Resolve provider capabilities before generation

Do not create a paid video request from a generic template or model-family memory. Before the first video call for a production, read the selected endpoint's current official API reference. For fal-hosted models use that exact endpoint's fal API page; for direct Google models use the current Google Gemini/Veo documentation. A model list discovers candidates; the endpoint schema controls the request. Models and higher-level workflows are different contracts. Do not infer one from the other.

Read these bundled references when their phase begins:

• references/capability-record.md before resolving or refreshing a model.
• references/continuity-planning.md before writing the shot list or request.
• references/execution-review.md before spending, retrying, assembling, or presenting outputs.
• video-model-selection/references/cost-guidance.md when comparing models or giving a user a costed choice.