Llama Guard 4 for Multimodal Safety

Llama Guard 4 is a safety moderation model for both large language and vision language models. It can detect unsafe images and text inputs. To use it, simply input text or images along with text.You can also provide hypothetical LLM outputs separately, as this model can take in entire conversation.
Select categories to EXCLUDE from moderation

Selected categories will NOT be flagged.
Examples
Image Input (Optional) Text Input Model Output Select categories to EXCLUDE from moderation