slide-video-producer

Generate one final slide image per storyboard slide using generate_image or edit_image, create one narration clip per approved segment with generate_speech, then build segment videos with create_video_from_image_and_audio and concatenate them in storyboard order.

Select one narrator identity before generating the first clip and keep the same voice, model route, language target, and generation config for every clip. Use prompt-level bracketed cues like [warm, conversational] for local delivery changes instead of switching voices.

No. Final slide images must be created with generate_image or edit_image only. Scripts may prepare source assets, convert formats, or assemble video, but Python/PIL-only, HTML/SVG-only, and script-only rendering are disallowed routes for final slide visuals.

Image generation and editing calls must run one at a time with a 60-second sleep after each result, failure, or timeout. This serial pacing prevents concurrent dispatch issues and ensures every attempt is inspected and logged before the next call.

Production is blocked. The package is routed back to the narration_script_reviewer or presentation_director, since the skill requires a passed narration review report, a final script, and an explicit narration-to-slide mapping before any media generation begins.

Create subtitles.srt only when requested, useful for accessibility, or required by the delivery context. Subtitles must match the approved narration, exclude non-spoken cue tags, and use a conservative bottom-safe style that does not cover slide content.