Azure AI Content Safety's image moderation capability analyzes images and returns, for each harm category (such as violence, sexual content, self-harm, and hate), a severity level (0-Safe, 2-Low, 4-Medium, 6-High). An application can configure allowed/prohibited severity thresholds per category so the service returns an accept/reject decision, letting harmful uploads be blocked automatically based on severity. Running OCR keyword scanning only detects text embedded in an image and misses purely visual harm; prompt shields defend against prompt-injection attacks rather than classifying image content; and blocklists match specific known terms rather than assessing visual harm severity.
Community Discussion