jjrCN

jjrCN / PanoWorld

Public

Official repo for the paper "PanoWorld: A Generative Spatial World Model for Consistent Whole-House Panorama Synthesis"

85
7
100% credibility
Found May 25, 2026 at 85 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

PanoWorld is an AI research project that takes a few 360-degree panoramic photos from different rooms in a house and automatically generates complete whole-house visualizations. It uses deep learning to understand how rooms connect and maintains consistent materials, colors, and geometry across the entire house. The system outputs both rendered images and interactive 3D point clouds that can be explored like a virtual tour. This is useful for real estate visualization, VR tour creation, or any application needing complete spatial understanding of indoor environments.

How It Works

1
🔍 You discover a new way to visualize houses

You hear about PanoWorld - an AI that can take a few photos of a house and generate complete 360-degree views of every room.

2
📸 You gather your panoramic photos

You collect 360-degree photos from different rooms in a house - the AI will use these as starting points to understand the space.

3
✨ The AI learns the whole house at once

Instead of processing rooms one by one, the AI understands how all rooms connect together, keeping walls, floors, and materials consistent everywhere.

4
Choose your quality level
🖼️
Standard quality (1024x512)

Faster processing with good results for most uses like previews or web tours.

🏠
High resolution (2048x1024)

Crisper details for professional real estate listings or detailed inspections.

5
🎨 Watch as new views come to life

The AI generates fresh perspectives of each room, filling in views you never photographed while keeping everything looking natural.

6
🕹️ Explore your house in 3D

Your results include interactive 3D point clouds you can fly through, like walking through a virtual tour.

🎉 You have a complete house visualization

From a handful of photos, you now have consistent 360-degree views of an entire house, ready for virtual tours or further editing.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 85 to 85 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 PanoWorld?

PanoWorld is a research project that generates consistent 360-degree panorama images for entire houses from floorplans and style references. Given a floor layout and a reference image, it autoregressively synthesizes room-by-room panoramas that maintain geometric consistency and material continuity across the whole space. Built in Python using PyTorch, it combines transformer-based neural networks with 3D Gaussian Splatting to render the final output. The official repository releases inference code, pre-trained model checkpoints at two resolutions (1024x512 and 2048x1024), and evaluation data from the RealSee3D dataset.

Why is it gaining traction?

The hook here is whole-house consistency. Most panorama generation tools handle single rooms in isolation. PanoWorld explicitly maintains cross-room geometry and material identity, which matters for virtual tour applications where users navigate between spaces. The floorplan-derived 3D shell provides structural guidance that most alternatives lack. It also outputs renderable 3D Gaussian point clouds alongside the panoramas, giving developers both the synthesized images and the underlying geometry for downstream applications.

Who should use this?

VR tour developers building real estate or architectural visualization platforms will find this most useful. Researchers in novel view synthesis or 3D scene reconstruction can use the released checkpoints as baselines. If you're building tools that need consistent multi-room environments from minimal input, this is worth evaluating. However, the inference-only release means you'll need to wait for training code if you want to fine-tune on your own data.

Verdict

PanoWorld addresses a real gap in spatial generation but comes with the expected caveats of a fresh research release. With only 85 stars and a 1.0% credibility score, this is early-stage work from a single company (Ke Holdings). The documentation is functional but sparse, and the full pipeline including training code remains forthcoming. Evaluate it for the novelty of the approach, but treat it as a research preview rather than production-ready infrastructure.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.