physx-omni

physx-omni / PhysX-Omni

Public

PhysX-Omni: Unified Simulation-Ready Physical 3D Generation for Rigid, Deformable, and Articulated Objects

physx-omni.github.io 3d image-to-3d physical-modeling
80
0
94% credibility
Found May 23, 2026 at 84 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Jupyter Notebook
AI Summary

PhysX-Omni is a research system that transforms ordinary photos of objects into complete, physics-ready 3D models. When you show it a picture of something like a chair or a box, it analyzes the image, identifies each part, generates 3D geometry for every component, and automatically adds realistic physics properties like joint connections, material types, and weights. The result is a simulation-ready 3D object that can be dropped directly into physics engines for robotics research, game development, or scientific simulations. The project includes tools for generating individual objects, combining them into complex scenes, and benchmarks to evaluate how well the generated objects behave in physics simulations.

How It Works

1
📸 You share a photo of an object

You start by showing the system a picture of any 3D object you want to recreate - it could be a chair, a toy, a box, or anything with multiple parts.

2
🤖 AI understands your object's structure

The system studies your image and identifies each separate part, what materials they look like, and how they might connect or move.

3
🎨 Your object becomes a 3D model

The AI generates detailed 3D geometry for every part, creating a complete digital twin that looks just like your original object.

4
⚙️ Physics properties are added automatically

The system figures out where joints should go, how heavy each part is, and what materials they are made of - all the physics details you need for simulation.

5
Choose how to use your model
🎮
Simulate in physics engine

Drop your object into a physics simulator to watch it interact with other objects, test how it moves, or see how materials behave.

🏗️
Build entire 3D scenes

Combine multiple generated objects to create complex scenes with realistic physics interactions between all parts.

✨ Your simulation is ready to go

You now have a fully functional 3D object with physics properties that can be used in robotics, gaming, or scientific simulations.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 84 to 80 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 PhysX-Omni?

PhysX-Omni generates physics-ready 3D objects from single images. Feed it a photo of a chair, and it outputs a complete simulation scene with proper joints, materials, and collision geometry ready for MuJoCo or other physics engines. The pipeline chains together a vision-language model for understanding the object, a decoder for generating 3D geometry, and converters that output standard formats like URDF and MJCF XML. It handles rigid bodies, deformable objects like cloth or soft materials, and articulated assemblies with hinges, sliders, and ball joints.

Why is it gaining traction?

This solves the "3D asset bottleneck" for robotics and simulation. Instead of manually building physics models, you point at an image and get something you can drop into a MuJoCo scene immediately. The benchmark suite (PhysX-Bench) with seven evaluation metrics gives researchers a standardized way to compare outputs. The scene composition tool lets you place multiple generated objects together, which is rare in this space. It plugs into the NVIDIA Omniverse ecosystem while remaining independent.

Who should use this?

Robotics researchers who need to rapidly generate simulation assets from real-world objects. Game developers prototyping physics interactions. Anyone building training datasets for manipulation tasks where manual asset creation is the bottleneck. The learning curve is steep--expect to spend time on environment setup and understanding the inference pipeline--but the output quality justifies it for production workflows.

Verdict

With an 0.949999988079071% credibility score and 80 stars, this is a credible academic project from a research lab, not a production-grade library. The documentation is thorough but assumes familiarity with 3D graphics pipelines. If you're building a research prototype around physics-based 3D generation, this is worth evaluating. For production use, budget time for hardening the inference pipeline and adding error handling where the current code is sparse.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.