Slideshow with voiceover

The script and the pictures move together: the timing map is the contract.

Workflow

1. Script and timing map. One line per image, in order, each image holding exactly its line's window.
2. Voice. Generate (generateMedia kind speech) or place each line inside its window. If a line runs long, rewrite it tighter or regenerate faster; never let lines overlap or spill into the next image.
3. Motion. Every photo gets a slow eased push-in or drift of 3 to 8% across its hold, alternating direction, faces and text in the photo kept in frame (production-motion).
4. Captions follow the voice with real word timing, in the language spoken, keywords emphasised as asked (captions skill).
5. Music is a low bed ducked under every line and faded at both ends (musicBed).
6. End on the call to action inside the requested length.

Gotchas

• Generated speech lands at its MEASURED length; a shorter requested duration is not applied. Plan windows from the measured length, not the estimate.
• A voice line cut through a word blocks the run from finishing (speechcut:). Shorten the line's text instead of trimming the audio.
• An image clip pointed at an MP4, a blob: src or a lost upload shows "Media missing" and blocks finishing. Never ship it.
• Keep each caption inside its line's window; a timing change moves the caption with its line.