Have you ever spent all day with ChatGPT?  Have you ever spent all day with ChatGPT and Gemini while hyper-focused on getting them both to code and create looping noise shaders in Open Shading Language?

This has been quite a learning day, from noticing the personality quirks in A.I. to learning about coding shaders to kind of starting to understand why looping noise is so hard to achieve through code.  The animation is now using two OSL noise shaders that ChatGPT and Gemini each coded.  Specifically, the noise shader that ChatGPT created is inside the noise shader that Gemini created.

Gemini

ChatGPT

Like I thought, the noise that ChatGPT created just ping-pongs back and forth which is a cheap ass way to make a loop.  But the math behind the scenes is doing the best it can to make that transition not so apparent, so still a good job.

Then Gemini after much finagling and tweaking finally explained to me that getting a loop the way ChatGPT was doing it was always going to result in a ping-pong effect, it’s just inherent to how the math generates the noise.  Gemini then went a step further and suggested another method to get looping noise could be by using cylindrical UVW mapping and then animating the coordinates around a circle.  Um.  Ok, I said.  Do it.  It did.  It worked.  It’s really simple, it just moves the generated noise texture around in a circular pattern and on its own it’s not very impressive, but when you start combining motions things can get interesting quickly.

And again Gemini went a step further and offered more unsolicited advice on how to break up the admitted monotony of its circular animation method.  It started suggesting to add distortions to the noise.  Um.  Ok, I said.  Do it.  It did.  It worked.

The code has tripled in size from where I left off yesterday and there’s been so much information overload and tweaking to the shaders today that I’m not even sure what exactly they’re doing at this point.  But they’re both working and they’ll be incredibly valuable tools to have.

So this is pretty neat.  

I wanted to add clouds to this city scene using a static VDB file.  Ok, no big deal but then I wanted to use an animated procedurally generated noise texture to give the cloud a little pizazz. 

That’s the problem.  

3ds Max has never had a simple, nonconvoluted way to animate any types of procedural noise seamlessly over a given amount of time/frames.  I’m not sure any of the big 3D packages can do it and I’ve never been sure why.

Almost 25 years ago there was a Max plugin which added a Loopable Noise Controller for some things, but that’s long gone and the last version out there is for 3ds Max r6 from 2001.  So, as I’m sitting there playing with the clouds and wishing I didn’t have to load up After Effects to generate an entire 300 frame image sequence of a seamlessly looping animated noise texture, it occurs to me…get ChatGPT to write a noise shader that loops. 

Autodesk maintains a Github of OSL shaders which is where I grabbed the noise.osl file.  I opened that file up in a text editor and copied the contents into a ChatGPT window.  I then asked ChatGPT to modify it by adding the ability to, and UI options for, seamlessly looping the noise over time.

It fucking works. 

The shader code that ChatGPT spat out freaking works.  I loaded up the OSL in Max as normal and there it is.  Everything after Amplitude is all ChatGPT’s code:

Here are ChatGPT’s instructions on how to use the shader:

Results for a 300 frame loop:

See, I’ve been wanting to be able to do this, like this, for a good long while but do not have the math or coding skills required to create it.

The yellow box object has an Arnold Properites modifier in the stack which uses our new noiselooper.osl to drive a displacement map.  The black and white inset animation is the actual map the noiselooper created.

I looked at the code a little but haven’t started hacking away at it to see what it does yet.  My concern is that the noise just kind of ping-pong’s back and forth to make a loop which could become noticeably distracting. 

All in all though, this is excellent alternative to manually setting up and using blending techniques between two animated noise maps in Max or using After Effects to render out an image sequence.

Good going, ChatGPT.  If I look at the code closer and it seems to be doing more than just ping-pong’ing and/or if there are any improvements to it, I’ll post the .osl here.  ChatGPT said it could achieve the noise loops a few different ways without relying on this exact same method.  I dunno, I’m already flipping shocked this worked, so who knows.

I’ve always loved these image comparison slider things.  There’s the 3ds Max RAM Player, for previewing images and sequences, that offers A/B image comparison and is what reminded me to try it here.  The UI is a little clunky, but it is free.