This is an official implementation of AAAI 2025 paper StoryWeaver: A Unified World Model for Knowledge-Enhanced Story Character Customization. The proposed StoryWeaver can achieve both single- and multi-character based story visualization within a unified model. For more detailed explanations and supplementary materials, please refer to our arXiv version. At the same time, we will continuously update the arXiv paper with more experiments.
- Adding more experiments results on more stories world, e.g., Flintstons, Pokemon,...
- Code Released.
- Paper Released.
The codebase is tested on
- Python 3.8
For additional python libraries, please install by:
pip install -r requirements.txt
use the shell script,
bash train.sh
use the shell script,
bash sample.sh
Tip
We highly recommend performing Knowledge-Enhanced Spatial Guidance on the coarse class label too for each character in the text descriptions. For example, the coarse label for character Loopy is beaver. Please refer to the code of the dataset for more details.
