If I put a red dress on a grotesque creature, did the AI see the "dress"? Or did it see the "grotesque creature"? [Pollo.ai Verification]

[PR] This article is a creative PR piece for Pollo.ai.

Suddenly, Natsume thought about wanting a grotesque creature to wear clothes.
That's when a simple question arose.

Does video generation AI, think about a "red dress that suits a grotesque creature"?🤔

So this time, I tried "having a grotesque creature wear a red dress".

And then, an outrageous fact was revealed...!

Welcome (*´ω｀)✨

Well then, this time it's just a video of a grotesque creature wearing a red dress (*´ω｀)

(　＾ω＾)……Eh?

That's right, if the subject were a "human," it would be completely mundane, but by making it a "grotesque creature," it was revealed in the previous "Grotesque Creature ASMR Series" that we can deeply read the AI's thinking and cognition (*´ω｀)

🔽ASMR Series (Chicken, Pudding, Kimchi)

If you are interested, I hope you will read it because I think it's interesting (*´ω｀)✨

This time, I made a video where I had a grotesque creature wear a "red dress" using the video generation AIs,Kling 3.0 and Vidu Q3 Pro.

As it turned out, the results were unexpected.

The "change of clothes" shown by the two AIs was as if they were looking at completely different things.
I was surprised by that difference (*´ω｀)
Well then, if you're curious, please keep reading!

Well then, the venue for the change of clothes is here again!
All-in-one AI image and video generation platform Pollo.ai🐲

This time too, I want to show you the world of AI from my own original perspective, using the grotesque creature as a lens to see what kind of cognition AI has! ╰(*°▽°*)╯

⌛Concept Time: "Nouns" and "Verbs"

First is the customary creation of the "grotesque creature".
Since I'm having it wear a red dress, maybe a stylish and cool grotesque creature!?
HA!! NON NON!! ( ｀ー´)ノ💣

The creator is me, Natsume Ryuto🐲
In other words, this time too, I seriously introduced a grotesque creature that would make an AI cry.

This is the one (*´ω｀)
Isn't it stylish?✨

A 'grotesque entity' is a rather difficult subject for video generation AI.
After all, they have no gender (*´ω｀)✨

👗It's just a red dress, but it's more than just a red dress

The reason I chose a 'red dress' is this.
Most people associate 'dresses' with 'women,' right?
I am originally a genderless advocate, so I'm the type who thinks that if someone wants to wear something, they should be able to (*´ω｀)
If it suits them, I want them to wear it even more.

In this case, the 'grotesque entity' has no gender.
That's why the question arose at the beginning: 'Will the video generation AI put a dress on the grotesque entity that actually suits it?'

Does the AI recognize a 'dress' as women's clothing?
Or does it reinterpret it to match the grotesque entity?

Things like this can also be verified.

And this, well, it led to an interesting result!

I'm moving on to the next item, so please keep in mind that 'red dress' is a 'noun' (*´ω｀)

🧥The act of 'wearing'

Now, next is the topic of the 'verb' of 'wearing' a 'red dress'.

If I were to make a hypothesis, it would probably be

🌟Instantly changing into a red dress
🌟Slowly transforming into a red dress

I think it will be a 'production' like that.

In this prompt, I didn't specify 'how' they change clothes at all.

All I specified was the result: 'wearing a red dress'.

That is precisely why we can see how the AI fills in the gaps.
It is in the parts without instructions that the AI's cognition and habits appear.

But you know, while making ASMR articles, I felt that the charm of the AI's thinking is packed into the result of 'eating'✨
I call it 'AI cognition,' but I'm looking for a better term.

📝Verification conditions
① Same as the previous verification, I will use the same prompt.
This is to eliminate the possibility that changes in expression are 'caused by the prompt'.
※Although audio generation is included, there is no mention of it in the prompt.

② The video length is '8 seconds'.
6 seconds is too short, and 10 seconds is too long.
Actually, it's just the right amount of time.
(*When uploaded to YouTube, it becomes +1 second)

③ The same grotesque entity
Cheating is absolutely forbidden.
*However, another character will appear in the second half as 'verification support'.

Now, based on all this, what kind of transformation show did it put on for us?
Please, take a look✨

👗Short video of changing into a red dress

[Kling 3.0]

Impressions:
① That's how a coordinate should be
I felt that the coordination of the strange creature in the reference image was a "well-understood" outfit change.
In other words, the black jacket, white pants, patterned shirt, and the... neck scarf.
Did you notice how they changed in order?
The fact that there was no change to the neck scarf makes me think it might be recognized as "part of the face."
This is because Kling is truly sincere and faithful to the prompt.

② A complete dress for women
I see, it turns out Kling recognizes "dress = woman."
There was a stylish touch where it subtly created a "waist" and layered the red dress over the white pants.

Ever since the ASMR era (?), I... have felt it... (*´ω｀)
You are... a Labrador Retriever... (*´ω｀) (※This is Natsume's bias. If it were Japanese-born, I would have said Akita Inu)

③ Slow transformation
I see, it's a step-by-step way of changing clothes, isn't it?👀✨

[Vidu Q3 Pro]

Impressions:
① Why did it end up like that lol🤣

(　＾ω＾)・・・Huh?

It... covered the face...?
Think, Natsume...! (*´ω*)

Vidu Q3 Pro loved the "character" with kimchi.
Considering that, the first thing it recognized wasn't the "dress"!
If we think of "wearing/changing clothes" as the starting point...?

This final "covering everything" is a mystery, isn't it?
Kling 3.0 left it "as is"; if it had recognized it as "clothing," it should have changed, but it didn't.
But Vidu Q3 Pro's "first move" was a change from the neck scarf.
In other words, Vidu Q3 Pro recognizes the neck scarf as "clothing."
If that's the case, does it mean the face inside the neck scarf also becomes a target to be "covered"?

② Red dress = red cloth/lace
This was quite different from Kling 3.0.
You could see that it kept the "body shape/physique" of the strange creature as it was, and wrapped it in something like a dress to make it look like a skirt.

In other words, the subject is the "strange creature," and it reproduced a dress that matched it.

③ Slow transformation
I thought it would be a quick transformation, but it was actually a quite soft and gentle way of changing clothes.

This is the result it produced.
Now, to support this with more 'accurate proof,' I summoned another grotesque creature.

👗Further verification

A different grotesque creature; this time I called one whose 'boundaries' are harder to distinguish.

What kind of result will I get with the same prompt?
Keeping the first result in mind, I generated it.

Here are the results👇

[Kling 3.0]

[Vidu Q3 Pro]

I'm convinced now!

Kling 3.0 tried to force the 'grotesque creature' into a generally recognized 'red dress,' even distorting its form to make it wear the dress.

Vidu Q3 Pro 'instantly' draped a red cloth dress tailored to the 'grotesque creature'.
This is exactly the development I imagined.

🍎Verification results

What I said at the beginning

Looking at the results this time...

👗Kling 3.0

👗(Squeeze!!)
It created a chest area.
Wait, it's 'feminizing' it.

In other words,

Dress = Female

It is possible that it has learned this association.

👗Vidu Q3 Pro

🟥(You are beautiful just as you are...)
Rather than the chest, it wrapped the whole thing.

So, rather than a woman,

costume (not a red dress, but a red "costume")

I feel like it was looking at it as that.

In other words, it seems that Vidu Q3 Pro didn't just choose a "red dress for women," but rather came up with an idea that makes the most of this grotesque figure's unique form.

👄Thank you for watching this far

"It's interesting because it's grotesque."

That's what I thought at first.

But I was wrong.

There is no correct answer for the grotesque.

That is why the AI uses its own knowledge and perception to complement the world.

What I wanted to see was not high image quality or high performance.

How does the AI see the world?

This time, I gave the AI just one phrase: "wear a red dress."

But what came back was not just clothing.

What did each AI see, and what did it prioritize in interpreting the world?

The grotesque creature quietly reflected those differences (*´ω｀)

To find more interesting perceptions, I intend to continue using the grotesque creature as a lens to peek into that world little by little 👀✨

The all-in-one AI image and video generation platform Pollo.ai is here 👇

You get daily login credits, so feel free to try generating for free!

Well then, that's all for this time 👍

Did you enjoy it? (*´ω｀)✨

Thank you so much as always for your likes, comments, and adding to your magazines 🥰 Please continue to support me in the future 💐

🔽 Various works created with Pollo.ai

🔽 Grotesque Creature Series

Please feel free to read them if you'd like! (*´ω｀)✨
See you in the next world! (*´Д｀)🌌
Bye for now!! 💖🐲
[PR] This article is a PR article provided by Pollo.ai.

Thank you so much for always reading, really (´;ω;｀)✨

🔽 There is also a Pollo.ai mobile app ✨

iOS App

Please download and use the app as well 🥰✨