YxuanAr

YxuanAr / Code-as-Room

Public

A MLLM-based agentic system converts a single room image into executable Blender code for 3D room reconstruction.

65
2
89% credibility
Found May 22, 2026 at 72 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Code-as-Room is an AI-powered research tool that transforms a simple top-down photograph of any room into a complete, renderable 3D scene. The system uses a multi-stage pipeline where AI agents analyze the room layout, generate 3D geometry code, apply realistic materials and textures, and set up professional lighting - all outputting a Blender Python script that produces photorealistic 3D rooms for architecture, robotics training, or game development.

How It Works

1
📸 Share a photo of your room from above

Take a top-down photo of any room - a bedroom, lab, office, or kitchen - and upload it to the tool.

2
🤖 Watch AI understand your space

The system studies your image to identify furniture, walls, and how everything is arranged, just like looking at a floor plan.

3
🏗️ Your 3D room takes shape

Behind the scenes, the tool writes code that builds your room in Blender - walls, floors, furniture, all positioned correctly.

4
🎨 Colors and textures come alive

The system adds realistic materials - wood grain on furniture, tile patterns on floors, paint on walls - matching your room's style.

5
✨ Your room is ready to render

A complete Blender script is generated that you can open in Blender to see your photorealistic 3D room come to life.

Sign up to see the full architecture

3 more

Sign Up Free

Star Growth

See how this repo grew from 72 to 65 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 Code-as-Room?

Code-as-Room is an MLLM-based agentic system that converts a single top-down room image into executable Blender code for 3D reconstruction. Built in Python, it uses a multi-stage pipeline that progressively transforms visual evidence into scene understanding, object layout, geometry code, materials, textures, and render settings. You feed it an image; it outputs a runnable Blender script.

The pipeline runs 12 stages, from scene classification through spatial analysis, code generation, material assignment, and final rendering. Optional extensions handle small-object geometry and texture generation via image models.

Why is it gaining traction?

The hook is simple: take a floor plan image, get a 3D scene without modeling anything manually. The agentic approach means the system self-corrects through validation loops, and the multi-stage design lets you resume from any checkpoint if something goes wrong. Batch processing support makes it practical for generating multiple rooms at once. The incremental layout engine handles collision detection and object placement automatically.

Who should use this?

3D artists who need rapid scene prototyping from reference images will find this most useful. Game developers working on level layouts can use it for quick iteration. Researchers exploring multimodal code synthesis will appreciate the structured pipeline. Blender users who want to automate scene creation from real-world floor plans will benefit most. It is less suited for production pipelines requiring deterministic outputs or teams without API budgets.

Verdict

This is a promising but nascent project with 65 stars and a credibility score of 0.8999999761581421%. The documentation is thorough and the pipeline architecture is well-thought-out, but the dependency on external LLM APIs means costs and quality vary. Best for experimentation and prototyping rather than production use.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.