Sunday, 16 March 2025

Pieces of AI ART


AI ART CONFUSION

ARTIFICIAL INTELLIGENCE ART has been innovating rapidly since 2022 (as well as in 2023 and 2024) -- but things are changing abruptly now that anyone can merely chat with the new multimodal AI programs to generate images (like in GOOGLE AI STUDIO).  



de CHIRICIO
Old School AI Creation

MIDJOURNEY is still the best program for manipulating STYLES with AI.  I altered this photo I took from the Albuquerque PRESS CLUB with a Moodboard model that I created of de Chrico's style (--profile 25juclq):

Midjourney
EzGIF
Animation of a de Chirico Albuquerque
Midjourney Moodboard (--profile 25juclq)

This photo that I altered with Midjourney AI,
which I took from the Albuquerque PRESS CLUB
when we drew there on February 15, 2025



However I got locked out of Midjourney again on March 14th, 2025 for another billing issue (which is a problem on their end).  





















RECRAFT

I then imported the first MIDJOURNEY image into RECRAFT and generated new more, cleaner variations:

Midjourney
Recraft
Variation generated in Recraft

Midjourney
Recraft
Variation generated in Recraft

I've used AI to generate fake de Chiricos before, back in 2022:

Fake de Chirico Architecture
Fake de Chrico architecture from 2022


WHISK

WHISK by Google allows me to somewhat blend images, and usually defaults to a more "realistic" output than Midjourney:



Covfefe







RemoveBG







MULTIMODAL LLM

The MULTIMODAL option in GOOGLE AI STUDIO came out on March 12, 2025, letting us upload and generate images -- using the "(Gemini 2.0 Flash (image generation) Experimental" option -- by simply having a conversation with the interface.  KREA incorporated "chat creation" (using Deep Seek) on February 7, 2025, and I tried it out the next day.


prompt: "Can you give a blue tone to this image?"

GOOGLE AI STUDIO
GOOGLE AI STUDIO 
does a good job changing colors

Chatting with a "multimodal" AI program does not generate the best images, not as good as MIDJOURNEY and WHISK -- yet.  However GOOGLE AI STUDIO does do specific things better, like adding and changing colors in an image.


MY STYLE

I also asked GOOGLE AI STUDIO directly about my drawing style.  I have done this before with other AI programs -- and AI often gets it wrong.  Both GOOGLE AI STUDIO below and GROQ think my I use "vibrant colors," which is NOT true of my pencil drawings:


Can you tell me what kind of style that KRRRL uses?

"Based on the images I've analyzed, KRRRL appears to use a style that blends graffiti art influences with elements of cartooning and illustration. Here's a breakdown of the key characteristics:

Bold Outlines: Their work often features thick, clearly defined outlines that give their characters and designs a strong visual presence, a common trait in graffiti.

Exaggerated Features: Like in cartooning, KRRRL frequently exaggerates features like eyes, expressions, and body proportions, lending a playful and dynamic feel to their subjects.

Vibrant Color Palettes: Their use of bright, often contrasting colors is another hallmark of both graffiti and many styles of cartoon art.

Character-Centric: Many of their pieces revolve around distinct, often whimsical or slightly mischievous characters.

Dynamic Poses and Movement: The characters are often depicted in action or expressive poses, conveying energy and personality.

Graffiti Lettering Integration (sometimes): While not always the central focus, KRRRL occasionally incorporates stylized lettering reminiscent of graffiti tags or bubble letters.

Clean and Polished Execution: Despite the graffiti influences, the execution often feels quite clean and deliberate, suggesting digital tools or a refined hand style.

In summary, you could describe KRRRL's style as a vibrant and energetic fusion of graffiti aesthetics, cartoonish character design, and illustrative elements, often characterized by bold lines, exaggerated features, and bright colors.

It's worth noting that an artist's style can evolve over time, so this description is based on the images I currently have access to."

RUNNING AI ON 
YOUR OWN PRIVATE
SUPERCOMPUTER

The immediate solution is probably to download the AI ART programs and run them locally on a consumer supercomputer.  We did that at Quelab in 2023, running Stable Diffusion and Deforum on a beefy computer with  of VRAM (12 Gigs?) on an RTX video card.

We would like to do that again at Quelab's new location.  Below are some suggestions of programs to install (which I first mentioned on March 3):

A lot of these programs are one-click installs on PINOKIO. However some of the PINOKIO sources are "not verified" -- is that a problem?

 AI IMAGES:


3D FILES:

AI VIDEO:

WAN 2.1 (from Alibaba) is "open source":

Installing WAN 2.1:

Convetional LLMs:
Most online AI ART programs censor me, and won't let me upload my figure drawings to alter them. Ultimately the goal is to generate FAKE KRRRL drawings with AI:

August Summary
AI generated FAKE KRRRL drawings
back in 2020

After I trained a model in Leonardo.AI,  it used to make the best fake KRRRL drawings.  I especially like the way it transferred my drawing style to sculpture:


January 27, 2024 -- BERNINI



But am I just making more "Low Effort Nothings" with AI?


Posted by at

No comments:

Post a Comment

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

Subscribe to: Post Comments (Atom)