filliptm

filliptm / HY-World-2.0-Playground

Public

Web UI for running Tencent's WorldMirror 2.0 locally — images or video → 3D Gaussian Splats, point clouds, depth, normals, cameras. Blackwell GPU and Windows-ready.

13
2
69% credibility
Found Apr 17, 2026 at 13 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

A user-friendly local web interface for Tencent's WorldMirror 2.0 that reconstructs 3D Gaussian splats, point clouds, depth maps, normals, and camera poses from images or videos with one-click setup and interactive viewers.

How It Works

1
🔍 Discover HY-World Playground

You find this fun tool that turns your photos or videos into amazing 3D models, like magic reconstruction.

2
📥 One-click setup

Run the simple install script and it prepares everything on your computer, no hassle.

3
🚀 Open your web playground

Click run and a friendly web page opens in your browser, ready to explore.

4
📸 Pick a sample or upload your own

Browse cute example scenes or drag your vacation photos and videos right onto the page.

5
⚡ Hit reconstruct

Click the big button and watch live progress as it builds your 3D world from every angle.

6
🕶️ Dive into interactive 3D

Spin around glowing splats, zoom point clouds, compare depth maps side-by-side – it's like being there.

💾 Download and share your scene

Grab your 3D model files, rendered videos, and camera info to use anywhere – your creation is ready!

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 13 to 13 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 HY-World-2.0-Playground?

This Python project delivers a local web UI for Tencent's WorldMirror 2.0, converting images or videos into 3D Gaussian splats, point clouds, depth maps, normals, and camera poses—all via drag-drop uploads or 22 bundled examples. Launch with `./install.sh` and `./run.sh` (or `.bat` on Windows), hit localhost:5173, tweak 28 params in collapsible panels, and track live progress with thumbnail previews. Blackwell GPUs and Windows get native support through auto-detected kernel compiles, pulling 2.4GB model weights on first run.

Why is it gaining traction?

It patches upstream gaps like Windows temp dirs, flash-attn fallbacks, and Blackwell arches (sm_90+PTX), enabling seamless code running web on new hardware without manual tweaks. Polished Vite frontend offers inline splat/point/depth viewers with frustum overlays, side-by-side inspectors, job history, and one-click reruns—far beyond CLI output. Like a github web ide for 3D recon, it exposes every modality and param without gradio boilerplate.

Who should use this?

Vision researchers iterating Gaussian splat pipelines from casual video clips. Game devs prototyping running web games with quick scene captures. ML engineers on RTX 50xx Blackwell cards needing a web running python playground for WorldMirror 2.0 inference.

Verdict

Worth forking for Blackwell/Windows users—early (13 stars, 0.699999988079071% credibility) but battle-tested patches and crisp UI outpace upstream. Maturity lags in tests/community; test on toy inputs first.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.