This project is part of my final assignment for a course on Web Mining and Cognitive Computing at my university. See below for the original wording of the assignment.
To run this app, you need a current version of Python, preferably 3.10 or higher.
Furthermore, you will need to install the following packages:
| Package Name | Installation Command |
|---|---|
streamlit |
pip install streamlit |
openai |
pip install openai |
langchain |
pip install --upgrade langchain |
langchain-openai |
pip install --upgrade langchain-openai |
tiktoken |
pip install --upgrade tiktoken |
Other packages needed by the script should come with the base installation of your Python distribution.
In case you are using a conda environment, make sure to activate the environment first before running the streamlit command.
You will also need your own OpenAI API key, which can be obtained on your OpenAI Account page.
To run this script on your local machine, you will need to clone this repository:
git clone https://github.com/NightmindOfficial/wmcc_storybot.git
After that, you can navigate in the folder and execute the app with:
streamlit run storybot.py
If everything is in order, your default browser should open the app automatically. In case you need to do this manually, use the following URL:
https://localhost:8501
Note: The program will ask you to provide your own API key since I cannot reliably ensure everyone has an API key in their envs.
On the sidebar, you can control which GPT models are used, log out, reset the chatbot, and view some debug information (e.g., total tokens used in the last prompt).
After the first two user inputs, the bot suggests three keyword options to continue the story from. Klicking the buttons directly continues the story with the selected prompt.
After the next story fragment is generated, you have to wait for the corresponding image to generate before continuing with the next story.
During generation, you can still update the current prompt by re-submitting a new prompt or clicking on one of the remaining buttons. This is by design to allow for instantaneous changes in case you misclicked something.
To have the bot write a (happy) end for your story, instruct the bot to "end the story now", or a similar prompt. Story endings will not have any keyword suggestions attached. However, the chat history does not clear, making it possible to continue a second story which intertwines with the plot of the previous story and allows your children to unleash their creativity to their fullest!
You can contact me directly via DM if you encounter any issues or should you have any comments.
Create a streamlit app that automatically creates endless bedtime stories for kids. The app should be visually appealing and create textual and visual content.
The use case is as follows:
- Parent enters a text: Explain the scenery of a fairytale castle on a dark and stormy night.
- The LLM should provide the textual description and an image. You should visualize it properly in the app. Make it user-friendly.
- After text and image have been visualized, ask again for a follow-up:
- Parent enters: A knight is riding slowly towards the castle; explain this scenery and provide feasible options for how the story should continue.
- Now, the LLM should take into consideration what happened before, create a scenery text and image, and options for how to continue. The options should be made visible as a Button using just 1-3 keywords describing the options.
- The parent selects options and enters again the next step of the story.
- This is an iterative process until the parent/user tells the system to complete the story. The context windows (token length) should never exceed 4000 tokens. This means that you have to summarize the text in between using appropriate tools.