso of her husband Arturo, was one of the five chosen versions:  

Vallarie Enriquez's photo 
is now hanging in the Mauritshuis Museum in The Hague


Meanwhile the Iron Man sculpture met his match on February 27th, when a truck ran into him on Lomas at 4th. The Iron Man faced off with three justice buildings, on other three corners of this intersection in Albuquerque:

February 27, 2023
The Iron Man went down
at Lomas and 4th in Albuquerque


I always find it ironic that at least one of the justice buildings was a crime, as the mayor and head of the state senate skimmed off a lot of graft from its construction. Maybe that is what the Iron Man was standing up against:

Wikipedia: "From 2005 to 2009, the Metro Courthouse was at the center of a high-profile fraud investigation, during which allegations emerged that a group of conspirators had siphoned off $4.2 million from the courthouse construction project in a scheme described by the Albuquerque Journal as "breathtaking in scope and star power." Eight people were eventually named as defendants in the case, including the former president pro tempore of the New Mexico State Senate and a former mayor of Albuquerque. The investigation ended with six of the accused pleading guilty to conspiracy and mail fraud, while the other two pleaded guilty to misprision of felony. "

 

PROGRAMS

I note that MattVidPro AI is from Albuquerque, according to his Twitter.



Posted by at

No comments:

Post a Comment

Note: only a member of this blog may post a comment.

Subscribe to: Post Comments (Atom)