AI coding assistants for arch viz scripting in 3ds Max and Blender

The repetitive work nobody budgets for

Every arch viz project contains hours of tedious scene work: renaming imported Revit objects, assigning materials by layer, replacing proxy trees, setting up twenty camera views with matching render settings, exporting passes with consistent file names. These jobs are rarely in the fee and are exactly what scripts are for.

Many artists never learned MAXScript or Python because the learning curve did not pay back on a single project. General-purpose language models, such as ChatGPT, Claude and Gemini, and editor-based assistants such as GitHub Copilot change that calculation. They can draft working pymxs scripts for 3ds Max or bpy scripts for Blender from a plain description.

They are also confidently wrong often enough that blind use is risky. A script that runs over every object in a 600 MB scene can do a lot of damage in one second.

Writing good requests

Good requests are specific about the software version, the language and the scene structure. Ask for a Python script for 3ds Max 2025 using pymxs, or for Blender 4.2 using bpy, and describe exactly what the scene contains. Something like all window frames are editable poly objects whose names start with WIN_, assign them the material named Alu_Dark_Bronze, skip any object on the layer Existing gives the model what it needs.

Ask for the script to report what it would change before it changes anything. A dry-run mode that prints object names and proposed actions is easy for the model to add and lets you check the selection logic on the real scene.

Paste error messages back verbatim. API names change between versions, and models frequently suggest functions that were renamed or never existed. The traceback is the fastest way to get a correction.

Testing before trusting

Always test on a copy of the scene. Save incrementally before running anything, and run first on a small selection. Check the undo stack: some operations in both 3ds Max and Blender are not undoable, especially file operations and certain modifier changes.

Read the script before you run it, even if you are not fluent. Look for loops over all objects, file deletion, saving or overwriting paths and anything touching network drives. If a line does something you did not ask for, ask the assistant to explain it.

Keep client data out of prompts. Describe the scene structure generically instead of pasting proprietary object lists, project names or file paths. Check whether your assistant's settings allow conversations to be used for training, and use an enterprise or no-training option for studio work.

Building a studio script library

The real value builds over time. Each working script, once tested, goes into a shared studio library with a short description, the software version it was tested on and an example of its use. Within a few months a small studio can have thirty reliable tools: batch camera export, material reassignment, proxy swap, naming cleanup, render pass setup.

That library is where AI assistance pays back, not in the single script. The assistant writes the first draft; the artist tests, documents and keeps it. The result is a pipeline that does less clicking and more rendering.