5.2. Reflection Checkpoint
💡 First Principle: Generation is the opposite arrow from understanding. Understanding takes an image in and reasons about it (image → text); generation takes a prompt in and produces media (text → image, text → video). The exam's vision domain dedicates a whole sub-skill to this production direction — text-to-image, text-to-video, and editing existing media — and it is easy to conflate with understanding if you only think "vision = analyze a picture."
Why care: the official skills outline explicitly lists generating images from text prompts and reference media, generating videos from text prompts and reference media, and editing workflows (inpainting, mask-based edits, prompt-driven modifications, editing generated video). A candidate who studied only captioning and OCR will miss an entire sub-skill. Knowing which direction a requirement points — analyze versus produce — is the first decision.
⚠️ Common Misconception: "An image model is an image model." Generation and understanding are different capabilities pointing in opposite directions. A multimodal understanding model describes a photo you give it; an image generation model produces a new photo from your text. Selecting a generation model to analyze an uploaded image (or vice versa) is a direction mismatch the exam probes.
