How to make notification action conditional on message content?

Configuration
1

I’m trying to use Gemini AI to send a notification to my phone if a person is detected by my video doorbell. I’ve got the automation sending images and descriptions to my phone, but I can’t work out how to ONLY send a notification if the AI description (in image_description.response_text) does NOT include the phrase “no people are visible”.
Can anyone tell me how I would need to modify the code below to add this condition?

alias: send_camera_snapshot
sequence:
  - variables:
      file_time: "{{ now().strftime('%Y%m%d_%H%M%S') }}"
  - data:
      filename: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    target:
      entity_id: camera.{{ camera_id }}
    action: camera.snapshot
  - action: llmvision.image_analyzer
    metadata: {}
    data:
      remember: false
      use_memory: false
      include_filename: false
      target_width: 1280
      max_tokens: 100
      temperature: 0.2
      generate_title: false
      expose_images: false
      provider: 01JQM1Y8JRXFCDFA9X5FHTVEZC
      message: >-
        Describe the image in a single sentence. If you see a person, describe
        them. If you see multiple people, give a count of the number of people
        and describe them. Try to determine if they are arriving or leaving and
        state which if you can. 
      image_file: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    response_variable: image_description
  - data:
      message: "{{ camera_id }} alarm - {{image_description.response_text}}"
      title: Alarm triggered
      data:
        image: /local/{{ camera_id }}_snapshot_{{ file_time }}.jpg
        channel: Motion
        ttl: 0
        priority: high
        importance: high
        visibility: public
        actions:
          - action: URI
            title: Open Image
            uri: >-
              https://mydomain/local/{{ camera_id
              }}_snapshot_{{file_time }}.jpg
    action: notify.mobile_app_ian_galaxy_s24_ultra
mode: single
2

Template condition:

{{ "no people are visible" in image_description }}

Sorry.
You need

{{ not "no people are visible" in image_description }}

Obviously.

1 Like
3

Thanks. I’ve tried to incorporate this but am struggling with the syntax. I am getting this error:

Message malformed: Unable to determine action @ data['sequence'][3]

This is the code:

Message malformed: Unable to determine action @ data['sequence'][3]

alias: send_camera_snapshot
sequence:
  - variables:
      file_time: "{{ now().strftime('%Y%m%d_%H%M%S') }}"
  - data:
      filename: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    target:
      entity_id: camera.{{ camera_id }}
    action: camera.snapshot
  - action: llmvision.image_analyzer
    metadata: {}
    data:
      remember: false
      use_memory: false
      include_filename: false
      target_width: 1280
      max_tokens: 100
      temperature: 0.2
      generate_title: false
      expose_images: false
      provider: 01JQM1Y8JRXFCDFA9X5FHTVEZC
      message: >-
        Describe the image in a single sentence. If you see a person, describe
        them. If you see multiple people, give a count of the number of people
        and describe them. Try to determine if they are arriving or leaving and
        state which if you can. 
      image_file: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    response_variable: image_description
  - data:
      message: "{{ camera_id }} alarm - {{image_description.response_text}}"
      title: Alarm triggered
      data:
        image: /local/{{ camera_id }}_snapshot_{{ file_time }}.jpg
        channel: Motion
        ttl: 0
        priority: high
        importance: high
        visibility: public
        actions:
          - action: URI
            title: Open Image
            uri: >-
              https://mydomain/local/{{ camera_id
              }}_snapshot_{{file_time }}.jpg
          - choose:
              - conditions:
                  - condition: template
                    value_template: {{ not "no people are visible" in image_description }}
                sequence:
                  - action: notify.mobile_app_ian_galaxy_s24_ultra
mode: single
4

No…
The condition must be before the notification.
Just a normal condition, no choose.

5

OK I’ve re-arranged it, as below. Still get this error:

Message malformed: Unable to determine action @ data['sequence'][3]

alias: send_camera_snapshot
sequence:
  - variables:
      file_time: "{{ now().strftime('%Y%m%d_%H%M%S') }}"
  - data:
      filename: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    target:
      entity_id: camera.{{ camera_id }}
    action: camera.snapshot
  - action: llmvision.image_analyzer
    metadata: {}
    data:
      remember: false
      use_memory: false
      include_filename: false
      target_width: 1280
      max_tokens: 100
      temperature: 0.2
      generate_title: false
      expose_images: false
      provider: 01JQM1Y8JRXFCDFA9X5FHTVEZC
      message: >-
        Describe the image in a single sentence. If you see a person, describe
        them. If you see multiple people, give a count of the number of people
        and describe them. Try to determine if they are arriving or leaving and
        state which if you can. 
      image_file: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    response_variable: image_description
  - data:
      message: "{{ camera_id }} alarm - {{image_description.response_text}}"
      title: Alarm triggered
      data:
        image: /local/{{ camera_id }}_snapshot_{{ file_time }}.jpg
        channel: Motion
        ttl: 0
        priority: high
        importance: high
        visibility: public
        actions:
          - action: URI
            title: Open Image
            uri: >-
              https://mydomain/local/{{ camera_id
              }}_snapshot_{{file_time }}.jpg
          - condition: template
            value_template: {{ not "no people are visible" in image_description }}
          - action: notify.mobile_app_ian_galaxy_s24_ultra
mode: single
6

Before the notification

alias: send_camera_snapshot
sequence:
  - variables:
      file_time: "{{ now().strftime('%Y%m%d_%H%M%S') }}"
  - data:
      filename: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    target:
      entity_id: camera.{{ camera_id }}
    action: camera.snapshot
  - action: llmvision.image_analyzer
    metadata: {}
    data:
      remember: false
      use_memory: false
      include_filename: false
      target_width: 1280
      max_tokens: 100
      temperature: 0.2
      generate_title: false
      expose_images: false
      provider: 01JQM1Y8JRXFCDFA9X5FHTVEZC
      message: >-
        Describe the image in a single sentence. If you see a person, describe
        them. If you see multiple people, give a count of the number of people
        and describe them. Try to determine if they are arriving or leaving and
        state which if you can. 
      image_file: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    response_variable: image_description
  - condition: template
    value_template: "{{ not 'no people are visible' in image_description }}"
  - action: notify.mobile_app_ian_galaxy_s24_ultra
    data:
      message: "{{ camera_id }} alarm - {{image_description.response_text}}"
      title: Alarm triggered
      data:
        image: /local/{{ camera_id }}_snapshot_{{ file_time }}.jpg
        channel: Motion
        ttl: 0
        priority: high
        importance: high
        visibility: public
        actions:
          - action: URI
            title: Open Image
            uri: >-
              https://mydomain/local/{{ camera_id
              }}_snapshot_{{file_time }}.jpg
mode: single
7

Hmmm…now I get:

Message malformed: template value should be a string for dictionary value @ data['sequence'][3]['value_template']

This is the code:

alias: send_camera_snapshot
sequence:
  - variables:
      file_time: "{{ now().strftime('%Y%m%d_%H%M%S') }}"
  - data:
      filename: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    target:
      entity_id: camera.{{ camera_id }}
    action: camera.snapshot
  - action: llmvision.image_analyzer
    metadata: {}
    data:
      remember: false
      use_memory: false
      include_filename: false
      target_width: 1280
      max_tokens: 100
      temperature: 0.2
      generate_title: false
      expose_images: false
      provider: 01JQM1Y8JRXFCDFA9X5FHTVEZC
      message: >-
        Describe the image in a single sentence. If you see a person, describe
        them. If you see multiple people, give a count of the number of people
        and describe them. Try to determine if they are arriving or leaving and
        state which if you can. 
      image_file: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    response_variable: image_description
  - condition: template
    value_template: {{ not "no people are visible" in image_description }}
  - action: notify.mobile_app_ian_galaxy_s24_ultra
    data:
      message: "{{ camera_id }} alarm - {{image_description.response_text}}"
      title: Alarm triggered
      data:
        image: /local/{{ camera_id }}_snapshot_{{ file_time }}.jpg
        channel: Motion
        ttl: 0
        priority: high
        importance: high
        visibility: public
        actions:
          - action: URI
            title: Open Image
            uri: >-
              https://mydomain/local/{{ camera_id
              }}_snapshot_{{file_time }}.jpg
mode: single
8

Sorry, see edit

9

Thanks for the update. It passed the validation, but the notifications are still being sent even when image_description contains the text “no people are visible”.
Any idea why that might be?

alias: send_camera_snapshot
sequence:
  - variables:
      file_time: "{{ now().strftime('%Y%m%d_%H%M%S') }}"
  - data:
      filename: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    target:
      entity_id: camera.{{ camera_id }}
    action: camera.snapshot
  - action: llmvision.image_analyzer
    metadata: {}
    data:
      remember: false
      use_memory: false
      include_filename: false
      target_width: 1280
      max_tokens: 100
      temperature: 0.2
      generate_title: false
      expose_images: false
      provider: 01JQM1Y8JRXFCDFA9X5FHTVEZC
      message: >-
        Describe the image in a single sentence. If you see a person, describe
        them. If you see multiple people, give a count of the number of people
        and describe them. Try to determine if they are arriving or leaving and
        state which if you can. 
      image_file: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    response_variable: image_description
  - condition: template
    value_template: "{{ not 'no people' in image_description }}"
  - action: notify.mobile_app_ian_galaxy_s24_ultra
    data:
      message: "{{ camera_id }} alarm - {{image_description.response_text}}"
      title: Alarm triggered
      data:
        image: /local/{{ camera_id }}_snapshot_{{ file_time }}.jpg
        channel: Motion
        ttl: 0
        priority: high
        importance: high
        visibility: public
        actions:
          - action: URI
            title: Open Image
            uri: >-
              https://mydomain/local/{{ camera_id
              }}_snapshot_{{file_time }}.jpg
mode: single
10

Post what image description is.
Maybe that is in the trace. Else you need to notify the whole image description and post it here

11

This was the latest response:

2025-04-01 12:33:11.702 INFO (MainThread) [custom_components.llmvision.providers] Response data: {'candidates': [{'content': {'parts': [{'text': "Here is a description of the image: Eye-level view of a residential street scene captured by a Reolink video doorbell, showing a row of brick houses on a sunny day; several cars are parked along the street, and a dark-colored Audi is parked near the camera's location.  There are no people visible in the image."}], 'role': 'model'}

(I changed the logic to look for “no people”)

12

So you need this I think:

value_template: "{{ not 'no people' in image_description.candidates[0]content.parts[0]text }}"
13

That gives this error:

Message malformed: invalid template (TemplateSyntaxError: expected token 'end of print statement', got 'content') for dictionary value @ data['sequence'][3]['value_template']

alias: send_camera_snapshot
sequence:
  - variables:
      file_time: "{{ now().strftime('%Y%m%d_%H%M%S') }}"
  - data:
      filename: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    target:
      entity_id: camera.{{ camera_id }}
    action: camera.snapshot
  - action: llmvision.image_analyzer
    metadata: {}
    data:
      remember: false
      use_memory: false
      include_filename: false
      target_width: 1280
      max_tokens: 100
      temperature: 0.2
      generate_title: false
      expose_images: false
      provider: 01JQM1Y8JRXFCDFA9X5FHTVEZC
      message: >-
        Describe the image in a single sentence. If you see a person, describe
        them. If you see multiple people, give a count of the number of people
        and describe them. Try to determine if they are arriving or leaving and
        state which if you can. 
      image_file: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    response_variable: image_description
  - condition: template
    value_template: "{{ not 'no people' in image_description.candidates[0]content.parts[0]text }}"
  - action: notify.mobile_app_ian_galaxy_s24_ultra
    data:
      message: "{{ camera_id }} alarm - {{image_description.response_text}}"
      title: Alarm triggered
      data:
        image: /local/{{ camera_id }}_snapshot_{{ file_time }}.jpg
        channel: Motion
        ttl: 0
        priority: high
        importance: high
        visibility: public
        actions:
          - action: URI
            title: Open Image
            uri: >-
              https://mydomain/local/{{ camera_id
              }}_snapshot_{{file_time }}.jpg
mode: single
14

Sorry missed the dots after [0]in two places. Make them [0].

15

Thanks. That passed validation but I get this error in the log when it’s executed:

2025-04-01 14:06:41.122 INFO (MainThread) [custom_components.llmvision.providers] Response data: {'candidates': [{'content': {'parts': [{'text': 'Here is a description of the image: A wide shot of a residential street on a sunny day, captured by a Reolink Video Doorbell WiFi, shows a row of brick houses, several parked cars, and a person in dark clothing walking away from the camera.'}], 'role': 'model'}, 'finishReason': 'STOP', 'avgLogprobs': -0.1503448486328125}], 'usageMetadata': {'promptTokenCount': 314, 'candidatesTokenCount': 54, 'totalTokenCount': 368, 'promptTokensDetails': [{'modality': 'IMAGE', 'tokenCount': 258}, {'modality': 'TEXT', 'tokenCount': 56}], 'candidatesTokensDetails': [{'modality': 'TEXT', 'tokenCount': 54}]}, 'modelVersion': 'gemini-1.5-flash-latest'}
2025-04-01 14:05:36.608 WARNING (MainThread) [homeassistant.helpers.script] Error in 'condition' evaluation:
In 'template' condition: UndefinedError: 'dict object' has no attribute 'candidates'

alias: send_camera_snapshot
sequence:
  - variables:
      file_time: "{{ now().strftime('%Y%m%d_%H%M%S') }}"
  - data:
      filename: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    target:
      entity_id: camera.{{ camera_id }}
    action: camera.snapshot
  - action: llmvision.image_analyzer
    metadata: {}
    data:
      remember: false
      use_memory: false
      include_filename: false
      target_width: 1280
      max_tokens: 100
      temperature: 0.2
      generate_title: false
      expose_images: false
      provider: 01JQM1Y8JRXFCDFA9X5FHTVEZC
      message: >-
        Describe the image in a single sentence. If you see a person, describe
        them. If you see multiple people, give a count of the number of people
        and describe them. Try to determine if they are arriving or leaving and
        state which if you can. 
      image_file: /config/www/{{ camera_id }}_snapshot_{{ file_time }}.jpg
    response_variable: image_description
  - condition: template
    value_template: >-
      {{ not 'no people' in
      image_description.candidates[0].content.parts[0].text }}
  - action: notify.mobile_app_ian_galaxy_s24_ultra
    data:
      message: "{{ camera_id }} alarm - {{image_description.response_text}}"
      title: Alarm triggered
      data:
        image: /local/{{ camera_id }}_snapshot_{{ file_time }}.jpg
        channel: Motion
        ttl: 0
        priority: high
        importance: high
        visibility: public
        actions:
          - action: URI
            title: Open Image
            uri: >-
              https://mydomain/local/{{ camera_id
              }}_snapshot_{{file_time }}.jpg
mode: single
16

In that case I’m lost.
I don’t know why this does not work.

17

OK thanks for your help anyway.

18

Is it this integration?

It seems according to this blueprint you should use image_description.response_text

Not sure though, but worth a try.

1 Like
19

Excellent - that seems to be working! Many thanks for your help with this.