entagma

Advanced Setups: Using Local LLMs Inside Houdini

We’re still in the process of building new Premium courses. In the meantime Chris is here to share a somewhat silly but also fun and interesting setup he built a couple of weeks ago: In this setup we’ll run a small large language model on our own hardware and control it from within Houdini to come up with some town names and generate some color ramps for us. Nothing that will change the world, but a fun showcase for some pretty new tech in the world of machine learning.

Advanced Setups 30 – PCA Magic 2: Dejitter animations

Over the past months we finished our first machine learning project here on Entagma. And in episode 9 we talked about a very special new node in Houdini: The PCA node! This node is extremely handy for feeding 3d data to a neural net, like we did in that course, but that’s far from everything that this node can do! So over the next two videos we want to show you two additional tricks you can do with PCA: Aligning objects and smoothing animations.

Advanced Setups 30 – PCA Magic 1: Aligning Objects

Over the past months we finished our first machine learning project here on Entagma. And in episode 9 we talked about a very special new node in Houdini: The PCA node! This node is extremely handy for feeding 3d data to a neural net, like we did in that course, but that’s far from everything that this node can do! So over the next two videos we want to show you two additional tricks you can do with PCA: Aligning objects and smoothing animations.

ML101 – pt. 13: Generating Poses With Vellum And PDG

Hello and welcome to the second addendum and the last episode of our Machine Learning Deformer Premium Course. After this episode and in the upcoming weeks, we’ll pay visits to some other courses on here to cover smaller topics and ideas. And in the near future, we’ll be back with a new Premium Course. We have a couple of course candidates in preparation, now it’s just on us to decide, what we’ll release next. More notes on that soon. In this video we’ll replace the simple wrinkle d...

ML101 – pt. 12: Addendum A: Stacking Additional Models

This week on ML101 we‘re not quite finished with our Muscle deformer yet! We still have some loose threads left and in this episode we want to take a look at how we can train multiple parts of our character in separate nets and combine all of them in one inference setup.

ML101 – pt. 11: Building the Inference Setup

In this Episode of Ml101 we will bring our trained Model back into Houdini and see the results of all the work we did for the last 10 Episodes! We will combine all the things we learned to create a Inference Setup.

ML101 – pt. 10: Reducing our Data – Encoding, Exporting and Training our Data-Set

In Episode 10 we are finally able to train our Neural Network with the date from our Capybara! But first we need to prep the data and change some inputs in our Python Script. Then we can train for the first time and have a look at how well our Data-Set is performing. In the end of this Episode we will have an onnx- File for our Capybara that we will use in Houdini in the next Episode!

ML101 – pt. 09: Reducing our Data – the PCA Node

This week is all about Principal Component Analysis! Chris will show you what PCA does and how it works in Houdini. He will go over this very abstract Node with you in a Demo Scene and visualize all the Steps that will reduce our Dataset a whole lot to a more manageable Size without any noticeable Quality Loss.

ML101 – pt. 08: Pose Prep 01 – Extracting Differences

In Part 8 of ML101, Chris will show you how to make the training data more digestable for our Neural Network. He will have some Tips for you along the way, since there are some pitfalls, that can still cost you a lot of training time, if you don’t take them into account. In the End we will have a nice, clean setup with our Training Data, that is ready to be processed further in our Neural Network.

ML101 – pt. 07: Section Overview and Pose Generation

This week we are finally back in Houdini and we will go over the upcoming Steps to get a Neural Net that actually works as a Muscle Deformer. We will prep some Poses of our Capybara for training and we will add a Wrinkle Deformer to get more realistic looking neck wrinkles!

ML101 – pt. 06: Finishing our Neural Net

Finally we are finishing our Neural Net! We will finish our python code and train a few different models at once. We will talk about how to train the best model out of the five even further, so we can use that model in Houdini for our muscle deformer in the next episode!

ML101 – pt. 05: Adding Tensorboard

In the last episode we completed our pipeline from Houdini to PyTorch and back to Houdini. And we created and trained our first neural net, which didn’t exactly solve our initial problem. So how do we get from some neural net to a good one? That’s what we’ll explore in the next two episode. In this one, we’ll use Tensorboard to get a much better judgement over how well our model is training and in the next episode we’ll modify our code to build and train multiple networks so we can ...

ML101 – pt. 04: From Python back to Houdini

In this episode we’ll complete two major steps for our project: We’ll finish our pipeline from Houdini to Python and back to Houdini and we’ll train our very first neural network!

Commands to install onnx in the Anaconda Prompt (make sure you activate your Python Environment first):
pip install onnx
pip install onnxscript

To stop training early hit “Ctrl-C” in the VS Code Console

New on Entagma Patreon: Course Collections

Dear Supporters,

for your convenience, all our videos are now sorted into collections that resemble the premium courses. Find the collections here:

https://www.patreon.com/entagma/collections

Thank you, Manuel

AHTYA 2.0 – pt.34: Course Completed!

And we’re done! We’ve covered everything we think is necessary to get you started in Houdini (and maybe a bit more) and it’s time to move on to new adventures and new courses!
As is tradition this marks the end of our biweekly schedule of new AHTYA videos. Also with this video we’re taking a small break from our general Premium Content upload schedule and will soon be back with an entirely new course for you to enjoy. Exciting times ahead!

AHTYA 2.0 – pt.33: SOPs vs DOPs

We finished the last big simulation project in the AHYTA Simkit and are now on the home stretch of this course. In this video Chris has some closing thoughts on using either SOPs or DOPs for simulations in Houdini.

Talks mentioned in this video:
Microsolvers and DOPs Data by Fuat Yüksel
Stacking Experiments Up: From A Sphere to Animation by Rich L...

AHTYA 2.0 – pt.32: Smoke Stick Part 3 – Rendering in Karma

This last part of our Smoke Stick Project is all about rendering our setup with Karma. Volume rendering comes with its on peculiarities, so we will have a look at those in this AHTYA episode.

AHTYA 2.0 – pt.31: Smoke Stick Part 2 – Adding Details

In this second part of the Pyro solver project we are going over how to get an even more high res looking sim without compromising on simulation speed. Also as always we are adding particles to make the simulation look more interesting.

AHTYA 2.0 – pt.30: Smoke Stick Part 1 – Pyro Setup

It’s time to start our final project in this AHTYA series – which will be a Pyro Solver project. In this first part, we’ll set up the base sim and talk about what makes this particular solver quite different to the ones we covered before.

AHTYA 2.0 – pt.29: Microsplash Part 2 – Adding Bubbles

It’s time to finish our FLIP fluids setup! And, as always, this involves adding lot’s of tiny details – in this case lots of bubbles! This time round, we won’t use POPs though and build our bubbles directly from our FLIP fluid particles.

AHTYA 2.0 – pt.28: Microsplash Part 1 – Flip Fluid Setup

In part 28 of our AHTYA 2.0 series, we are creating a microfluid simulation and taking a look at Flip Fluids in Houdini. In part 1 of this project, we will setup the splashing effect and talk about the parameters that you can tweak to replicate any micro fluid behavior, that you want.

AHTYA 2.0 – pt.27: Vase Smash Part 2 – Adding Particles

This is part 2 of our Vase Smash Project. In this video we are going to add particles to our existing setup to tie this animation together and get the details to make this look even more realistic. We take a look at two different ways of spawning particles from an rbd sim and adding further details like angular velocity.

AHTYA 2.0 – pt.26: Vase Smash Part 1 – RBD Setup

It's time for the next project in our AHTYA Sim Kit! In this episode we are creating a vase smashing simulation using the RBD Bullet Solver. We will go over what goals we want to achieve to make our sim realistic and interesting and what parameters we need to tweak to get to our desired result.

Reposting missing AHTYA episodes on Patreon

Dear supporters,

as you know you have two alternatives to watch our premium content. You can access our premium content directly here on patreon or you can choose to use our website https://entagma.com. You can login over there easily using your patreon credentials. The website may make finding content and watching videos easier and more convenient.

Unfortunately, due to a technical problem, some of the AHTYA videos were...

ML101 – pt. 03: Transferring training data from Houdini to Python

In this episode we will do more Prep-work for our neural network. This workflow allows you to transfer attributes in a efficient and fast way from Houdini to make it readable in Python. And we will write a Python File together that can be a starting point for more complex setups in the future.

ML101 – pt. 02: The Python Install Episode

In the next episode we’re finally jumping into Python to build our neural net with PyTorch. Unfortunately this is not something we can do within Houdini, so it’s time for another “How to install Python” episode!

ML101 - pt. 01: DIY Neural Net & Regression Basics

In this episode we link the two main concepts we need to build our muscle deformer: Training a neural net and fitting a function to a dataset. Along the way we build our own simple neural net in Houdini with Vex.

ML101 – pt. 00: Introduction

This course is about making Machine Learning the next tool in your Houdini toolbox. Just like some problems are best solved with vector maths or Vex or Python, some problems get way easier with ML and we want to be able to spot these problems and solve them using your own custom networks.

To do so we create our own training data in Houdini, we build and train our own neural nets with Python and PyTorch and finally integrate them into our Houdini Workflows using ONNX. Additionally we wan...

AHTYA 2.0 – pt. 25: Dissolving Pill Part 2

This is second part of our dissolving pill project. In this video we are adding small air bubbles to our pill and taking a deeper look into the DOP Solver and what makes the workflow with DOPs quite different compared to other solvers in Houdini.

AHTYA 2.0 – pt. 24: Dissolving Pill Part 1

In this video we will build the first part of our dissolving pill project you’ve seen in the teaser. We will take a look at the SOP Solver in a simple example to get a feel for the workflow. Then we put that knowledge to use on our pill setup to create a realistic dissolving effect.