Thursday, 24 November 2022

November 23, 2022 -- Second Veterans Exhibition drawing session

Donald Johnson (Lt Commander Donny Deron Johnson Jr Esquire --USN Retired) arrange for  the second life drawing session in the Fine Art Building on the ground of Expo New Mexico (State Fairgrounds) in Albuquerque, as part of the annual New Mexico Veterans Art Exhibition.

On November 9th we enjoyed the first drawing session as part of the New Mexico Veterans Art Exhibition 2022.

I can work with these AI Art generations

However overall, the AI Prompt programs 
generated too much descriptive text,
which probably hurt my drawings


The DRAWINGS

















Ken Romig
drew me in the background



Artificial PROMPT Generators

THIS YouTube VIDEO encouraged me to work somewhat backwards -- from Image to Prompt back to Image. 

I uploaded the drawing below to generate prompts online in two different AI programs.  And then I uploaded those AI prompts to both MIDJOURNEY and "TEXT 2 DREAM" (Deep Dream Generator) to generate more figures:


This is the drawing that I uploaded
to generate AI prompts

*****

PROMPT

I uploaded my drawing to img2prompt
and generated the AI Prompt below:

Img2prompt: "a drawing of a woman sitting on the ground by Hans Bellmer, pixiv, figuration libre, pencil sketch, stipple, biomorphic"


Then I uploaded that AI Prompt above to generate more figures in Midjourney (see below):


MIDJOURNEY

"AI Prompt only -- and "Image and AI Prompt" variations:

Without drawing uploaded (top row)
 and with original drawing uploaded (bottom row)
to Midjourney --
plus the uploaded img2prompt AI prompt


AI Prompt only
uploaded to Midjourney --
The nicest result


AI Prompt and the original drawing uploaded
to Midjourney  --
Closer to my original drawing


"Text 2 Dream"
 (Deep Dream Generator)

With the same above img2prompt prompt, I generated two versions in "Text 2 Dream" (Deep Dream Generator) -- "Without," and "With the Original Drawing" uploaded:

AI Prompt only (first image)
and AI Prompt and original drawing uploaded (second image)
to "Text 2 Dream" (Deep Dream Generator)






STABLE DIFFUSION 1 Demo

I also entered the AI Prompt in the STABLE DIFFUSION 1 Demo on Hugging Face:



Olivio Sarikas on YouTube announces that Stable Diffusion 2.0 was just released (Nov 23rd), but the Hugging Face Stable Diffusion 2 demo was so busy that I could not successfully upload my AI Prompt.

STABLE DIFFUSION 2 DEMO

Stable Diffusion 2 was so new that THE DEMO was too popular.  I finally got access to it on Nov 26th.  I uploaded the same prompt generated by img2prompt:



The results were a lot better than in the previous "Stable Diffusion 1 Demo":




*****
PROMPT

Next I used the Colab notebook program -- clip_interrogator.ipynb -- to generate a larger and most sophisticated AI Prompt after uploading my drawing:

I generated the prompt  below


Colab clip interrogator ipynb -- "a drawing of a woman in a dress, a drawing, inspired by miles johnston, figuration libre, acupuncture treatment, seated, ffffound, hairy bodies, wounds, 'untitled 9 ', from 2001, cd cover artwork, scratches, in salvia divinorum, woman is sitting, hives, baselitz"


MIDJOURNEY

Midjourney did not like the word "wounds," so I took that out before uploading the AI Prompt to Midjourney:

"wounds" is a banned word in Midjourney


Without drawing uploaded (top row)
 and with the original drawing uploaded (bottom row)
 to Midjourney --
plus the clip_interrogator.ipynb AI Prompt


AI Prompt only

AI Prompt and original drawing uploaded

"Text 2 Dream"
 (Deep Dream Generator)

I did not take the word "wounds" out before uploading it into "Text 2 Dream" in Deep Dream Generator:

AI Prompt only (first image)
and prompt and original drawing uploaded (second image)
to "Text 2 Dream" (Deep Dream Generator)




While using AI to generate prompts is rather interesting and novel, those AI Prompts are not then generating the best figures in Midjourney and "Text 2 Dream" in Deep Dream Generator (the prompts seem to be optimized for stable-diffusion (clip ViT-L/14)).

Also note that both AI Prompt generators labeled my drawing "Figuration Libre," which seems to be a French school of bad drawing in the 80s.


Just for Fun

I mixed things up just for fun, uploading the image -- created in "Text 2 Dream" with the clip_interrogator.ipynb AI prompt -- to Midjourney with the img2prompt AI prompt:




 Back to Hans Bellmer --
have the extra steps 
taken us anywhere different?




PHOTOS


The drawing session that night





Tom McGill
won third prize in the painting category


Jim McManus
showing in the Veteran's Exhibition


No comments:

Post a Comment

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