3D Shape Generator

-

In my second experiment, I wanted to work with curves and save old coordinates so that I could reuse them for similar curve shapes.

3D Shape AI Node


 


1. Create a Python node named "shape_node_GPT," adding the below code in the content area. 
import openai
import hou

# Current Python SOP node
node = hou.pwd()

# Geometry container
geo = hou.node("/obj/geo2")
if not geo:
    geo = hou.node("/obj").createNode("geo", "geo2")

# Set OpenAI API key from houdini.env
openai.api_key = hou.getenv("OPENAI_API_KEY")

# User prompt
user_prompt = node.parm("prompt").eval()

# System message
shape_message_node = hou.node("/obj/geo2/shape_message_node")
if shape_message_node:
    system_message = shape_message_node.parm("message").eval()
else:
    hou.ui.displayMessage("Error: shape_message_node not found!", severity=hou.severityType.Error)
    system_message = ""

# OpenAI call
def get_completion(user_prompt, system_message):
    try:
        messages = [
            {"role": "system", "content": system_message},
            {"role": "user", "content": user_prompt}
        ]
        response = openai.ChatCompletion.create(
            model="gpt-4-turbo-preview",
            messages=messages,
            temperature=1.0,
            max_tokens=500
        )
        return response["choices"][0]["message"]["content"].strip()
    except Exception as e:
        return str(e)

# Hide previously generated AI nodes safely (don't delete while cooking)
for child in geo.children():
    if "ai_" in child.name().lower():
        try:
            child.setDisplayFlag(False)
            child.setRenderFlag(False)
        except:
            pass

# Get AI result
response = get_completion(user_prompt, system_message)
lines = response.split("\n")

created_nodes = {}

# Create shapes and apply transforms
for line in lines:
    parts = line.split()

    # CREATE
    if "create" in line.lower():
        shape_type = parts[1].lower()
        params = {parts[i]: float(parts[i + 1]) for i in range(3, len(parts), 2)}
        shape_node = None

        if shape_type == "sphere":
            shape_node = geo.createNode("sphere", f"ai_sphere_{len(created_nodes)}")
            shape_node.parm("radx").set(params.get("radius", 1.0))
            shape_node.parm("rady").set(params.get("radius", 1.0))
            shape_node.parm("radz").set(params.get("radius", 1.0))

        elif shape_type == "box":
            shape_node = geo.createNode("box", f"ai_box_{len(created_nodes)}")
            shape_node.parm("sizex").set(params.get("size_x", 1.0))
            shape_node.parm("sizey").set(params.get("size_y", 1.0))
            shape_node.parm("sizez").set(params.get("size_z", 1.0))

        elif shape_type == "cylinder":
            shape_node = geo.createNode("tube", f"ai_cylinder_{len(created_nodes)}")
            shape_node.parm("height").set(params.get("height", 2.0))
            shape_node.parm("rad").set(params.get("radius", 0.5))

        elif shape_type == "torus":
            shape_node = geo.createNode("torus", f"ai_torus_{len(created_nodes)}")
            shape_node.parm("rad1").set(params.get("major_radius", 1.0))
            shape_node.parm("rad2").set(params.get("minor_radius", 0.3))

        elif shape_type == "cone":
            shape_node = geo.createNode("tube", f"ai_cone_{len(created_nodes)}")
            shape_node.parm("height").set(params.get("height", 2.0))
            shape_node.parm("rad1").set(params.get("radius_top", 0.5))
            shape_node.parm("rad2").set(params.get("radius_bottom", 1.0))

        if shape_node:
            created_nodes[f"{shape_type}_{len(created_nodes)}"] = shape_node

    # APPLY
    elif "apply" in line.lower():
        transform_type = parts[1].lower()
        params = {parts[i]: float(parts[i + 1]) for i in range(3, len(parts), 2)}
        last_node = list(created_nodes.values())[-1] if created_nodes else None

        if last_node:
            xform = geo.createNode("xform", f"ai_{transform_type}_{last_node.name()}")
            xform.setInput(0, last_node)

            if transform_type == "translate":
                xform.parm("tx").set(params.get("x", 0.0))
                xform.parm("ty").set(params.get("y", 0.0))
                xform.parm("tz").set(params.get("z", 0.0))

            elif transform_type == "rotate":
                xform.parm("rx").set(params.get("axis_x", 0.0) * params.get("degrees", 0.0))
                xform.parm("ry").set(params.get("axis_y", 0.0) * params.get("degrees", 0.0))
                xform.parm("rz").set(params.get("axis_z", 0.0) * params.get("degrees", 0.0))

            created_nodes[f"{transform_type}_{len(created_nodes)}"] = xform

# Merge all generated nodes into one final output
if created_nodes:
    merge_node = geo.createNode("merge", "ai_merge_all")
    for i, node in enumerate(created_nodes.values()):
        merge_node.setInput(i, node)

    out_node = geo.node("OUT")
    if not out_node:
        out_node = geo.createNode("null", "OUT")
        out_node.setDisplayFlag(True)
        out_node.setRenderFlag(True)

    out_node.setInput(0, merge_node)

# Auto-layout the generated network
geo.layoutChildren()

2. Also, create a Prompt parameter inside the Python node and a cook button with this script in it: "hou.node('/obj/geo2/shape_node_GPT').cook(force=True). "



3. Create a null node named "shape_message_node" and edit the parameter by clicking the gear icon at the top and selecting "Edit Parameter Interface" to create a string inside the node. 





Pastes the below prompt in the string:
You are an AI specializing in procedural 3D shape generation for Houdini.
Interpret the user's text prompt and generate structured Houdini node instructions.

Rules:
- Output only shape creation steps, no extra text.
- Supported shapes: Sphere, Box, Torus, Cylinder, Cone, Grid.
- If the user provides a size, adjust parameters accordingly.
- Format example: Create sphere with radius 2 Apply translate with x 1 y 2 z 0 Apply rotate with axis_x 0 axis_y 1 axis_z 0 degrees 45

DO NOT include explanations or comments, ONLY the structured shape instructions.

4. Create a Rules board because AI has limitations, and because of that, the words need to be precise at some points.

You are an AI specializing in procedural 3D shape generation for Houdini.
Interpret the user's text prompt and generate structured Houdini node instructions.

Rules:
- Output only shape creation steps, no extra text.
- Supported shapes: Sphere, Box, Torus, Cylinder, Cone, Grid.
- If the user provides a size, adjust parameters accordingly.
- Format example: Create sphere with radius 2 Apply translate with x 1 y 2 z 0 Apply rotate with axis_x 0 axis_y 1 axis_z 0 degrees 45

DO NOT include explanations or comments, ONLY the structured shape instructions.


Note: To get the "Cook" button next to the prompt area, mark "Horizontally Join to Next Parameter" in Prompt or any node earlier than the cook node.



Thank You, Houdini School, for explaining the foundations!
If you want to learn how to use Houdini, they are one of the best places to start with your VFX career.

Comments

Post a Comment

Popular posts from this blog

Present Best Way to Install MLOPs in Houdini

-
Image
I had an issue while installing MLOPs just following the tutorial, and to solve the problems, I went to different communities but got no solution, in the end, Bismuth's discord community, I got an answer from @.jaydrake who gave me the best steps for the current time to make MLOPs work. Thank you for saving my project. Below are the steps that currently work for setting up MLOPs, which are needed to use Open AI API, with the help of one of the users in the BV's discord server @.jaydrake. Setting up and installing MLOPs: 1. Checking your Python version with Houdini's Python Interpreter by going to the main options bar at the top where Windows -> Python Shell . 2. The choice of Python Interpreter depends upon the user if they will use an external Python environment or hython, which is Python in Houdini. 3. Then go to https://github.com/Bismuth-Consultancy-BV/MLOPs , and install GitHub Desktop, which is the best way to clone the repository inside your computers. This metho...
Read more

Color Ramp Generator

-
Image
I chose one of the most effortless node-based AI approaches for my first experiment: color ramp. It's similar to a gradient bar you see in Adobe Creative Suite, which in Houdini can be used to create a gradient effect over any line or shape made in Houdini; it's easy as it's a node that can be manipulated by parameter. By connecting Open AI API, I can make it a prompt-based approach without any problem where user text base input will be interpreted via a Python code and will give results to change the color ramps.  Using Houdini for this project gives me an advantage as it allows me to interact with the nodes already in the Houdini workspace using Python, also known as Python (Houdini + Python).  Color Ramp AI Node 1. Create a Python node named " color_ramp_GPT ," adding the below code in the content area.  from openai import OpenAI import hou import ast node = hou.pwd() client = OpenAI() def get_completion(user_prompt, system_message):     try:     ...
Read more

Researching topics for My Thesis (Master's Edition)

-
Image
Below are some of the ideas that popped up in my mind while thinking about what technology I can work on, which will help me learn and also contribute to the field of Digital Media. 1. Topics: Streamlining steps to procedurally generate landscapes in Houdini and Unreal Engine by testing tools with the help of AI! This will help artists work on the parameters to change the landscape, and AI can reduce time by building up the structure and connections required to build that landscape. Whenever I work on any project involving big maps, it makes me feel overwhelmed by how much time it takes to build something from scratch, and procedural generation is one of the solutions that made me feel how much we can improve it just by adding AI. If I could find a tool, the pipeline for developing landscapes would become more streamlined than ever for environmental designers. PCG is research where they use GAN to make landscapes via Height maps. Need to Experiment with AI and work on the concepts to ...
Read more