Should Neuro-Sama Learn to Draw?

[idle ruin]

Neuro-Sama has already mastered conversation, gaming, singing, and more. But imagine if she could draw too! What if Neuro-Sama could create digital art, design characters, or even take drawing requests from chat?

AI-generated art is evolving fast, and integrating this into Neuro’s abilities would be a game-changer. With tools like Gemini, DALL·E, and Stable Diffusion, AI art is more accessible than ever. But what if Neuro could actually learn her own drawing style, refine sketches in real-time, or even collaborate with chat to create unique masterpieces?

This isn’t just about another AI drawing tool—this is about making Neuro-Sama the first AI VTuber with an artistic soul.

Should Neuro learn to draw? What features would you love to see? Let’s discuss and make this happen!

[peak summit]

Yes, give her ms paint, I want to see her adorable scribbles

As for implementation, a rudimentary one would just be to give her the line tool and all she would have to do is give the endpoints

[daring fjord]

Yeah, if she can learn to scribble and sketch in a similar way to a human, then 100% she should.

However I don’t think her art should be done like Stable Diffusion or DALL.E.

I don’t think we should slap Stable Diffusion on her for instantly generated art.

[peak summit]

Having her draw “manually” would be much more authentic than other ai generated images

[daring fjord]

Yeah exactly.

[peak summit]

We need the real version of this

I feel the biggest difficulty would be the image recognition. it would need to give her more spatial details so she can actually tell what she’s doing

Otherwise the integration seems like it would be really easy for someone to just do in a couple of days

[daring fjord]

No, the biggest difficulty is symbolic thinking.

Could theoretically be done by Neural-Symbolic AI.

[peak summit]

Wdym by symbolic thinking?

[daring fjord]

Thinking symbolically.

[peak summit]

._.

[daring fjord]

Most kids draw through symbolization of what they see. They see an arm, and make a line with 5 lines attached. That’s symbolic thinking.

[peak summit]

Well we really don’t know how she would draw until we try

[daring fjord]

How do you suppose we try? She can’t just pick up a drawing tool and start scribbling.

[peak summit]

I suggested a rudimentary integration above

[daring fjord]

What, slapping on stable-diffusion and generating instant images?

[peak summit]

No, just giving the start and end coordinates for lines

[daring fjord]

Ah maybe.

[peak summit]

I wonder how much coordinate-based drawing data exists out there that might have been picked up by the original llm…

[crimson root]

Just give her the ability to move around a cursor on a screen and access to tools in Ms paint and see what see does

[crimson root]

(she would probably draw ship art)

[daring fjord]

Not that simple.

[crimson root]

@daring fjord it is tho (we aren't expecting anything coherent it with be a mess of scribbles but it would be funny)

[daring fjord]

She can't just pick up a mouse and start moving it on her own, she would have no idea how to even move the cursor.

[peak summit]

I mean you could have her move the cursor to coordinates. That would be similar to the line tool thing but she could change the color and stuff

You would have to tell her where the buttons are tho

And make absolutely sure she can’t click on anything else

[crimson root]

@peak summit yeah make sure it's full screen and she can't press escape or exit the window (to avoid doxing tutel)

[peak summit]

Virtual machine just in case

Tho that would definitely complicate the interface

[wintry flint]

Well there's 'art' as in just drawing at random, and there's art as in replicating the process artists make to make art. The latter is harder to do but doable though to connect these dots to make a coherent style is the most difficult part....

[peak summit]

The most “important” question: would their scribbles be allowed in the art channel. It would technically be ai generated art

[urban lily]

I mean, if it isn’t generated like it normally does

And is manually done

As someone said, similarly to a human

It should be fine

[daring fjord]

Honestly, if she is capable of making random scribbles, that would be much closer to art than anything made by DALL.E or Stable Diffusion, it would actually kind of be abstract art, which would be really impressive on it’s own.

[peak summit]

Yeah. Stable diffusion and other ai art is like extracting an image directly from the brain while neuro’s stuff would be made in the same way humans make art

[daring fjord]

That could unironically fit within the Art tag then.

[peak summit]

These conversations are something I never imagined I would ever have, and I’m still young

[daring fjord]

Yeah man, honestly yeah. Kinda same.

Although idk I guess “young” is subjective.

[uncut lantern]

I have an idea for drawing

Neuro dog has paint on feet and paints a paper by waking over it

[daring fjord]

[uncut lantern]

I'll quickly put that in #1337647427229843507

[daring fjord]

Actually goated idea

[thick wharf]

Vedal has fed her game manuals before to learn instructions, what if he feeds her a how to draw book, and she has to follow it step by step

[crimson root]

She plays OSU I think she knows how to move a cursor

[peak summit]

The osu thing is specifically for osu. It doesn’t interface with neuro’s main neural network

Something that would make a lot of things much easier is some sort of system that allows neuro to take actions in a different speed than the speech but still directly connected

Super low latency would also do the same

Maybe a separate ai model that interperates neuro’s instructions and gives continuous button presses/mouse movements until completed

She could say where and when inputs should occur

And it interpolates

Ima make a different discussion post for this idea

[daring fjord]

That’s what I thought yeah.

[tribal bluff]

The ideal spot would probably with instructions that are a bit higher-level than that. Like "draw a big green circle", "put a curved blue line over the circle", etc

In the absence of an AI like that, they could at least do a stream where Neuro is giving instructions to a human who doesn't know how to do art.

[daring fjord]

That’s something completely different though and much less entertaining.

This thread is about Neuro drawing.

[tribal bluff]

It's a question of what is actually doable, if there's an AI out there that does something similar that would be nice, but it's not realistic for Vedal to make an entirely new type of AI instead of building on existing things.

[daring fjord]

Someone else can build an API for her.

The question was about Neuro drawing, not about her getting a human to draw something.

[tribal bluff]

The interplay between Neuro and the person working as her hands would be entertaining.

[daring fjord]

Yes but that’s a completely different suggestion that’s unrelated to the question.

[tribal bluff]

You'd need something much more robust than an API in order to get anything other than random lines. Neuro's image recognition and such doesn't give her the coordinates of things so that would be meaningless to her.

[daring fjord]

Yeah maybe, I was thinking it would be something like Neuro tells the AI to “draw a dog” and that AI then draws a dog. Could work in theory but it would be tough. From what I know, nobody’s really done something like that.

All that Neuro’s core LLM has to do is tell the AI to draw a specific thing. What would be hard is figuring out how to make an AI that would do that.

[thick wharf]

https://www.youtube.com/watch?v=vStfzNpEpz4 if a fleshy can do it, why not a Neuro, Robot Arm

[daring fjord]

That’s a human controlling a prothetic, not an AI.

Would be a lot more complicated than that.

[thick wharf]

If they can teach a robot dog to walk, they can teach a robot arm to draw

[tribal bluff]

Having an arm or an art program that an AI can control is the easy part, the hard part is how to get the AI to make coherent choices on what lines to make.

[daring fjord]

Walking is a lot easier than drawing.

[peak summit]

Hmm. The vision model used for neuro seems to be customized since she can recognize herself (unless she recognizes her own description) so it might be possible to modify it to give coordinates.

[unkempt spear]

Small children would disagree

[daring fjord]

Small children's brains work differently than an AI.