nepfaff

nepfaff / scenesmith

Public

Code for "SceneSmith: Agentic Generation of Simulation-Ready Indoor Scenes"

scenesmith.github.io
264
15
100% credibility
Found Feb 12, 2026 at 163 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

SceneSmith uses AI agents to automatically generate complete, physics-ready indoor scenes from text prompts for robotics simulation and virtual environments.

How It Works

1
🔍 Discover SceneSmith

You find SceneSmith online through a research paper or project website and get excited about creating realistic indoor spaces from simple text descriptions.

2
📥 Gather your pieces

You download free scene building blocks like furniture models and materials, then connect smart AI helpers to make everything work smoothly.

3
✏️ Describe your dream space

You type a natural description like 'a cozy living room with a sofa, coffee table, and plants near the window' to tell the system what you want.

4
🚀 Watch magic happen

With one simple command, the AI agents build your entire scene step by step – floor plan, furniture, decorations – all physically realistic and ready to use.

5
👀 Preview in 3D

You open an interactive viewer to walk around your new space, check details, and make sure everything looks perfect.

6
💾 Export for action

You save the scene in formats ready for robot simulators, games, or any physics engine to bring it to life.

🎉 Your scene is simulation-ready

Now you have a fully interactive indoor world where robots can navigate, objects can be picked up, and physics works just like real life – all from your words!

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 163 to 264 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 scenesmith?

SceneSmith is a Python code github repository that turns text prompts into fully simulation-ready indoor scenes, handling everything from floor plans to furniture placement and small objects. Using agentic AI workflows, it generates houses or rooms complete with physics properties, articulated cabinets, and PBR materials, outputting Drake SDF files ready for robotics sims or export to MuJoCo/USD. Developers get complete, interactive environments without manual modeling—perfect for code github ai experiments in scene generation.

Why is it gaining traction?

It stands out by automating the full pipeline: multi-agent planning critiques placements for physical realism, supports generated or retrieved assets (HSSD/Objaverse), and includes Docker for easy GPU setup with multi-GPU rendering. The code github readme details CLI commands like `python main.py +name=my_experiment` for quick starts, plus branching/resuming for iterative tweaks. No more stitching meshes manually—scenes are stable and sim-ready out of the box.

Who should use this?

Robotics engineers building manipulation tasks in sim, RL researchers needing diverse indoor training envs, or sim devs prototyping robot homes/kitchens. Ideal for those tired of ProcTHOR's limitations or hand-crafting scenes in Blender/Drake.

Verdict

Try it for agentic indoor scene gen—strong docs, tests, and Docker make setup painless despite 64 stars and 1.0% credibility score signaling early maturity. Polish physics exports and scale to production for a winner.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.