FairyTailor: Multimodal Generative Framework for Storytelling

Human-in-the-loop visual story co-creation.

Users can create a cohesive children's story by weaving generated texts and retrieved images with their input.
With co-creation, writers contribute their creative thinking, while generative models contribute to their constant workflow.
FairyTailor adds another modality and modifies the text generation process to help producing a coherent and creative story.


Set-up (development)

After cloning the repository:

Client (Vue 2.6)

Install and check that the client compiles:

cd client
npm i
npm run build

Backend (FASTAPI)

Install and activate the environment (conda provided):

conda env create -f environment.yml
conda activate MultiModalStory

Install environment globally in the directory:

pip install -e .
pip install git+https://github.com/openai/CLIP.git

After installation run:

python -m spacy download en_core_web_sm

In python terminal:


Large Data Management (dvc)

Our large data files are stored on IBM's Cloud Object Storage, and to pull data files from that platform you will use a special, read-only .dvc/config file.

dvc pull -f

Which will pull:

  • backend/outputs (five preset stories)
  • backend/story_generator/downloaded (transformers)
  • client/public/unsplash25k (styled images)

Running the framework during developemnt


cd client
npm run devw

Backend (with server auto reload):

uvicorn backend.server:app --reload --reload-dir backend

Open the uvicorn server localhost:8000 in your web browser

Modifications Ideas:

New huggingface transformer

  • Place the transformer in backend/story_generator/downloaded directory.
  • Update the current model path by changing the constant FINETUNED_GPT2_PATH in backend/story_generator/constants.py.

New images folder

  • Replace the folder client/public/unsplash25k/sketch_images1024 with yours.
  • Update the current path by changing the constant IMAGE_PATH in client/src/components/Constants.js.

API functionalities

  • Add functions to the backend endpoint at backend/server/main.py.
  • Update client/src/js/api/mainApi.js to call the backend endpoint from the client.
  • Update the corresponding user components in client/src/components.