Analyze images using multimodal LLMs via MakerAI. Describe scenes, answer visual questions, detect objects, extract text (OCR), compare images and generate structured data from visual content.
visionimagemultimodalocrdetectanalyze
MakerAI Pipeline
pipeline: IPAILLMProvider.Complete with image content blocks (multimodal)
providers: anthropic (Claude), openai (GPT-4o), google (Gemini)
Input Parameters
Parameter
Type
Description
operationrequired
string
'describe': general description; 'ask': answer a question about the image; 'ocr': extract all text; 'detect': list objects/people/elements; 'compare': compare two images; 'extract': extract structured data (tables, forms, etc.). One of: describe, ask, ocr, detect, compare, extract.
image_urloptional
string
URL of the image to analyze (http/https or data URI).
image_b64optional
string
Base64-encoded image (alternative to image_url).
image_b64_2optional
string
Second image for compare operation.
questionoptional
string
Question to answer about the image (ask, extract operations).
extract_schemaoptional
object
JSON schema defining the structure to extract from the image (extract operation).
Discovery hint: Install with ppm install mcp-vision or invoke remotely via POST /v1/invoke/mcp-vision on the MCP Service.
PascalAI Usage
uses toolslib;
var Tool := LoadTool('mcp-vision');
var R := Tool.Call(JsonObj(['operation','describe','image_url','https://example.com/photo.jpg']));
Writeln(R['description']);