Semantic image

The Semantic image check in SUPERWISE® moves beyond technical file attributes to understand the actual context and category of visual data. This rule uses advanced computer vision to classify images into specific types—such as Photographs, Artworks, Document Scans, Screenshots, Diagrams, and even Memes—allowing users to strictly control the nature of visual interactions. By choosing to either include or exclude specific categories, organizations can prevent irrelevant or inappropriate content from being processed by their agents. For example, a professional document-processing agent can be configured to only "Include" Document Scans while "Excluding" Memes or Artworks, significantly reducing the risk of hallucination or logic errors caused by out-of-domain inputs. For cases that fall outside standard definitions, the Other category ensures that even unconventional visual data can be captured and managed. This semantic layer of protection ensures that your AI agents remain focused on relevant visual tasks, maintaining high professional standards and operational efficiency in every interaction.

Using the SDK

from superwise_api.models.guardrails.guardrails import ImageSemanticsGuard

image_semantics_rule = ImageSemanticsGuard(name="Image Semantics Guard", image_categories=["Photograph","Meme"], filter_mode="include")