Can AI Teach SideFX Houdini? | Exploring AI in 3D Software Learning

Introduction

Hi everyone, this is Roland from graphandinmotion.com. Have you ever wondered whether modern AI can teach us professional 3D software like Houdini? In this article, we explore exactly that. There's been a lot of buzz in the 3D artist community over the last few years about whether AI will take over our jobs. As a 3D artist and educator, I really hope that this will never be the case because I love my job, sharing knowledge with the community, and creating tutorials.

To test this, we conducted a little experiment using ChatGPT, the latest version of OpenAI's language model, to guide us through two different effect setups in Houdini.

Before diving into the experiment, I would like to ask for a small favor. I have been running this YouTube channel for 13 years and have not yet reached 100,000 subscribers. If you like my content, please consider subscribing to my YouTube channel. For further support, check out my Patreon, where I provide lots of exclusive and premium content about Houdini.

Now, let's see how well AI can teach us. I used ChatGPT-4, and provided it with a specific prompt: "Act as a trainer for SideFX Houdini. I want to learn Houdini and I am a beginner. Write a step-by-step guide on how to create the following effect: cover a 3D object with many spheres, scale and colorize the spheres using noise, and animate the noise to change the effect over time. I don't need any camera lights and rendering, just focus on the effect."

ChatGPT generated a structured step-by-step guide, including code snippets and explanations. We followed these steps in Houdini:

1. Setting Up a 3D Object:

   • Create a geometry node.
   • Import or create a 3D object, for example, a sphere.

2. Cover the Object with Spheres:

   • Append a scatter node to the sphere node to create points on the object.
   • Adjust scatter node parameters to control the number of points.

3. Copy Sphere to Points:

   • Create a small sphere.
   • Use a 'copy to points' node to copy the small sphere to each scattered point.

4. Scale and Colorize Using Noise:

   • Add an attribute wrangle to create noise-based scaling.
   • Add another attribute wrangle to create noise-based coloring.

5. Adjustments and Viewing:

   • Fine-tune the parameters and preview the animation.

The simple setup worked perfectly, but the second effect—a balloon inflation—proved more challenging. The prompt for this was: "Create a balloon inflation using vellum. The balloon should fall like an empty balloon, inflate over time, and rise."

ChatGPT’s initial instructions resulted in issues because vellum constraints couldn’t be animated directly.

After pointing out mistakes, ChatGPT provided a different method involving a Point Wrangle to manipulate pressure attributes, but even then, errors prevailed. Despite multiple corrections, the AI's guidance was ultimately insufficient for more complex setups.

In the end, manually tweaking and adding some custom nodes to Houdini solved the problem. We animated the rest length scale using a Vellum Constraint Properties node and then added appropriate forces to simulate inflation and rising.

The outcome? AI is not yet ready to replace 3D artists and tutorial creators, especially for more complex tasks.

Keywords

• 3D Software
• Houdini
• ChatGPT
• AI in Art
• Vellum Constraints
• Attribute Wrangle
• Balloon Inflation
• Animation
• 3D Effects

FAQ

Q1: What was the purpose of this experiment? A1: The purpose was to explore whether AI, specifically ChatGPT-4, can guide users through learning and setting up complex 3D effects in Houdini.

Q2: How did ChatGPT handle the simple setup? A2: ChatGPT handled the simple setup of covering a 3D object with spheres, scaling, and colorizing them with noise quite well.

Q3: What issues arose with the more complex setup? A3: ChatGPT struggled to animate vellum constraints over time and often provided incorrect or outdated methods, leading to failed simulations.

Q4: How were the issues resolved? A4: The issues were resolved by manually adding and correcting nodes, specifically using a Vellum Constraint Properties node to animate the rest length scale and adding appropriate forces.

Q5: Is AI ready to replace 3D artists? A5: No, the experiment demonstrated that AI still has a long way to go before it can handle complex 3D setups and replace 3D artists and educators.

If you want to explore more detailed tutorials and continue to learn about Houdini, please consider subscribing to my YouTube channel or supporting me on Patreon. Thank you for your time!