A 3D modeling system has a small BlenderGPT plugin that generates content based on tasks defined in natural language. The command input interface is designed as an additional tab called 'GPT-4 Assistant' in the 3D View sidebar, where arbitrary instructions (for example, 'create 100 cubes in random locations', 'take existing cubes and make them different sizes') can be entered and results received immediately. The code is written in Python, spans just over 300 lines, and is distributed under the MIT license.
The working principle involves sending a test request to the ChatGPT chatbot, which uses the GPT-4 model, through the OpenAI public API, appending the user-specified text with the note 'Can you please write Blender code for me that accomplishes the following task'. Then, the Python code is extracted from the response and executed as a script in Blender. An OpenAI API access key is required (specified in the plugin settings menu). The generated code can be monitored through the console (Window > Toggle System Console).
Forrás: opennet.ru
