This project is a procedural dungeon generator designed for Shadowdark RPG.
It leverages a combo of AI-driven prompts and random inputs to generate a dungeon.
- AI-driven ancient history generation.
- AI-generated, history-specific factions.
- Procedurally created concept map of the dungeon.
- AI-driven room descriptions.
- Flux.dev AI-powered image generation for dungeon rooms.
Ensure you have Python 3 installed along with the required dependencies.
pip install -r requirements.txt
In addition, you'll need Ollama and a local model installed. The script currently implements openthinker:7b, an uncensored distillation of deepseek-r1, as the primary model.
Run the generator with:
python main.py
Valid inputs are yes
, no
, and debug
. You can also supply those at runtime with -y, -n, and -d flags.
Specify no
if you have some ideas about the dungeon you want to generate.
Specify debug
if you want to skip the first three steps of generation.
- Dungeon Content not being written to final doc.
- Final doc has formatting problems.
- Sometimes, the dungeon map adds nodes outside the view.
- Hardcoded models, temperature, and context window.
- Somewhat questionable room content outputs.
- Image generation requires significant GPU memory (at least 8GB VRAM).
- Fix Known issues.
- Implement loops.
- Generate random encounter tables.
- Further refine AI prompts for better room outputs.
Pull requests are welcome! Not looking for bug reports; I got enough problems ;)
All rights reserved for now.