Automatically shuffles every 5 minutes or narrates when the collection image changes.

Collection image narrator automation.yaml

id: collection_image_narrator
alias: Collection image narrator
description: Automatically shuffles every 5 minutes or narrates when the collection image changes.
mode: restart
triggers:
  - trigger: state
    entity_id:
      - image.media_collection
    id: state_changed
  - trigger: time_pattern
    minutes: "/5"
    id: timer_trigger
conditions:
  - condition: template
    value_template: "{{ states('image.media_collection') not in ['unknown', 'unavailable'] }}"
actions:
  # Action 1: If triggered by timer, shuffle to a new image
  - choose:
      - conditions:
          - condition: trigger
            id: timer_trigger
        sequence:
          - action: collection_image.shuffle
            target:
              entity_id: image.media_collection
          - delay:
              seconds: 2

  # Action 2: Extract EXIF metadata (Can be disabled if you don't want Python!)
  - action: shell_command.extract_exif_data
    data:
      image_file: "{{ state_attr('image.media_collection', 'entity_picture') }}"
    response_variable: exif_output

  # Action 3: Save local snapshot frame for AI vision task
  - action: image.snapshot
    target:
      entity_id: image.media_collection
    data:
      filename: /media/collection_snapshot.jpeg

  # Action 4: Ask LLM model to narrate the image
  - action: ai_task.generate_data
    data:
      task_name: wall tablet picture narrator
      entity_id: ai_task.ollama_ai_task_ministral_3_8b
      instructions: >-
        Look at this image and write one tiny, friendly line(s) for a wall
        tablet. Use EXIF context if available: {{ exif_output.stdout if (exif_output is defined and exif_output.stdout is defined) else 'No EXIF data' }}. Try to
        tell from where this picture is place/city based on latitude & longitude
        also and when it was taken if there is EXIF data. Use very simple words.
        If it is a famous landmark or popular art say the name and what it is.
        If you are not sure, say that it is only a guess. Keep it above 200 but
        under 255 characters. No markdown.
      structure:
        caption:
          description: One short wall-tablet caption, fact, or riddle.
          required: true
          selector:
            text: null
      attachments:
        - media_content_id: media-source://media_source/local/collection_snapshot.jpeg
          media_content_type: image/jpeg
    response_variable: picture_story

  # Action 5: Save description to our helper
  - action: input_text.set_value
    target:
      entity_id: input_text.collection_image_story
    data:
      value: >-
        {{ picture_story.data.caption | default('New picture. Tiny robot is thinking.', true) | string | truncate(255, true, '') }}

  # Action 6: Show persistent notification
  - action: persistent_notification.create
    data:
      message: "{{ picture_story.data.caption }}"