2.4.1. Image and Video Generation
💡 First Principle: Image and video generation are production tasks (text prompt → media output), distinct from vision understanding (media input → text/analysis). Same modality, opposite direction — and the exam cares which direction a requirement runs.
Foundry's catalog includes image-generation models and video models (the Sora family supports text-to-video and early video-to-video transformation). You prompt with a description and optional parameters (size, style, count) and receive generated media. For agentic scenarios, an agent can call image generation as a tool to produce a visual as part of a larger task.
⚠️ Exam Trap: A requirement to "analyze what's in this photo" is understanding (vision domain), while "create an illustration of X" is generation (this domain). Picking an image-generation model to answer a "describe this image" question — or vice versa — is a direction error.
Reflection Question: Why is "summarize the contents of this uploaded image" handled by a different capability than "generate an image of a summary," even though both involve images and text?
