Generate a batch of storyboard images in one call (async). No user auth.

Items may reference other items in the SAME batch, so a first_frame can reuse the character and product images generated alongside it. The worker runs in dependency order — anything with no references goes first, then whatever depended on it.

Headers: x-api-key: <STORYBOARD_API_KEY> (or ?api_key=<STORYBOARD_API_KEY>)

Body: storyboard_id: your run identifier — optional, echoed back and stored items[]: angle: angle slug, e.g. "a1" beat: beat id, e.g. "b01"; null for angle-level (char/product) slot: char | product | first_frame | end_frame — also the filename ratio: "9x16" | "1x1" | "4x5" … (URL-safe, no colons) model: nanobanana | nanobanana_pro | nanobanana_2 | seedream | gpt_image_2 prompt: full generation prompt references: keys of other items in this batch to use as reference images, in @image1..@imageN order. Key is "{angle}/{slot}" or "{angle}/{beat}/{slot}".

Each result is saved to: storyboard/{task_id}/{angle}/{slot}/{ratio}/{slot}.png (beat null) storyboard/{task_id}/{angle}/{beat}/{slot}/{ratio}/{slot}.png (beat set)

Returns: dict: task_id, storyboard_id, accepted, status