7-15-2024: ai in art (in-work)

artgraphic-db to develop a "funny jellyfish" to digital image ... (coming soon)
This post is currently "in work"...

Funny Jellyfish

(Machine learning and art)

Outlines:

  • Introduction is about the exercise of the orange astronaut cat (Illustrator made both the cat and the equipment orange),

            which serves as an introduction about AI and machine learning;
  • How to know, How to understand the AI black box?

  • The Funny Jellyfish follows the same idea than the orange astronaut cat,

            with a more graphical technical application or standpoint

Introduction

The question of using an AI tool has met recently artgraphic-db. AI for Artificial Intelligence, tends to spread over the world of creativity.

The idea has popped up due to new capabilities that Adobe has wanted to have along with Illustrator.

So, they now have an awesome “illustration with generative AI” option, as depicted by the example snipped below, that allows to generate illustrations based on one’s desire, in terms of typed in request.


In the demo, a request of (creating an illustration of) an orange astronaut cat is typed in and two proposals come out generated by

AI generative AI (as snipped below).


That’s cool, and what’s behind the AI engine, how it is made, the assumptions made, and the question

of how to make it myself rose.

Of course, it is not pure imitation to do like Illustrator or at least, not only.

Other posts mentioned the use of AI tools, like, for instance the magic eraser AI tool by pixelcut. And, the writer of this blog likes, personally, to use AI search engines (like MS copilot or Google Gemini).


So, in this section, I am not focusing on a “funny jellyfish”. I am trying to get some best practices on
how I could make my own AI, or at least how I could follow a sort of AI way, or some sort of
machine learning methodology to train a method that could be, then, derived in terms of a computer
program or routine.

Keeping on with the idea of the “orange astronaut cat” I have tried two other tools to generate possible

visuals of “orange astronaut cats”.

  • First, I have tried Open Art AI (OpenArt AI), I don’t remember how I got this particular app of this particular patron.

            There are plenty of examples of implemented AI.
Following the idea, I used the app to create these “orange astronaut cats”, as seen below,
for comparison with the AI generative AI.

I have requested an orange cat, which was created (left, an orange cat) before I have realized the weakness of the requirement.

Then, I requested an orange astronaut cat (right, an orange astronaut cat by Open Art AI).



In fact, AI creator studio gives several degrees of freedom to create its own generated picture, among

“what not to see”, lighting options, Angle and framing options, lens & capture, film (in terms of

photograph film), art style, vibes, mood and scale.


online-convert.com has many tools online to handle web media, including converters, vectorization,

formats and editing that can be installed as chrome extensions.


Thus, I wanted a watercolor art style but, I kept the photograph style and the requirement that

watercolor and/with photography showed incompatible, and I got the photographs as seen below, left

and right, variants of “an orange astronaut cat” by AI creator studio.


Based on these experiences, I, with artgraphic-db, imagined (wrongly or rightly, based on my own

understanding of machine learning) that these AI’s based their learning on some sorts of visual

datasets, of cats, and/or astronauts, and/or -why not- astronaut cats, and/or -why not- orange

astronaut cats.


Thus, I made a quick internet search and came out with the two following search results:

Google search Orange cat

Google Search Orange Astronaut Cat


So, finally, perhaps my assumption was not stupid and such datasets of visuals would be good training

sets. An orange cat popped up quite naturally from the first dataset -in fact, I just imitated a sorting

engine with some intuitive criterion of activity and action (as we imagine astronauts can be-. From the

second dataset, another picture popped out of what we can fairly imagine as an orange astronaut cat

(with equipment, somewhere lost in space, amidst celestial infinity).


Finally, I imitated a little further and created an image, mainly based on, from, the first sample (I

imagined that the picture could obtain a very good score and match with some “orange astronaut cat”

expectation).

However, some penalties, probably, had to be overcome. The character would have been in a space

environment because, I was trained to see astronauts in space, (more than office), and trying to move

or fly, (so, that I have put the cat in a more astronautical attitude).

The score improved quite impressively for an “orange astronaut cat” based on these good matches

(space environment, busy, floating toward, …), such that it was not necessary to dress the cat within a

specific equipment, nor the need of creating a spaceship closeby. 


An orange astronaut cat by artgraphic-db:

  1. Generate an orange cat in any environment (in this case it is an existing image from any database, or training set);

  2. Select the cat (using intelligent scissors);

  3. Modifying (here, rotating);

  4. Give the illusion of astronautics (here a background of “cell noise” with convenient settings -at

    this step, the cell noise has been already looking like cosmos- is artificially generated, which needed
    an invert, then, a threshold color treatment).


“Orange Astronaut Cat” Summary:

Left to right: (a) AI Generative AI, (b) Open Art AI, (c) AI Creator Studio, (d) artgrahic-db


Clue: Do you think that a “style” requirement might influence the generation of the image? If you think

“yes” perhaps you made up your mind and thought that the AI has not a lot to do this time and could

generate an image like the following (left), with a minimum of change, or an AI generated like the latter

(right) (with more training including, generating an orange cat, style photograph, generating an

equipment and other stuff, style cartoon, generating a background seeming cosmos, generating details

corrections like whiskers).


Drill 2: How do you think a machine or an engine has to be trained to generate the following image?

How to know or even, How to understand the AI black box?

Corporate

It is not big news, AI, AI for Artificial Intelligence has invaded under this wording most aspects of our

daily environment, including art, through generative AI techniques, but also business and corporate.


Where can we see AI?

How to learn to recognize AI?

Since the beginning, computing is a matter of repetition (of tasks), (reliable and infinite) memory,

accuracy and speed. So, it is not a big surprise to assign computers to automation (sequence of

commands without human beings intervention).

Computers are not the only machines that can be automated, and generally speaking it is the matter of

rocessor programming (automats, robots, androids…).


In addition, AI is also about decision making, in place of human beings, or, say, instead of instant

human-type decision making, so, AI is also about prediction.

Said in two bullets, AI has something to see with:

  • automation

  • prediction


The world of business sees the introduction of AI for lots of reasons, so let us revert to two examples.


SAP (SAP SE is a German multinational software company based in Walldorf, Baden-

Württemberg. It develops enterprise software to manage business operation) has made use of

AI art (in artgraphic-db understanding) to talk about advantages of AI to business (video):

  • AI can do something useful (labels categorization, objects recognition) 

  • Business stronger

  • Decisions faster

  • People better [artgraphic-db guesses in the sense in better mood or in better disposition,

            because AI makes things easy]


In the video presented, the cat has the style of a Van gogh Impressionist famous painting with features

that are to be kept all along the video, so if we think how it was made, it was probably a matter of

training to an impressionist style to apply to make the cat in the same way all along the video. In fact,

even in 3D making, it should be necessary to create the texture and tune it to the motion of the cat

according to some graphic trick (this is where there is AI, probably, in this video).


IBM has called their AI solutions under the vocable watson. First, they remind us that AI is

not new. Sometimes called cybernetics, sometimes called nothing special, computer

programming has always involved automated tasks, and some of us made AI without being

aware of it, as long as they needed to generate cases, discriminate or sort them and reproduce

some conditions to imitate some good or desired circumstances.


Why do leaders want to adopt generative AI?

– Make better and faster decisions by applying generative AI to their data

– Use AI to automate decision-making

– Democratize AI for all AI value creators and builders–regardless of skill-level or role

– Amplify growth by automating manual tasks, uncover hidden patterns in their data, and create new

business lines, products and services

– Support the next-generation of the workforce with data-driven insights

– Enhance customer experiences with more hyper-personalized interactions

data-driven insights

Most of us will have to meet data and to understand data-driven Decision Makings, in the understanding of IBM Watson.

Democratic

I have made some search about how an average citizen can get information to make up his mind

about AI. Perhaps a good start is to search the internet, so, here my google searches on organizations

feeding the citizen debate about technology and about AI, and how machine are learnt to imitate

circumstances to produce results and make decisions faster:

Search AI and democracy non profit

Search technology and democracy non profit


Referring to the previous paragraph (about business AI), things go faster than the framework, maybe,

but, nothing prevents from making an opinion. Of course, AI is going to pervade until the government

(https://ai.gov/apply/).


About talking on technology, and AI in particular, and “development of AI products and services”, there

is information online  https://cdt.org/cdt-ai-governance-lab/ provided by the Center of Democracy and

Technology which has other initiatives like TechTalks; in particular there is this podcast Talking Tech with Nabeel Gillani on The Black Box of AI in Education (about AI in or within education).

They have practical actions, including recently about AI Act in Europe

Correlation, models ? What is behind, that is the question.


Not specifically about AI, but there is AI too, on this website

https://www.tutorialspoint.com/artificial_intelligence/index.htm


AI as invaded the chatbox, including at Mozilla (the ones who are related with MDN and SVG tutorials

and support (quoted here and there in the posts or the website)

https://developer.mozilla.org/en-US/


Diagrammatic (alternative perspective to data-driven), used by SVG filters website (just the idea if

diagrams could be handled by AI…) 

https://www.jointjs.com/


I have got an email Welcome to Pinecone about Vectorization, generally speaking. It is a real practical

email about building a database and how AI algorithms use vectors in a representation space that are

general (it is not a matter to generate a cat but an output “Y”, in the representation space of felines

(X1), colored (X2), with or without equipment (X3)). Plus, we already know that it will be a piece of

code of Python programming language.

In fact, Pinecone intends to propose a private ecosystem to build training sets (which are open source

until today).


This section is quite far of the topic of Art tools.

Technical


For the very novice, artgraphic-db has proposed a selection of youtube videos to familiarize with terms and ways of doing needed with a computer WITHOUT human being intervention.


Below, is an example of a sequence of lines to use with the Vector Graphics program Inkscape to convert automatically Vector Graphics to PNG. As it was used later and as it was talked in other posts, this is a sequence of lines to run a software from a shell without a human interface (except the shell).

@echo off

for %%i in ("%~dp0*.svg") do (

echo %%i to %%~ni.png

"C:\Program Files\Inkscape\bin\inkscape.exe" --export-type="png" "%%i"

)


AI in Art


https://openai.com/index/robust-adversarial-inputs/


Generative AI (AI Art Generators) has also a certain interest to the art amateur. This category of AI,

said generative, aims to create from a text requirement, illustrations or digital images.

Additionally, a fork of the 3D vector graphics and animation software blender (see post) has come out

under https://neuralblender.com/ for the generative part.


The Funny Jellyfish

Google search Funny Jellyfish

Finally, back to the Funny Jellyfish in this section. The set of jellyfish images is also used in another post.

Part 1: Define the framework

Funny

[7-30-2024] I have all of these new things in mind about machine learning and all that stuff and I see

major things to overtake: 1. Find a vectorization (I guess it is about features that can describe a “funny

jellyfish”) 2. Start to think about a way to categorize and select features that make a “funny jellyfish”.

I realize that I have to include some semantics in my AI. What is funny, or what can make something

funny (if different that making a funny jellyfish).

(I find a a starting point here, with this webpage: https://www.greatworklife.com/funny-things-to-do-and-say/

Maybe, this jellyfish will have to do or say something funny?)

Automatic conversion to SVG images

(Nobody is supposed to know, but here the results of an automatic run of converting the selection of

jellyfish pictures (presented in the post about filter effects), to an SVG format).

It was hard for my potrace to work in grayscale but there is something interesting here. The shadow of

the jellyfish has characteristics that make it unique (they are jellyfish for sure).


The question rose if the generative AI had to work on Vector Graphics or raster images or both?


Jelly #8 case to feed the discussion

Original grayscale bitmap (what potrace sees).

Versions v2 and v3  (what potrace sees on the left).

Color Palette

What colors using?

I guessed the AI had to learn how to make a palette and reduce the number of colors, to approach a

style “cartoon” -which adds to the funniest-.

The colormap, 64 colors, as a whole on all the “dataset” images.

A first “AI” generated

[8-2-2024] An attempt to test the preliminary schemes (identify a funny scene or storytelling, generate

a jellyfish with outline and relevant colors) ended in the sample result below, using the identified

techniques: jelly #10, vectorized, recolored according to a jellyfish palette, and funny features added):


Notice the additional elements, not parts of the jellyfish color features. Perhaps, the AI could come up

with something similar ?.




Link to document


Comments

Popular posts from this blog

5-25-2024: LEDs