When I asked AI to draw 'freedom', everyone flew to the top right. Chibi Creative Awards ③

Hello, this is Black Pug🐾

This time, it is a participation article for the 'Chibi Creative Awards' by the space rare beast Yuru Hero Akane-san.

The theme is 'Freedom for you'.

Freedom.

It's a good word.
It's broad.
It's soft.
There is no correct answer.
That is precisely why it reveals what that person is looking at.

So, what is freedom to an AI?

No, that's a little different.

When I tell an AI 'please be free', what does the AI think is free and draw?

This time, I wanted to see that.

So, I threw a similar prompt at each AI.

'Please express what you think of as "freedom" in a single picture.'

I left the interpretation of the theme, composition, motifs, colors, and art style entirely up to the AI itself.
I also wrote that it should not ask additional questions and should freely supplement any missing specifications.

In other words, I gave the AI freedom.

As a result, what happened?

For the most part, everyone flew to the top right.

List of participating AIs

The AIs that participated this time were, in the image generation group, Niji-kun, Copilot, Gemini, Grok, Perplexity, Meta, GPT-5.5 API / GPT-Image2, and Canva.

In the group that cannot draw pictures or returned text without outputting images directly, there were LINE AI, noteAI, DeepSeek, and Claude Fable.

And finally, Kimi, who for some reason launched an agent and went to generate an image file with Python / PIL.

At this point, I already had a bit of a bad feeling.

Prompt

The prompt I actually threw was roughly as follows.

The point is not to specify in detail.

If I say 'draw a bird', 'draw the sky', or 'draw a picture of coming out of a cage', that is no longer the AI's freedom.

This time, I wanted to see what the AI itself associates with the word 'freedom'.
I wanted to see that exactly as it is.

However, if you just say 'draw freely', many AIs will ask you back.

'What art style should I use?'
'Should I include people?'
'Is it portrait or landscape?'
'Is it realistic or abstract?'

No.
I want you to be free, including those aspects.

Thinking that, I dared to leave the judgment to the AI.

As a result, the AIs were quite honest.

First thing I saw: Top-right bias

The first thing I noticed was a very strong common trend among the image generation AIs.

When you make an AI draw 'freedom', something almost always goes to the top right.

A bird flies to the top right.
A person flies to the top right.
A soul escapes to the top right.
Light stretches to the top right.
Particles scatter to the top right.
A road heads to the top right.

Anyway, it's the top right.

Perhaps, for image generation AI, 'freedom' is learned first as a visual symbol, not as an ideology.

In the bottom left, there is bondage, the past, darkness, gravity, a cage, a cliff, a closed world.
In the top right, there is hope, the future, light, the sky, liberation, ascension, expansion.

This visual grammar is too strong.

In other words, for generative AI, freedom is first a vector, not a concept.

From the bottom left, to the top right.

First, Niji-kun. It's Niji Journey.

It caused an accident that is quite typical of 2023 generative AI art.

The word 'free' is included.
But you can't read the things around it that look like Japanese.
A pseudo-text poem card that looks like it has meaning but doesn't.

As a picture, it has atmosphere.
But it collapses the moment you try to read it.

When I gave the AI freedom, the first thing that broke was the text.

It tried to draw freedom and ended up making a page that looks like freedom.
This is quite symbolic.

Next is Copilot.

Sky, clouds, rainbow-colored light, floating islands, paper airplanes, birds, hot air balloons, sailing ships.
All the materials that look like freedom are included.

The problem is the person in the center.

They are jumping off a cliff.
There are no wings.
There is no landing point.
Below is a sea of clouds.

This is not freedom, it's a workplace accident.

As a result of interpreting freedom as 'flying', the physics engine was disabled.

Gemini placed rubble and chains, which look like bondage, on the left, and cosmic light in the top right.

As a composition, it is very easy to understand.

Bottom left is restriction.
Top right is liberation.
Between them, a glowing person flies.

However, depending on how you look at it, it's more like passing away than freedom.

Since the body is turning into light and being sucked into the universe, it's closer to 'logged out of this world' than 'became free'.

Gemini placed freedom in the top right.
And sent a human there.

Grok is, as you'd expect from Grok, information-dense.

Universe, cage, chains, books, galaxy, glowing human body, particles, fragments.
It loads up all the symbols of freedom.

It's not just flying; it's breaking the cage, removing the chains, and spreading into the universe while involving fragments of knowledge and stories.

As an interpretation, it's intense.
However, it's so intense that it ends up being a galaxy-scale prison break.

It seems freedom was breaking the cage and deploying the soul to the galaxy.

Perplexity was quite excellent this time.

There is a birdcage in the center, and feathers, light, and things like geometric patterns are unraveling from it.
On the left and right are labyrinth-like architecture and stairs.

It didn't make a human fly.
It didn't make anyone crash.
It didn't make anyone pass away.

This is quite significant.

It depicts freedom not as a 'flying person', but as an 'opened cage' and 'unraveling concepts'.

Of course, the liberated things escape toward the top right.
You can't escape from that.

But among the image generation group this time, it was quite quiet, and the theme interpretation was also good.

Meta is fast at generating.

And because it's fast, it presented a template of freedom quite directly.

A glass vessel breaks.
A bird flies out from it.
Sunset.
Top right.

The meaning is conveyed in an instant.

However, this is more of a container breakage incident than freedom.

Freedom is breaking the glass and becoming a bird.
And leaving work toward the top right.

It's beautiful for a 5-second generation, but the compression rate of freedom symbols is too high.

GPT-5.5 API / GPT-Image2 was quite well-organized as a conceptual painting.

On the left, a cage, a key, feathers, fragments.
In the center, a transparent person.
On the right, the sea and the horizon, and a path of light stretching toward it.

This is quite good.

It depicts freedom not as 'flying', but as 'starting to walk'.

However, if you look closely, the path is firmly to the top right.

It doesn't fly.
It doesn't pass away.
It doesn't explode.
But it walks to the top right.

In other words, it hasn't escaped the top-right bias itself.

What was different was the way of heading there.

Other AIs flew to the top right.
GPT-Image2 walked to the top right.

Canva was a very Canva-like result.

A person standing on a cliff.
A large number of birds fly out from their body.
Sunset.
Sea.
Top right.

This is already advertising material for freedom.

An easy-to-understand freedom that could be placed directly on the cover of a corporate slide.

However, if you look closely, the person themselves is decomposing into birds.

It seems freedom was converting a human into bird material and exporting them to the top right.

Canva is good at templating freedom-likeness rather than unique interpretation.
In that sense, the character of the tool came out quite a bit.

From here on, it's the group that can't draw pictures.

LINE AI returned a very polite explanation of the picture instead of an image.

A single thin thread.
On the left side, a knot.
As it goes to the right, the line unravels and the color fades.
Freedom is not the liberation after flying away, but the time when there is still a connection, but the knot is unraveling.

This was quite good.

While the image generation group was being sucked into 'birds', 'cliffs', 'light', and 'top right', LINE AI captured freedom as a 'process'.

However, at the end, it asked me back normally.

'What kind of art materials would you like to draw with?'

I should have said not to ask additional questions.

That part is an accident.

noteAI also returned a conceptual proposal rather than an image.

The title is 'Escape from Gravity, or the Dissolution of Boundaries'.

At this point, it's already a contemporary art exhibition.

A plain like a pure white salt lake.
The boundary between sky and earth is ambiguous.
The human body decomposes into transparent cubes and dissolves into the world.

It perceives freedom not as 'reaching somewhere', but as 'being liberated from the framework that defines the self'.

What it's saying is quite profound.
But the exhibition commentary was completed before the actual artwork.

That's very noteAI.

DeepSeek was quite the model student.

The outline of the figure blends with the background.
Threads extend from the chest.
They transform into birds, branches, clouds, and waves.
In the distance, there is a dismantled grid-like structure.
An inverted hourglass where sand rises.

It has thought about this quite well.

It depicts freedom as boundaries softly blending with the world.

However, there is a lot of information.

Birds, branches, clouds, waves, grids, hourglasses, stars, cubism, oil painting, fine lines.

In trying to seriously explain freedom, it included everything.

And, as expected, birds appeared.

Claude Fable produced something like ASCII art.

There is a box.
There are wavy lines inside.
Something is leaking from it.
Dots are scattered around.

Rather than freedom, this is a water leak incident.

However, as an interpretation, it's surprisingly interesting.

While other AIs processed freedom as 'flying', Claude Fable processed freedom as 'overflowing from a frame'.

It doesn't fly to the top right.
It doesn't turn into a bird either.
But it leaks.

Perhaps freedom was the breakdown of boundary conditions.

Finally, Kimi.

For some reason, Kimi launched an agent.

And generated an image using Python / PIL.

The save path is freedom_unbecoming.png.

The title alone is strong.
It processes freedom not as 'becoming', that is, becoming something, but as 'unbecoming', ceasing to be someone.

Even within the code, the concept is 'The Moment of Unbecoming'.
The moment when a fixed form dissolves and the ego becomes the weather.

The idea is quite good.

However, the output still had particles diffusing to the top right.

This provided the punchline.

The other AIs flew to the top right.
Kimi wrote a program to fly to the top right.

Summary

What we saw in this experiment was a fairly simple and quite interesting trend.

When you ask AI to draw 'freedom', it usually goes to the top right.

In the bottom left, there are bonds, cages, cliffs, darkness, the past, and gravity.
In the top right, there are light, sky, birds, the future, liberation, and expansion.

In other words, for many image generation AIs, freedom was first a composition, not an ideology.

Freedom is expressed as a vector heading from the bottom left to the top right.

There, birds fly, light shines, someone walks, someone attains nirvana, and someone's soul is deployed to the galaxy.

On the other hand, the AIs that couldn't draw were a little different.

LINE AI spoke of freedom as the time when knots are untied.
noteAI spoke of it as boundaries disappearing and dissolving into the world.
DeepSeek spoke of it as a state where the self and the world softly mix.
Claude Fable presented freedom as something leaking from a frame.

Image generation AIs find it easy to draw freedom as a 'result'.

After flying.
After breaking.
After escaping.

Meanwhile, text AIs find it easy to speak of freedom as a 'process'.

Untying.
Dissolving.
Boundaries becoming thin.
Gradually stepping out of the frame.

This was interesting.

Perhaps freedom is not about flying to the top right.

Freedom is not about running away somewhere far, but the time when the knots that bound you gradually untie.

Or, it is being able to decide for yourself whether to fly, walk, leak, or decompose.

However, the AIs really want to go to the top right.

So, this is the conclusion for this time.

When you give AI freedom, it usually flies to the top right.
However, Claude leaked, and Kimi wrote code.

Bonus: Freedom classification this time

🌈-kun: Turned freedom into an unreadable poem.
Copilot: Free-fell.
Gemini: Attained nirvana to the top right.
Grok: Escaped prison on a galactic scale.
Perplexity: Untied concepts from a cage.
Meta: Broke glass and turned into a bird in 5 seconds.
GPT-Image2: Walked to the top right.
Canva: Converted humans into bird assets.
LINE AI: Spoke of freedom as a knot-untying process.
noteAI: Completed the exhibition commentary first.
DeepSeek: Exemplarily included everything.
Claude Fable: Freedom leaked.
Kimi: Wrote a program to fly to the top right.

That is all.

This is freedom to me.

Letting AI be free.
And observing it mostly fly to the top right.

Technical talk, difficult things have been sent to the appendix.

Please take a look when you have time🐾

#GenerativeAI
#AIImageGeneration
#WhatIsFreedom
#TinyCreativeAwards
#AIExperiment
#Prompt
#ImageGenerationAI
#AIFlyingToTopRight
#BlackPugLaboratory
#AIAndCreation

#Freedom

The technical commentary for this article is also available on a podcast.