15:12Machine-translated from Azerbaijani by AIRead in Azerbaijani
LDraw Nova: AI Tool for Creating LEGO Models Unveiled
The open-source “LDraw Nova” tool is designed to use artificial intelligence models to create structures from thousands of real “LEGO” pieces. Models such as “GPT-6 Astra” and “Claude Opus 5.5” can generate detailed computer-aided design (CAD) files based on a text prompt. One of the demonstration examples is the 2175-piece “Sakura Garden” model. Developer Carlos Antelo released the first version on Friday after 3 previous attempts failed. However, none of these designs has yet been assembled with real bricks.
The process begins with the creation of a JSON plan describing the model’s structure and individual parts. The AI agent then writes a “Python” program. That program creates the main file in the “LDraw” format and lists each “LEGO” piece on a separate line. The agent can review the model’s image, correct errors and repeat the process until a satisfactory result is achieved. The completed files can be opened in “LDView”, “LeoCAD” and “Studio”. “LDraw Nova” is a web application that runs through “Docker”. It works with the “OpenAI”, “Anthropic” and “OpenRouter” services. The application’s gallery indicates which AI model and text prompt were used to create each structure. Examples include “Cathedral”, “Tidal Observatory”, the unfinished “Atlas Crane” and the “Copper Bean” residential buildings. The “Astra” and “Claude Opus 5.5” models were also used to develop the tool. The project’s main drawback is that it does not test the physical stability of structures. “LDraw Nova” prevents pieces from passing through one another, but it does not calculate whether the finished model will remain standing under real-world conditions. Creating a “Technic” mechanism with “Astra” costs approximately $5 (approximately 9 manat) in tokens, meaning the cost is calculated based on the volume of text in AI requests. Powerful AI systems are currently required for large and precise models, and the process also takes considerable time. The tool also works with “Meta Quest 3”, but speed and stability issues remain. The project differs from “BrickGPT”, a tool trained on more than 47 000 “LEGO” structures. “BrickGPT” also checks the physical stability and accuracy of the designs it creates. There are plans to use “LDraw Nova” to create minifigures, “Technic” cars and engines in the future. The developer wants the tool to work with less powerful AI agents as well. For now, all the models remain CAD files, but producing one of the smaller examples with a 3D printer is being discussed.
The process begins with the creation of a JSON plan describing the model’s structure and individual parts. The AI agent then writes a “Python” program. That program creates the main file in the “LDraw” format and lists each “LEGO” piece on a separate line. The agent can review the model’s image, correct errors and repeat the process until a satisfactory result is achieved. The completed files can be opened in “LDView”, “LeoCAD” and “Studio”. “LDraw Nova” is a web application that runs through “Docker”. It works with the “OpenAI”, “Anthropic” and “OpenRouter” services. The application’s gallery indicates which AI model and text prompt were used to create each structure. Examples include “Cathedral”, “Tidal Observatory”, the unfinished “Atlas Crane” and the “Copper Bean” residential buildings. The “Astra” and “Claude Opus 5.5” models were also used to develop the tool. The project’s main drawback is that it does not test the physical stability of structures. “LDraw Nova” prevents pieces from passing through one another, but it does not calculate whether the finished model will remain standing under real-world conditions. Creating a “Technic” mechanism with “Astra” costs approximately $5 (approximately 9 manat) in tokens, meaning the cost is calculated based on the volume of text in AI requests. Powerful AI systems are currently required for large and precise models, and the process also takes considerable time. The tool also works with “Meta Quest 3”, but speed and stability issues remain. The project differs from “BrickGPT”, a tool trained on more than 47 000 “LEGO” structures. “BrickGPT” also checks the physical stability and accuracy of the designs it creates. There are plans to use “LDraw Nova” to create minifigures, “Technic” cars and engines in the future. The developer wants the tool to work with less powerful AI agents as well. For now, all the models remain CAD files, but producing one of the smaller examples with a 3D printer is being discussed.