Hackster is hosting Hackster Holidays, Ep. 6: Livestream & Giveaway Drawing. Watch previous episodes or stream live on Monday!Stream Hackster Holidays, Ep. 6 on Monday!
Kenneth Yang
Published © Apache-2.0

Blender AI Assistant

Create in 3D with plain English

BeginnerFull instructions provided30 minutes642

Things used in this project

Hardware components

AMD Radeon Pro W7900
This professional GPU has 48 GB of VRAM allowing for complex content creation while also supporting large LLMs such as Llama 3 70b locally for fast and private LLM usage.
×1

Software apps and online services

Blender 3D
The content creation tool for which the add-on in written for.
Ollama
Quickly setup and serve local LLMs.
Ubuntu 22.04.4 LTS
Linux OS distribution that supports ROCm.

Story

Read more

Code

AI Assistant add-on

Python
Blender add-on that enables the AI assistant feature.
bl_info = {
    "name": "AI Assistant",
    "author": "Kenneth Y (Microbob)",
    "blender": (2, 80, 0),
    "version": (0, 0, 1),
    "category": "3D View",
}

import bpy
import requests
import threading
import traceback
import functools


result = ""


class ASSISTANT_PT_Panel(bpy.types.Panel):
    bl_label = "AI Assistant"
    bl_idname = "ASSISTANT_PT_panel"
    bl_space_type = "VIEW_3D"
    bl_region_type = "UI"
    bl_category = "AI Assistant"

    def draw(self, context):
        layout = self.layout
        scene = context.scene
        assistant_props = scene.assistant_props

        layout.prop(assistant_props, "input_text")
        layout.operator(
            "assistant.submit",
            text="Generating" if assistant_props.is_generating else "Submit",
            emboss=not assistant_props.is_generating,
        )
        layout.label(text="Response:")
        for line in assistant_props.output_text.split("\n"):
            layout.label(text=line)


def send_request(prompt, system, flag, recurse_count=0):
    url = "http://localhost:11434/api/generate"
    headers = {"Content-Type": "application/json"}
    data = {
        "model": "llama3:70b",
        "prompt": prompt,
        "stream": False,
        "system": system,
    }

    global result

    try:
        response = requests.post(url, headers=headers, json=data)
    except requests.RequestException as e:
        result = f"Request failed: {e}"
    else:
        if response.status_code == 200:
            result = response.json().get("response", "NO RESULT")

            if result == "NO RESULT":
                flag.set()
                raise ValueError("No result field found in API call.")
        else:
            result = f"Error {response.status_code}: {response.text}"
            flag.set()
            return

        # Extract code fragment.
        try:
            start_of_fragment_index = result.index("```")
        except ValueError:
            # Show response if no code.
            flag.set()
        else:
            end_of_fragment_index = (
                result[start_of_fragment_index + 3 :].index("```")
                + start_of_fragment_index
                + 3
            )
            code = result[start_of_fragment_index + 3 : end_of_fragment_index]
            result = code

            # Try to run the command
            try:
                exec(code)
            except Exception:
                # Get the error and request a fix.
                error = traceback.format_exc()

                # Stop trying after a few times (avoid infinite recursion).
                if recurse_count > 3:
                    result = error
                    flag.set()
                else:
                    # Recurse with new prompt and setup.
                    send_request(
                        f"Command: {prompt}\nError: {error}",
                        "You are a programming assistant for Blender 3D's Python API. I will give you a Blender Python command that didn't work and the associated error it produced. Fix the command with the correct Blender Python code and place it between three tick marks like this '```'. Do not explain your answer. No need to import bpy. If a command is not possible, respond with `not possible` and a one sentence description of why. If my question is not related to Blender, respond with `not possible` and a one sentence description of why.",
                        flag,
                        recurse_count + 1
                    )
            else:
                # On successful exec.
                flag.set()


class ASSISTANT_OT_Submit(bpy.types.Operator):
    bl_idname = "assistant.submit"
    bl_label = "Submit"

    def execute(self, context):
        scene = context.scene
        assistant_props = scene.assistant_props

        # Skip if already generating.
        if assistant_props.is_generating:
            return {"FINISHED"}

        assistant_props.is_generating = True

        # Generate object-location dictionary.
        scene_info = {}
        for object in scene.objects:
            scene_info[object.name] = object.location

        # Run the request in a separate thread.
        thread_flag = threading.Event()
        thread = threading.Thread(
            target=send_request,
            args=(
                assistant_props.input_text,
                "You are a programming assistant for Blender 3D's Python API. I will ask you to perform actions in Blender and you will respond with the corresponding Blender Python commands surrounded by three tick marks like this '```'. Do not explain your answer. No need to import bpy. If a command is not possible, respond with `not possible` and a one sentence description of why. If my question is not related to Blender, respond with `not possible` and a one sentence description of why. Here is a list of objects in the scene and their locations: {scene_info}",
                thread_flag,
            ),
        )
        thread.start()

        check_thread = functools.partial(self.update_scene, context, thread_flag)
        bpy.app.timers.register(check_thread)

        return {"FINISHED"}

    def update_scene(self, context, flag):
        if flag.is_set():
            global result

            context.scene.assistant_props.is_generating = False
            context.scene.assistant_props.output_text = result

            # Reset after using it.
            result = ""

            return None  # Return None to indicate the callback should not be repeated
        return 0.1


class ASSISTANT_Props(bpy.types.PropertyGroup):
    input_text: bpy.props.StringProperty(
        name="Input Text", description="Describe what you'd like to do", default=""
    )
    output_text: bpy.props.StringProperty(
        name="Output Text", description="Response from the assistant", default=""
    )
    is_generating: bpy.props.BoolProperty(
        name="Is Generating",
        description="Indicates whether the assistant is generating a response",
        default=False,
    )


def register():
    bpy.utils.register_class(ASSISTANT_PT_Panel)
    bpy.utils.register_class(ASSISTANT_OT_Submit)
    bpy.utils.register_class(ASSISTANT_Props)
    bpy.types.Scene.assistant_props = bpy.props.PointerProperty(type=ASSISTANT_Props)


def unregister():
    bpy.utils.unregister_class(ASSISTANT_PT_Panel)
    bpy.utils.unregister_class(ASSISTANT_OT_Submit)
    bpy.utils.unregister_class(ASSISTANT_Props)
    del bpy.types.Scene.assistant_props


if __name__ == "__main__":
    register()

Blender AI Assistant

GitHub repo where this project is kept

Credits

Kenneth Yang

Kenneth Yang

10 projects • 100 followers
Maker, developer, 3D content creator

Comments