Controllable Procedural Generation of Landscapes

Paper SupplementaryMaterial Video

This is the main repo for the ACM MM24 paper "Controllable Procedural Generation of Landscapes".

The Unity client repo: https://github.com/omegafantasy/ControllableLandscape_UnityClient

Environment

Python 3.8 (not tested with other Python versions, but it should be OK as long as the dependencies are alright)

Install the dependencies: pip install -r requirements.txt

An OpenAI API key is required to enable LLM queries. Put it in the .env file.

Structure

Codes are mainly in main.py and pcg.py.

sys_msg_en.py lists the prompts for LLM queries.

data/data.json stores some information about the used landscape models. (Please also refer to the Unity client repo)

The generation results will be stored in outputs/ .

Landscape Generation

Basic usage: python main.py

Possible arguments:

• -q The generation will use LLM queries. When not adding this, the generation use the default parameters and does not need LLM.
• -t [description text] The user input.
• -n [number of scenes] The number of scenes to generate. By default, the program only generate one scene at a time.
• -s [the random seed] Specify it if needed.
• -c [checkpoint name] A checkpoint stores a part of the generation result. (You can actually ignore this)

The direct outputs will be stored in outputs/. Move them to the Unity client repo and you can finally see the landscape scenes in Unity. (Please refer to the Unity client repo)

Others

If you encounter some bugs, please create an issue.