Gemini media generation

Paths below use {base} as shorthand for this skill's base directory, provided automatically when the skill loads.

This skill turns a rough user request into a high-quality engine brief, picks the right model and effort level, then delegates the actual media work to Google's Gemini models: Nano Banana for stills, Veo for motion. The skill's job is to interrogate the request, expand it into a precise prompt, recommend the model/effort, lock the correct format for the channel, get explicit sign-off, and only then call the worker scripts.

This skill is deliberately brand-agnostic. It carries no house style of its own. If the user has a brand voice or visual system, ask for it and fold it into the brief — never assume one.

Core principle: brief first, generate second

Never send a raw user request straight to an engine. A one-line prompt produces generic output. The core job of this skill is prompt-building — turning "make an image of a coffee shop" into a precise, art-directed brief the engine can execute well. You do that by asking a few sharp questions (only when needed), then expanding the answer into a dense, technical prompt. Always build the prompt, always confirm, before spending an engine call.

Runtime prerequisites

• A Google Gemini API key in the environment as GEMINI_API_KEY, on a paid/billing-enabled project (image and video models are not on the free tier). Veo access enabled for video.
• Python 3 with the google-genai package available to the worker scripts.

If the key is missing, say exactly that and stop.