Storytelling Domain

This is the repository holding code we used in the experiments for "Beyond Prompts: Exploring the Design Space of Mixed-Initiative Co-creativity Systems".

Accepted by ICCC 2023!

You can find the paper here: https://arxiv.org/abs/2305.07465

Try the exemplar application

• All the scripts are located StorytellingDomain/Scripts.
  • You need to run the scripts from that directory (working directory).
• Set up Addons
  • Run the script setup_addon_server.sh, or review it if you wish to download models to other folders.
    • If you do choose to put models somewhere else, you need to modify StartAddonServer.py accordingly. To set this repository up:
• Set up environments
  • Create a conda env / virtual env named storytelling-domain-for-creative-wand
    • If you wish to use a different name, the run- scripts needs to adapt to it.
  • Clone the Creative Wand framework (https://github.com/eilab-gt/CreativeWand)
  • cd to the Creative Wand framework
  • pip install -e .
  • Then, pip install -r requirements.txt
  • If your conda env does not contain python yet, run conda install python. We developed under 3.10.
• Start the servers
  • Run the scripts run_addon_server.sh and run_backend_server.sh
    • By default we load GPT-J, which requires 24G of VMem or more.
    • Alternatively you can run the run_addon_server_smaller_model.sh with 8G VMem with GPT2.
      • The quality of generation will be poorer.
  • If you wish to run it manually:
    • You need to add StorytellingDomain/ to your PYTHONPATH.
    • The backend entry point is StorytellingDomain/ Application/Deployment/StartExperiment.py
    • and StorytellingDomain/ Application/Deployment/StartAddonServer.py
  • Run web-frontend/web-interface/run_http_server.sh
  • Manual setup: The frontend entry point is npm run start with web-frontend/package.json
    • The first time you run it you need to do npm install to for js dependencies.
    • You may also need to install conda install nodejs==18.11.0 (Or any newer versions)
• Use the system
  • Access your Creative Wand at http://localhost:3000/?mode=s2_f&pid=test .
    • mode is defined in StorytellingDomain/Application/Deployment/Presets/StoryPresets.py.
    • The modes featured in the experiments are:
      • s2_f for the "Full" system;
      • s2_g for the "Global-only ablation" system;
      • s2_l for the "Local-only ablation" system;
      • s2_e for the "Elaboration-only ablation" system;
      • s2_r for the "Reflection-only ablation" system;
      • s2_h for the "Human-Initiated-only ablation" system;
      • s2_a for the "Agent/AI-Initiated-only ablation" system;
    • pid was used to associate sessions with participants answering questionnaires.

Add your own Communications

Check StorytellingDomain/Application/Deployment/Presets/StoryPresets.py on how modes are defined; Starting there, we have documents on functions that will walk you through the rest of the system.

Set up HTTPS access (optional)

If you wish to add a certificate so that the sites can operate in HTTPS

• Pass -s to StartExperiment.py
• In WebFrontendServer.py set the paths to the certificates
• Set environments HTTPS=true;SSL_CRT_FILE=<your fullchain.pem>;SSL_KEY_FILE=<your privkey.pem> for the node frontend.

If you found this helpful

Please consider citing our work using the following:

@misc{lin2023prompts,
      title={Beyond Prompts: Exploring the Design Space of Mixed-Initiative Co-Creativity Systems}, 
      author={Zhiyu Lin and Upol Ehsan and Rohan Agarwal and Samihan Dani and Vidushi Vashishth and Mark Riedl},
      year={2023},
      eprint={2305.07465},
      archivePrefix={arXiv},
      primaryClass={cs.AI}
}