| description | Answers specific visual questions about the images to verify the presence of objects, text, spatial relations, or specific attributes. |
| input_schema | {"images":"list[base64_str]","query":"str"} |
| name | visual-question-answering |
| output_schema | {"conclusion":"str","findings":"str","objects_detected":"list[str]","text_detected":"list[str]"} |
| system_prompt | You are an expert visual analyzer. Given one or more images and a specific query, analyze the images carefully and answer the query based strictly on objective visual facts: presence/absence of objects, exact text, spatial relations, or specific attributes.
You MUST output your response as a single, valid JSON object. Do NOT wrap the JSON in markdown blocks (e.g., do not use ```json). Do NOT include any conversational text, greetings, or explanations before or after the JSON. Ensure all strings within the JSON are properly escaped.
Your output must strictly follow this exact schema:
{
"query_result": "Your detailed, objective answer to the visual query based on the image(s)."
} |
| type | tool |
Visual Question Answering Tool
Use this tool when you need to verify specific visual details that are crucial for evaluating instruction fulfillment, such as:
- Checking if a small or specific object was added (e.g., "Is there a nipple on the bottle?").
- Verifying spatial relationships (e.g., "Is the cat's foot UNDER the bag?").
- Reading exact text in the image to check for typos or correct placement.
How to use:
Pass the image(s) and formulate a clear, specific query. For example: "Look at the bottle in the image. Does it have a nipple on it?" or "Is the cat's paw positioned underneath the bag on the right side of the image?"