New Streamlit App: GPT-enabled storybook with images

utkrshm · June 14, 2023, 7:51pm

Check out this awesome Streamlit app I built -
https://story-gpt.streamlit.app/utkrshm/storygpt/main/app.py

It’s a multimodel AI program that creates a story, and changes it as you go, as per your decisions (It’s kind of like Goosebumps, except, get any genre/theme you want).
All with images, and all…

Do check it out, and let me know how I can improve it.

P.S. Here’s the GitHub repo: GitHub - utkrshm/StoryGPT: An interactive storybook built with the help of ChatGPT and Stable Diffusion.

Sully · June 16, 2023, 5:46am

I think your idea here is really fantastic. Your design looks nice and the code seems straightforward and well written – I like your clever session code for handling the input elements status as you progress through the story. It looks and feels polished. I want to see the prompts it generates for stable diffusion – that’s half the fun!

I tried it out and my story didn’t work out so well, but It was fun to play with:

I made an app with a similar concept-- mine does Tarot readings, but I setup prompts similar to yours that taught the model how to give me back some commands that I can parse out, so I know what to ask the user or when to draw tarot cards. It’s at https://emilytarot.com / GitHub - msull/emilytarot: AI Tarot Readings streamlit app if you are interested. One thing I added recently was checking the response to ensure I actually got back commands that make sense when I expect then, and regenerating the response if not:

github.com

msull/emilytarot/blob/main/src/streamlit_app.py#L297-L320


      
          with st.spinner("Generating AI response"):
              got_valid_response = False
              attempts = 0
              max_attempts = 3
              while not got_valid_response:
                  attempts += 1
                  response = _get_ai_response(chat_session)
                  try:
                      _extract_commands(
                          response["choices"][0]["message"]["content"]
                      )
                      break
                  except Exception:
                      print("Got bad response, trying again")
                      st.session_state.bad_responses.append(response)
                      st.session_state.total_tokens_used += response["usage"][
                          "total_tokens"
                      ]
                      if attempts >= max_attempts:
                          print("Out of attempts")

This file has been truncated. show original

You could consider something similar in your code to help handle situations like in part3 of my story, where presumably no questions were returned.

Cheers!

utkrshm · June 16, 2023, 4:38pm

@Sully thank you for trying out StoryGPT. I will most definitely work on that, and see that the stories all return options to choose from.

I really like the idea of adding a certain number of attempts to check if the prompt came out all right, and I will definitely add some of that functionality. Thanks for the suggestion!

Sully · June 23, 2023, 5:42pm

I totally hijacked your idea – I’ve been using ChatGPT with my kiddos to make little fun stories and things for a while now, particularly about some adventurous cats So I made a choose your own adventure generator for cat adventure stories. It doesn’t do any artwork yet but it’s still pretty fun to play with!

https://littlecattales.com

utkrshm · June 28, 2023, 1:11pm

I absolutely adore it!
This is more amazing than mine!

Thanks for sharing

Sully · June 28, 2023, 3:26pm

Hey thanks. I’ve been continuing to tinker with it. It generates 2 pictures per story now, and I added a browser for previous stories onto the main page last night. I really enjoy seeing all the art that it comes up with of the cats on their adventures

I also got the idea for another interactive fiction while working on this that I’ve been cooking up for a few days now-- hopefully I’ll have something to share soon!