Tuesday 15 November 2022

Having Coffee with the Masters -- AI Art


How much influence can I have over an AI generated art result?



So I experimented in Midjourney (version 4) by uploading one of my drawings follow by the prompt: "woman in a coffee shop in the style of ___," -- where I filled the blank out with the name of a famous artist.


I uploaded this drawing to Midjourney


prompt: "woman in a coffee house
in the style of (Artist)"

Below are the results generated by Midjourney when I uploaded my drawing followed by the prompt: "woman in a coffee house in the style of ___," where the blank is the artist.

Then I uploaded the prompt only, with the name of that artist, and posted the result generated side-by-side with what was influenced by my drawing.

Also, some of these images were also colorized and boosted in Palette FM, an online AI color program:


Midjourney prompt: "woman in a coffee house
 in the style of R Crumb" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Hans Bellmer" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Victor Moscoso" --
without and with my uploaded drawing


Google Lens (no results)

Midjourney prompt: "woman in a coffee house
 in the style of Paul Wunderlich" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Giorgio de Chirico" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Salvador Dali" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Diego Rivera" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Thomas Hart Benton" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of John Currin" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Otto Dix" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Norman Rockwell" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Rick Griffin" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Hendrick Goltzius" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Michelangelo" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Lawrence Alma-Tadema" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of John Everett Millais" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Ilya Repin" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Hans Holbein" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Thomas Eakins" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Jacques-Louis David" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Caravaggio" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Raphael" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Leonardo da Vinci" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Mucha" --
without and with my uploaded drawing



Midjourney prompt: "woman in a coffee house
 in the style of Picasso" --
without and with my uploaded drawing


CONCLUSION

How much influence can I have over an AI generated art result?

My drawing did influence the AI image, as Midjourney generated full body figures and followed the composition somewhat.  The images generated with the same text prompt alone, without uploading my drawing, were often, if not always, better.

However many of the "better images" looked alike, seeming to default to the same style (Thomas Hart Benton, John Currin, Norman Rockwell, Rick Griffin), so perhaps my drawing added a little character to the generated image.

Also Midjourney seems to generate a result in the mirror image of my composition.

I also wonder if my Midjourney history is influencing the generations, since I feel that some of the results for other people are not as good.  Is Midjourney training on my history?  Is Midjourney creating a bubble around my generations?




Psychedelic Hans Bellmer

Perhaps the next step is to paint these generated images...or send them off to be printed or painted:
And then exhibit all the AI Art in a Faker Faire.


UPSIDE DOWN
R Crumb

I then uploaded my drawing upside down, and added the same prompt: "in the style of R Crumb."  The images generated were quite delightful.









From the second generation batch


TEXT 2 DREAM
Drawing Variations

I also altered my drawings in Deep Dream Generator, using their new "Text 2 Dream" option, and uploaded my results in the following blog posts:


Disclaimer:  This blog post was created at two coffee houses in Albuquerque:

No comments:

Post a Comment

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