GitHub radar
Turn one photo into a flyable 3D video scene
Guangyu Zaojing (光域造景) turns a single photo and short prompt into a 3D scene you can fly through like a video — no 3D modeling skills needed.
A system that takes one image and a description, then builds a scene out of a point cloud (millions of colored dots) you can watch as a flythrough video.
When it helps
Good for quickly showing a look-around of a room or landscape for a review or a class. Building your own custom scenes needs a strong NVIDIA GPU.
Pros
- View a ready-made 3D scene in a browser, no GPU needed
- Comes with finished example scenes: anime cabin, a girl, more
- Download the point cloud as a PLY file for SuperSplat or MeshLab
- Depth estimation, scene generation, and rendering are already wired together
Cons
- Making your own scenes needs an NVIDIA GPU with 16+ GB VRAM
- Interface is in Chinese, so you'll need to translate as you go
- Full mode downloads about 30 GB of model weights on startup
How to set it up — step by step
- 1Clone the deepopen-com/deepopen-3d-webs repository (download the project code) to your computer.
- 2Open your AI agent (Claude Code, Codex) in the project folder.
- 3Send the agent the text from the block below.
- 4Wait for the agent to run demo mode via docker compose up -d --build (container build command) — no GPU required.
- 5Open http://localhost:8090 in your browser and check the result.
- 6Look at the Example Gallery section with ready-made scenes.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up Guangyu Zaojing from https://github.com/deepopen-com/deepopen-3d-webs — bring up the no-GPU demo mode with docker compose following the README, open http://localhost:8090, and show me what's in the Example Gallery tab.
Browsing the ready examples and spinning the point cloud in the browser works on a regular laptop with Docker and 8 GB of memory — no graphics card needed. Generating your own scenes needs an NVIDIA GPU with 16 GB or more, which a laptop cannot do; you would need to rent a server.
Open on GitHub▌ More finds