I don't see a Houdini&AI thread. Dump your thoughts here. I'm curious where it goes.

Maybe you can compare with the changes that happened in web development for the lst 20 years. In the early 2000’s developers had to learn how to hand code HTML, maybe hack some tables for layout, eventually learn what CSS is. With time templates arrived and now online services will literally ”generate” new sites from a few selections. Those who code manually are expected to know so much more about a vast array of frameworks.

In my speculation and guessing for computer graphics in a distant 10 year time (2034) I’d expect something similar, but also different and more complex. The basic things such as modeling, texturing and some generic rigging and animation will likely be done with templates and generative machine learning. Especially low budget productions will probably be done with these things and the small studio will be able to do more. Also things that are done manually today will probably be split into stages that allow a quicker start using templates and machine generated setups. Work will likely not be much easier though. This is because someone else can do those things you do as well. Outsourcing will probably not be done to low salary countries but rather to less paid machines.

For higher budget projects human intervention, probably genuine art skills will probably be worth the money and time. Those who code will much likely focus on overall planning and maybe rough algorithms, but low level coding is expected to be done fast, reliably and consistently by machines.

Those ”fact errors” currently present in chat systems will probably be solved or worked around somehow. Meanwhile I’d expect there to be some kind of movement in society that opposes the spreading of machine learning so that I’d expect ”glitches” and unexpected results coming out of those systems now and then.

These are just some quick and rough hypotheses. I just think history repeats itself somehow, because humans are the way they are, and then there is always a twist.

Please do not take this as something serious or a prediction in some sort. It is just for fun and casual.

This is the kind of thing I'd like to have built-in, a framework for machine learning. You wouldn't need to know the math behind it, it would come with big building blocks like In PyTorch and other frameworks, but exposed as Houdini nodes.

But maybe this is too far off of what Houdini is, and training your own networks might just not be much of a thing anymore because you can't keep up with the development in this area anyway. It might be more important to have the ability to somehow run existing models instead of building them yourself.

The sooner the better I guess.

With very primitive 3D AI already popping up here and there, it looks like suddenly AI is moving very fast towards 3D as well. I really hope we’ll soon be able to make use of this new tech in Houdini.

In combination with "Gaussian Splatting", Real Time Radiance Field rendering and Houdini beeing the best environment for particle sim and managing point clouds, combining this tech with AI in Houdini would create so many new and exiting possibilities.

Other than that, I think it would be great if AI could be used to enhance the workflow and GUI. It could filter Operators based on the selection and be smart enough to suggest Ops named differently then the search term, especially with the silly naming of some Ops in Houdini. But currently Houdini is giving me bopkus when using the LMB over an Edge, Poly or Point.

And for mere mortals like myself that have a blind spot for scripting and coding, it would be a big relief if it would be possible to use prompts instead of having to write Vex or Python code for simple tasks.

Personally, I think SideFX should concentrate, at least for the next couple of releases, on CFX tools, like KineFX!

AI, in my opinion, would only be useful for those who wish for a do-it-all magic button, while most of us, at least IMHO, prefer to do the hard work, the future of AI in VFX IMHO resides in animation, and there are already tools for that, like Cascadeur and Audio2face!

Something like this for Houdini would be interesting: https://youtu.be/LNtrJVlHObs?si=GjoP4ZQ6yybdJ8J3 [youtu.be]

toonafish
The sooner the better I guess.

Yes. A.I. is racing forward. Any call to slow down Houdini in this area is far beyong naive. It´s like when the internet came up and you ask Amazon to not sell stuff over the internet for a few years so normal retailers don`t have to catch up that fast.
But hey, on the other hand - there are people out there who still think a.i. can`t do humans hands correctly.

Sygnum

toonafish
The sooner the better I guess.

But hey, on the other hand - there are people out there who still think a.i. can`t do humans hands correctly.

… as usual human intelligence is highly overrated, how many people out there can draw hands correctly, and then in less the 60 sec. ;-)

I think it is a misconception to compare machines with humans. It is somehow equivalent of saying a rocket (made my human) can fly faster than man can.

You should not even compare human children’s thinking with adult minds (refering to Piaget for example).

First, ML is only a system that is a human creation, by a creative being that in turn originally was designed by a creator (it is my point of view).

However, machines are great for doing things that we as humans find booring and simply time consuming.

For example when you want to create some code that solves a problem you likely need to search for information. This a machine can support in doing. It has long been done by search engines and now the next level is here.

Trying to make or use generic ML/AI is probably not really motivated in the field of CG. However, as we know those things like rotoscoping, masking and set extensions are really time saving features.

However, the concept of ML/AI has brought a new phenomenon to the world and that is the abstraction layer. What this mean is that it would be possible for different AI/ML to work together. This is a scary part because with enough development this can very much threaten many job opportunities.

Cheers and peace!

I think that from marketing point of view Houdini 20 is like anniversary and its perfect moment for rebranding. First of all like in advertising campaign look is imortant. So new users can see that something new is coming. New look means new refreshed UI. So theres clear message that it changes (for artist it could mean its now easier to learn so they will give it a try) New look should be like upgrading from Windows 95 to Apple esthetics.

Second important message should be AI. So everybody knows that Houdini is really the future and savior for whole industy. Its not any marketing slogan but it is actually true. Thanks to Houdini modular approach it is the best choice for AI. So anything related to AI should be included in Houdini 20 from marketing point of view. Even stable diffusion nodes.

As for AI in Houdini in near future i see it as tool for working with many AI models. By this i mean using and connecting existing models, training my own models in whole spectrum of possibilities and using generative models.

First there should be something like CopyCat from Nuke. The philosophy is to train own 3D AI models inside Houdini without using Tensorflow on Linux console or other not artist friendly shit

with such AI node, VFX studios could use their existing 3D data and train AI models to speedup their workflows. Artist could use it for experimenting and making artisting things that nobody care about TDs could use their knowlage, make models and sell them on Orbot or other marketplace. So yeah it would be cool for start.

Then there could be integrading of existing models fe. from Nvidia for hairgrooming in realtime all sorts of neural simulation, neural rendering etc. when it will mature. So artist could combine different models and make tools as they like. I see Houdini more as construct and combine tool for AI.

So artist could use generative 3D AI model for geometry and make sculpt tool that finishes details and assist with sculpting. Then it quicly automatic make UVs and texture it using generative 2D, setup materials etc.. hmm damn... but it could be fun after all

Last thing is to easilly distribute such AI training on renderfarm or better on cloud.

So this is basicly how i see it and think about AI in Houdini 20.

PS. more artist, animators will use Houdini the better it will become for artists. So all hail to Houdini

First and foremost, I feel I need an AI MENTAT / TEACHER that is highly integrated and embedded in every aspect of the Houdini ecosystem, and super finely tuned with its documentation, its libraries of code and general algorithms for 3D.
This AI should assist me in creating anything within Houdini, understand what I want to do, and provide clear explanations for nodes, functions and workflows, and produce examples on the fly, all without hallucinating.

Developing such an AI helper for Houdini would be a WIN-WIN situation for SideFX, given the extremely intricate nature of the app.
In fact it would be beneficial for both making Houdini more accessible for newcomers (so more customers would be attracted) and also enabling already experienced users to advance their skills in short time and create more complex HDAs or scenes.

Digipiction
This is the kind of thing I'd like to have built-in, a framework for machine learning. You wouldn't need to know the math behind it, it would come with big building blocks like In PyTorch and other frameworks, but exposed as Houdini nodes.

But maybe this is too far off of what Houdini is, and training your own networks might just not be much of a thing anymore because you can't keep up with the development in this area anyway. It might be more important to have the ability to somehow run existing models instead of building them yourself.

This is a fascinating idea.

Do you guys see a place for a Houdini version of Nuke's CopyCat? In 2D comp work, it makes a ton of sense. I'm trying to imagine how something like that tool might translate to 3D.

I looked up Onnx, and it seems solid, there are free open source pre-trained models for it, there are even online tutorials on using it to create tools, I think that this is a very nice addition to Houdini!

https://onnx.ai [onnx.ai]

AnsonSavage
Do you guys see a place for a Houdini version of Nuke's CopyCat?

Isn't it what OpenFX and BorisFX Sapphire suite brings to Houdini, a Nuke copycat? Just wondering here!