EESJGong

EESJGong / Graph-CAD

Public

Learning Hierarchical and Geometry-Aware Graph Representations for Text-to-CAD

3d-generation 3d-modeling blender blender-python cad
18
5
100% credibility
Found Mar 27, 2026 at 18 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Graph-CAD is a research tool that converts natural language descriptions into hierarchical 3D CAD models and runnable Blender scripts via a structured graph planning process.

How It Works

1
📰 Discover Graph-CAD

You hear about a cool tool that turns simple words into 3D designs, like describing a gadget and seeing it built automatically.

2
📥 Gather your tools

You download the ready-made brainpower files and set up a quiet workspace on your computer so everything can think and create.

3
💡 Describe your dream design

You type a fun instruction like 'make a router with antennas' and hit go, feeling excited as the magic starts.

4
🔄 It plans and builds step by step

The tool thinks through the shape, pieces, and how they fit together, creating a blueprint just from your words.

5
🖼️ View your 3D creation

Beautiful pictures of your design appear from different angles, ready to explore in a 3D viewer.

6
📊 Check the quality

You review how well it matches your idea, seeing scores on shape, position, and instructions followed.

🎉 Celebrate your custom model

Now you have a perfect 3D model from words alone, ready to print, share, or tweak – pure creation joy!

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 18 to 18 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is Graph-CAD?

Graph-CAD takes natural language instructions like "create a router with antennas" and generates executable Blender code for CAD models. It solves the fragility of direct text-to-code generation on complex assemblies by using a hierarchical graph of parts and geometric constraints as an intermediate step, followed by action planning and code output. Built in Python with fine-tuned Qwen models served via LlamaFactory, it processes CADBench.jsonl benchmarks or single prompts through a simple CLI pipeline.

Why is it gaining traction?

It outperforms end-to-end LLMs and baselines like BlenderLLM on CADBench metrics for attributes, spatial relations, and instruction fidelity, especially geometric constraint satisfaction via its graph cad drawing approach. Developers dig the three-stage inference—text to graph cad, actions, then bpy code—that cuts syntax errors to 2% and handles long-horizon builds better. Model weights on ModelScope make it dead simple to spin up locally for graph code experiments.

Who should use this?

CAD engineers turning specs into Blender prototypes without manual scripting. AI researchers benchmarking text-to-3D or learning hierarchical graph representations, akin to cross-modal gesture gen. Hobbyists prototyping graph cad vs usd-style visualizations or cad currency graph tools in 3D.

Verdict

Grab it if you're into text-to-CAD research—benchmarks crush alternatives, docs guide setup to renders and evals cleanly. At 18 stars and 1.0% credibility, it's pre-release raw (TODOs linger), so expect tweaks, but weights and CLI deliver immediate value for local tinkering.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.