Building a Latent Space in Houdini: Part 1 – Theory

This week we want to give you a lot more insight into what a latent space actually is and how it gets created. In this case not by using Stable Diffusion, but by building and training our own neural net inside Houdini, who’s only job is to create a very simple latent space for us to explore.

In this video we’ll go over the main concepts and the theory of our latent space. And on Thursday in the Advanced Setups Course on Patreon, we’ll build the entire setup using Houdini and Pytorch.

You can find all visualisations from this video to explore by yourself in the scene file. Make sure you have access to Pytorch and Numpy in your Houdini Python environment, for example by installing the MLOPs Addon.

Download MLOPs

Parts of this setup are inspired by this article by Eugenia Anello

[Billary]

This is so cool! Such an intuitive way to visualize what's going on

[felixblacke]

Wonderful, I can't wait to see how you all work AI into Houdini.

[vincentthomas]

Now we speak! 😊 Exactely the informations i was looking for right now, quality content, crystal clear, really remarkable presentstion, thanks you guys! 😍

[rashshawn]

The visuals are so awesome!!! Is there part 2?

[jinchaohan]

please keep going on this tutorial!!! this is so cool!

[JulianAbreu]

finally making models in houdini, so cool!

[PawelGrzelak]

Great explanation, thank you for this

[roybatty]

When will the next part be up on Patreon? Unforeseen delay? Really looking forward to this one!

[Create-The-Imaginable]

Wow! Is that 3D or is it 4D with Time?

[rendermanpro]

Imagine Midjourney visualization of words looking like a galaxy and you navigate it. Like a starship flying through the words galaxy and it create video of it imagination in realitime. In 20 years or sooner.

[DistructiveElements]

Can you actually create models in houdini?

[rendermanpro]

Or imagine these flat layers with added one dimension, and it not flat but in 3D