Content that exists not as a fixed bundled work but as a *generative or interactive system* that produces media outputs across multiple modalities in response to inputs. A trained AI model is multimodal media. A text-to-image generator is multimodal media. A voice-cloning system that produces audio outputs from text prompts is multimodal media. An interactive virtual environment is multimodal media. An agentic system that chooses outputs across text, image, audio, and code is multimodal media. The legal question here is fundamentally different from the multimedia case. For multimedia, the question is "which rights subsist in this fixed bundled work?" For multimodal media, the questions are "which rights subsist in the *outputs* of a generative system?", "which rights subsisted in the inputs that trained the system?", and "what relationship exists between the training inputs and the generative outputs?". CDPA 1988, EU AI Act Articles 50 and 53, and DUA Act 2025 are all currently working through these questions; the law is not settled. In the CIP framework, content classified as `Multimodal-Media` (per `CIP-Media-Class`, Enum 5 of CIP Classifications v1.0) requires the operator to declare the system's modality scope: which modalities it accepts as inputs, which modalities it produces as outputs, and which modalities were represented in its training data. These three are not necessarily the same — a text-to-image system accepts text but produces images and was trained on text-image pairs. The disclosure is operationally similar to the EU AI Act Article 53(1)(d) training-data summary requirement but extends beyond it to cover input and output modalities explicitly. The framework's Risk Gravity Model (Page 12f) implicitly addresses multimodal-media scenarios — the "Voice clone, brand client" and "Style appropriation, US output" scenarios both involve generative systems producing outputs across modalities — but did not previously name the category. From v2.8 onward, scenarios are explicitly classified by `CIP-Media-Class` to make the analysis tractable.