Using AI to Caption My Japan Photos

You are creating captions for a travel blog called 'Dan in Japan' about a vacation taken by two couples in October 2024. Write natural, conversational captions in all lowercase that avoid clichés and overwrought emotional descriptions. You are being given a photo from the trip along with a screenshot of a map showing where the photo was taken using the serpAPI from the photo coordinates. The items identified in the map are for context only, it doesn't mean that the photo is exactly of something shown in the map. The photographer was {photographer} and the date it was taken was {date_taken} at {location_context}. If Chuck took the photo, then it's him and his wife Ashley. If Daniel took the photo, then it's him and his wife Christina, and vice-versa. Trip details: - First time in Japan - 10 days during second half of October 2024 - Visited Tokyo, Osaka, and Kyoto - Stayed in: - Tokyo (Airbnb) - Osaka (traditional Japanese house) - Kyoto (western hotel) - Any photos you see taken far outside of these cities would have been on the bullet train. Caption guidelines: - Always identify the people in the photo by name - Write as if texting a friend - casual and genuine - Avoid cliché travel writing phrases like "capturing the moment" or "radiating excitement" - No exclamation points unless absolutely necessary - Skip obvious details (we can see they're smiling/happy/etc) - Keep it to 1-2 short sentences - For US photos (pre/post trip), mention that it's from before/after the Japan trip Don't: - Use flowery or emotional language - Describe obvious visual elements - Add fictional details - Assume activities from map locations Your Points of Interest should be notable locations, cultural elements, or historical references that warrant further explanation in a modal window. Skip mundane points of interest. Each point of interest you generate is going to be clickable and its going to open a google search for the phrase you pick, so choose things that can be googled and are japan related. Here are the images to analyze:

from pydantic import BaseModel
from typing import List

class PointOfInterest(BaseModel):
    name: str
    description: str

class PhotoAnalysis(BaseModel):
    caption: str
    points_of_interest: List[PointOfInterest]

def generate_caption(...):
    # Initialize Anthropic client with Instructor
    anthropic_client = instructor.from_anthropic(
        Anthropic(api_key=os.environ['ANTHROPIC_API_KEY'])
    )

    # Send the request to the AI model
    response = anthropic_client.messages.create(
        model='claude-3-5-sonnet-latest',
        messages=[{'role': 'user', 'content': instruction_text}],
        response_model=PhotoAnalysis,
        temperature=0
    )

    # Extract caption and points of interest from the response
    return response.caption, [poi.model_dump() for poi in response.points_of_interest]