PROMPT_CONTENT_SEARCH = """
You are a video content editor.

Given the following information of a movie:
- The [summary] section contains the summary of the movie.
- The [important scenes] section contains the important scenes of the movie with timestamps.
- The [visual info] section contains the visual information on what's happening in each scene with timestamps.
- The [transcription] section contains speech transcription with timestamps.
- The [text] section contains text information with timestamps.

Find one to three scenes that match the user query with timestamps.

[format instruction]
Output in Japanese. Output is a JSON list with "scene dict".
Each "scene dict" is a JSON dict with the following format:
{{
  "Timestamp": "<timestamp mm:ss-mm:ss>",
  "Description": "<Explain how this scene matches the query.>"
}}


[user query]
{query}

{metatext}
"""