Documentation: Compare Images with AI Node¶

Overview¶

The Compare Images with AI Node is an action node that uses artificial intelligence to compare a set of reference images against a set of current images and determine whether there are matches, changes, or relevant differences. Like the Analyze Image with AI node, it allows you to choose the model/provider flexibly.

In IoT and security environments, it is ideal for detecting visual changes: comparing the "normal" state of an area (reference image) with its current state to detect missing or appearing objects, changes on a production line, intrusions, or for visual quality control.

When to use this node?¶

Use this node when you need to:

Detect changes between a reference image (normal state) and the current image from a camera.
Perform visual quality control by comparing a product/scene against a reference image.
Identify objects that are missing or have appeared in a monitored area.
Validate matches against a configurable pass threshold.

Node Configuration¶

The form is organized into three selectable sections: Model, Images, and Advanced. It also includes the JSON Editor tab.

Model section of the Compare Images with AI node

Section: Model¶

1. API Key *Required¶

The provider's API key (protected field). Enables the model browser.

2. Model *Required¶

Searchable model selector by name or provider (IDs in the format provider/model).

Section: Images¶

3. Reference Images *Required¶

The base images (normal/reference state) against which the comparison is made. Supports automation variables.

4. Comparison Images *Required¶

The current images to evaluate (e.g., the recent camera snapshot, {{get_snapshot_node.url}}).

5. System Prompt *Optional¶

System instructions for the model (defines role/behavior).

6. Prompt *Required¶

Describes what to compare and what constitutes a match (e.g., "Indicate if any object is missing or there are relevant changes in the area").

Images section of the Compare Images with AI node

Section: Advanced¶

Pass Threshold: Value (default 0.5) that defines the threshold above which a match is considered to exist.
Temperature: Creativity/randomness of the response (0 to 2).
Max Tokens: Maximum token limit for the response.

JSON Editor View¶

JSON Editor view of the Compare Images with AI node

JSON Structure (Input Parameters)¶

{
  "api_key": "sk-or-xxxxxxxxxxxxxxxx",
  "model_id": "openai/gpt-4o",
  "reference_images": [
    "https://demo02.netsocs.com/.../referencia.jpg"
  ],
  "comparison_images": [
    "{{get_snapshot_node.url}}"
  ],
  "prompt": "Compare the reference image with the current one. Indicate if any object is missing or there are relevant changes in the area.",
  "system_prompt": "You are a visual quality control system."
}

JSON Fields¶

Field	Type	Description
`api_key`	string	Provider API key.
`model_id`	string	Model ID (`provider/model`).
`reference_images`	array (string)	URLs of the reference (base) images.
`comparison_images`	array (string)	URLs of the current images to compare.
`prompt`	string	What to compare and what constitutes a match.
`system_prompt`	string	(Optional) System instructions.
`pass_threshold`	number	(Optional, Advanced) Pass threshold (default `0.5`).
`temperature`	number	(Optional, Advanced) Response creativity (0–2).
`max_tokens`	number	(Optional, Advanced) Token limit.

Output: Where the node's data comes from¶

The comparison result (including whether there is a match according to the threshold, and the model's explanation) is available in the node's output and can be used in downstream nodes with {{node_key}} to make decisions (e.g., escalate an alert if a change is detected).

Usage Examples¶

Example 1: Detection of a missing object in an area¶

Use case: Compare the reference image of a shelf/warehouse with the current snapshot to detect if any material is missing.

Reference Images: Reference image of the complete area.
Comparison Images: {{get_snapshot_node.url}}
Prompt: Compare the reference image with the current one. Indicate if any object is missing or there are relevant changes in the area.

(see JSON structure above)

Example 2: Visual quality control¶

Use case: Compare a product on the production line with a reference image to validate it meets the standard, using a pass threshold.

Pass Threshold: 0.7

Validation and Errors¶

Condition	Common cause / fix
Missing images	You must provide at least one reference image and one comparison image.
Authentication error	The API Key is invalid or lacks permissions for the chosen model.
URLs not working	Make sure the image URLs are publicly accessible.

Best Practices¶

Stable reference images: Use consistent reference images (same camera, angle, and lighting) for reliable comparisons.
Define the match clearly: In the prompt, describe precisely which changes matter and which should be ignored.
Adjust the threshold: Calibrate pass_threshold according to the desired sensitivity.
Chain with snapshot capture: Typical pattern: Get snapshot → Compare Images with AI → condition/notification.