FW Picture
==========

Streamlit + LangChain app to suggest a picture plot for a diary entry.

- Select among popular OpenAI and Google models (order is persisted).
- Enter diary text and get a suggested picture plot.

API keys:
  - Set environment variables ``OPENAI_API_KEY`` and ``GEMINI_API_KEY``
    or enter them in the sidebar inputs.

::

  import os
  import json
  from pathlib import Path

  import streamlit as st

LangChain components

::

  from langchain_core.prompts import ChatPromptTemplate
  from langchain_core.output_parsers import StrOutputParser

Prompt

::

  PICTURE_PROMPT = """
  You are a creative assistant. Given a personal diary entry, 
  propose a single, vivid, camera-ready picture plot that could illustrate the entry. 
  Keep it concise (60–120 words). Include: subject, setting, mood, style/medium, key visual elements. 
  "Avoid first person; write as a scene description.
  """

Model specifications

::

  MODEL_ORDER_DEFAULT = [
      "openai/gpt-5",
      "google/gemma-3-27b-it",
      "openai/gpt-5-mini",
      "google/gemini-2.5-pro",
      "google/gemini-2.5-flash",
      "google/gemini-2.5-flash-lite",
  ]

Extracts provider and model name from a model ID.
Args:
  model_id: The model ID in the format "provider/model".
Returns:
  A tuple (provider, model).
::

  def extract_model_spec(model_id: str) -> dict:
      provider, model = model_id.split("/")
      return {"provider": provider, "model": model}

See: PersistedList_

.. _PersistedList: PersistedList.py.html
  
::

  from PersistedList import PersistedList

LLM
---

::
    
  def get_llm(model_id: str):
      spec = extract_model_spec(model_id)
      provider = spec["provider"]
      model_name = spec["model"]
      temperature = 0.1

      if provider == "openai":
          try:
              from langchain_openai import ChatOpenAI
          except ImportError:
              raise RuntimeError(
                  "Missing dependency: pip install langchain-openai"
              )
          # ChatOpenAI will read OPENAI_API_KEY from env
          if not os.environ.get("OPENAI_API_KEY"):
              raise RuntimeError("OPENAI_API_KEY is not set.")
          return ChatOpenAI(model=model_name, temperature=temperature)

      if provider == "google":
          try:
              from langchain_google_genai import ChatGoogleGenerativeAI
          except ImportError:
              raise RuntimeError(
                  "Missing dependency: pip install langchain-google-genai"
              )
          # ChatGoogleGenerativeAI will read GEMINI_API_KEY from env
          if not os.environ.get("GEMINI_API_KEY"):
              raise RuntimeError("GEMINI_API_KEY is not set.")
          return ChatGoogleGenerativeAI(model=model_name, temperature=temperature)

      raise ValueError(f"Unknown provider: {provider}")


  def build_chain(llm):
      prompt = ChatPromptTemplate.from_messages(
          [
              (
                  "system",
                  PICTURE_PROMPT,
              ),
              ("human", "{diary}"),
          ]
      )
      return prompt | llm | StrOutputParser()

main

::

  st.set_page_config(page_title="FW Picture", page_icon="🎨")
  st.title("🎨 FW Picture")

Persisted order

::

  model_order_persisted = PersistedList(".fw_picture")
  order = model_order_persisted.sort_by_pattern(MODEL_ORDER_DEFAULT)

  # Determine current selection id (persisted across reruns)
  selected_id = st.session_state.get("model_selected_id")

  # Build options by current order
  option_labels = order

  # Always show the last-selected model first in the list (index 0)
  new_selected_id = st.selectbox("Model", options=option_labels, index=0, key="model_select_label")

  # Update selected_id and persist ordering if changed
  if new_selected_id != selected_id:
      model_order_persisted.select(new_selected_id)

  diary_text = st.text_area(
      "Freewrite",
      height=200,
      key="diary_text",
  )

  if st.button("Suggest picture plot", type="primary"):
      if not diary_text.strip():
          st.warning("Please enter a diary entry.")
          st.stop()
      try:
          llm = get_llm(st.session_state.model_selected_id)
      except Exception as e:
          st.error(f"Model initialization failed: {e}")
          st.stop()

      chain = build_chain(llm)
      with st.spinner("Thinking up a picture plot..."):
          try:
              suggestion = chain.invoke({"diary": diary_text})
          except Exception as e:
              st.error(f"Generation failed: {e}")
              st.stop()

      st.subheader("Suggested picture plot")
      st.write(suggestion)