Chapter 120: How Jina Creates Living Things

Jina looked at the various complex monsters filling the screen for a moment, then suddenly, an idea came to her.

“Couldn’t I extract features from the model’s form as prompts?”

Jina hurriedly moved her fingers, quickly typing new code onto the monitor.

The automatic rigging algorithm so far extracted a point cloud from the modeling and automatically embedded joints in appropriate positions based on its density.

But this alone was not enough.

It couldn’t efficiently process models with complex structures.

Furthermore, it couldn’t semantically classify monsters of various forms.

‘What if I automatically extract biological features and morphological information from the 3D model itself.’

‘And based on that, classify the species of the creature.’

‘Then, train AI models separately for each category to perform rigging and skinning?’

“Species-specific animation seems sufficiently implementable.”

It wasn’t an theoretically impossible story.

In an era where deep learning models that extract prompts from 2D images are widespread.

She thought it would be sufficiently possible to extract biological features from 3D models as prompts and generate corresponding skeletal structures.

“…This is it.”

The fingers tapping the keyboard get faster and faster.

Caffeine and dopamine race through her veins.

As the fog in her head caused by fatigue clears brightly.

Her stiff eyes glaze over blankly.

Code instantly filled the other screen.

It was an algorithm that divided the model data into a 3D grid and automatically generated a biological profile by analyzing the morphological features of each position.

For example, if it has many tentacles, a translucent surface texture, and flexible movement, it’s classified as an invertebrate.

If it’s long and thin, with many joints generated in a straight line on its torso, it’s an arthropod.

If it has a clear four-legged structure, the AI was trained to classify it as a quadrupedal mammal.

After several trials and errors like that, Jina’s new AI model was completed.

For testing, Jina dragged a pre-made model file and dropped it into the program.

First, a jellyfish-type monster spread across the screen.

[Analyzing form…]

Not long after the AI scanned the form, data was output onto the screen.

[Core Structure: Non-Skeletal fluidic core]

[Recommended Algorithm: Fluid Dynamics-based…]

Next, confirming that the desired result was output even when different types of monsters were input, Jina smiled faintly.

Not only did the result come out faster than expected, but it was so accurate that it was almost anticlimactic.

Of course, looking at the whole process, this was just the first step.

“First, the tentacle type.”

The moment the jellyfish-type monster on the hologram waved its tentacles.

The model surface seemed to undulate, then crumpled into a bizarre shape.

At that unnatural movement, a frown full of irritation formed between Jina’s eyebrows.

“This, I need to rewrite the surface deformation algorithm.”

Jina’s eyes, lost in thought, sharpened.

After a moment’s hesitation, Jina began tapping the keyboard.

The existing neural network that described human muscle and skin structure had to be completely overhauled.

She newly modeled it so that the surrounding tissues move hydrodynamically around a flexible, curved ‘core structure’ that forms the center of the tentacles, rather than a skeleton.

Here, she added a little imagination.

When it comes to tentacles, there are many scenes that can be staged.

Just off the top of her head, there are dozens of such and such scenes.

…She absolutely, never imagined anything that Pasta’s Cleanbot would red-flag.

‘Of course, it might accidentally, coincidentally burrow into clothes!’

‘And since it’s an aquatic creature, getting wet with liquid is also something that can’t really be helped!’

Like that, Jina created a kind of virtual skeleton and wrote formulas expressing the fluid’s soft movement, embedding them into the model.

It was, so to speak, a fluid dynamics-based weight model.

On top of this, she added a function where the neural network analyzes the model’s surface tension and dynamically adjusts values so that the tentacles bend naturally with each movement.

After implementing the formulas and running the training, a light test.

Theoretically, she thought it was perfect, but as always, reality had many variables.

In the first test, the tentacles were excessively stiff and unnatural.

Next, they stretched endlessly, undulating like jelly.

“Did I set the range too wide?”

Jina re-examined the data several times, repeatedly adding and removing various parameters to the weight values applied to the tentacle movements.

Even while adjusting the formulas, she didn’t touch the terms related to ‘that’ at all.

Sexiness is the identity(?) of a demon.

Simultaneously, it’s Jina’s aesthetic and philosophy.

This is, perhaps, like a king-size bed in an inn in a fantasy world.

After endless repetitions of corrections, Jina found the optimal parameters.

Watching the flexibly and smoothly moving tentacles, Jina let out a dry sigh.

“Finally done…”

One mountain after another, the next model, the arthropod-type monster, was even more perplexing.

Watching the surface unnaturally crumple each time a joint moved, Jina’s face also contorted fiercely.

For now, the problem was clear.

There were too many joints defined within the model, and the connections to the surface were thin, so simply transferring existing animation data would never create natural movement.

A robotics-related paper she had casually skimmed once vaguely comes to mind.

Did it say that arthropod movements, like precise robot joints, engage and move with exact angles and tension…

Naturally, she doesn’t remember the formulas. She doesn’t even remember the title.

In the first place, Jina isn’t a mechanical engineering PhD. Even less so an entomology PhD. More than that, she doesn’t have a PhD at all.

“Would it be better to have the artificial intelligence learn on its own and find the weights…”

When in doubt, just throw it at the AI and think, and half the battle is won.

Jina designed a semi-autonomous weight optimization model so that the deep learning neural network could repeatedly learn the optimal weights between the skeleton and the surface.

After all, looking at the movement of each joint, it’s all just moving left and right on a 2D plane.

Like that, she had it repeat the movement of each leg hundreds of thousands of times to find the optimal weights.

A task that would take a lifetime if done manually by a human, the AI can do in just a few hours.

While the model was training, Jina placed the wolf-type monster on a new viewer.

“You, well, you’ll be finished quickly.”

The algorithm was already clearly defined.

She developed an algorithm that crawls quadrupedal animal videos online, extracts movement data from millions of frames, and reconstructs it onto a 3D model.

As she trained two AI models simultaneously, she could feel the expensive cloud GPU server she rented screaming.

She started thinking she wanted to build a personal server computer soon.

And while at it, she wanted to move to a slightly larger room too.

“Maybe I should look for a room after the Jinro World content is over…”

The money she earned as a freelancer in the past year is staggering.

Just last year, she was worried about how to pay next month’s rent.

“Maybe it’s okay to move to the countryside altogether at this point.”

Her current room was just a single room she got solely for commuting to grad school.

If she’s going to continue doing remote-centered work like this, she thought it might be okay to leave the metropolitan area altogether.

“But if I move far from Seoul, I won’t be able to get drinks from Daon or Youngjun, that would be a bit unfortunate.”

While planning her life and carving NPC modelings, the training of the two AI models was completed.

Jina displayed the arthropod model on the left monitor and the wolf model on the right monitor, checking the animation.

Left. The monster’s legs moved slowly.

It even felt like the distinct segmented feel and taut elasticity peculiar to insects.

Right. The wolf began to gallop agilely across the meadow as if hunting prey.

The movement, capturing both the tautness and flexibility of the muscles, is dynamic.

On top of this, the realistic expression of the growling snout is perfect.

“Now… I need to work on the auto-skinning models for other categories too.”

Also, after the AI model design is finished, a mass production pipeline needs to be established.

It’s natural for another job to start when one job ends.

Unaware that the sun had set and risen again beyond the curtains, Jina, lost in deep immersion, tapped the keyboard.

She wasn’t just creating technology.

She was creating an ecosystem, the life of a new world.