Question

Your team runs a public marketing image generator built on Azure OpenAI. Legal requires that any generated image containing prohibited symbols (for example, competitor logos or certain regulated iconography) be automatically blocked before delivery to the user, and the block decision must be logged for audit. The prohibited symbol set changes frequently and is maintained by the compliance team. Which approach best enforces this visual policy rule while remaining maintainable?

Accepted Answer

Add a post-generation step that runs Azure AI Vision / Content Understanding to detect and classify visual elements against a compliance-managed reference list, and reject and log any image matching a prohibited symbol before returning it.

Answer

Rely solely on the Azure OpenAI content filter severity thresholds to reject any image containing prohibited symbols.

Answer

Append the prohibited symbol names to the negative section of the generation prompt so the model avoids producing them, and return all generated images directly.

Answer

Fine-tune the image generation model on examples that exclude prohibited symbols and skip any downstream inspection.