Image Analyze

Overview

The Image Analyze node accepts an image input and runs it through a configured vision model, returning structured descriptions, classifications, or extracted metadata. Use it in battle workflows to evaluate visual outputs from contenders or to extract scene data for downstream decision nodes. Credentials for the chosen provider must be configured via the workspace's BYOK settings before the node executes. If analysis fails (e.g. unsupported format, model error), the node routes to its error output with a typed error payload.

Configuration

Field	Type	Required	Description
provider	enum	Yes	Vision provider to use. Accepted values: openai, anthropic, google, azure_openai.
model	string	Yes	Model identifier for the chosen provider (e.g. gpt-4o, claude-3-5-sonnet-20241022, gemini-1.5-pro-vision).
prompt	string	Yes	Instruction sent to the model describing what to analyze or extract from the image (e.g. 'Describe the composition and dominant colors.').
output_format	enum	No	Shape of the returned result. Accepted values: text (default), json. Use json to receive a structured object suitable for downstream nodes.
max_tokens	number	No	Upper limit on tokens in the model response. Defaults to 1024.
detail_level	enum	No	Image resolution hint passed to the provider where supported. Accepted values: auto (default), low, high.

Inputs

Port	Type	Description
image	string or binary buffer	The image to analyze. Accepts a URL (http/https or signed storage URL), a base64-encoded data URI, or a binary Buffer.
prompt_override	string	Optional. When present, replaces the static prompt from config at runtime. Useful for dynamic analysis instructions from upstream nodes.

Outputs

Port	Type	Description
output	object	Analysis result from the model. Contains at minimum a text field (string). When output_format is json, the model's parsed JSON object is merged into this payload under a data key.
error	object	Emitted when analysis fails. Contains code (string), message (string), and optionally provider_error (raw upstream error object).

Example

{
  "nodeType": "image_analyze",
  "config": {
    "provider": "anthropic",
    "model": "claude-3-5-sonnet-20241022",
    "prompt": "Identify all visible UI elements and describe the layout structure. Return a JSON object with keys: elements (array), layout (string), dominant_colors (array).",
    "output_format": "json",
    "detail_level": "high"
  }
}