Product Introduction Video Generator

Problem It Solves

Turning a product script and a set of images into a publish-ready video usually requires manual editing, voiceover, subtitle timing, and presenter alignment. This skill targets that repetitive workflow: provide a script, images, a presenter photo, or voice samples, and receive a subtitled MP4 with narration.

How It Works

Instead of generating the whole video with a large model, it breaks the pipeline into controllable steps:
- Script-driven: each script segment maps to one image; starting with “口播：” enables presenter mode.
- Asset synthesis: uses the API key in config.json for auth and calls its own platform APIs for voice, uploads, and generation tasks.
- Voice options: supports edge-tts male/female voices, or clones “my voice” from a public or local audio file.
- Subtitles and visuals: supports soft subtitles, hard burn-in subtitles, or none; missing images can fall back to text-to-image, with a local Pillow fallback when needed.
- Outputs: final video, separate SRT subtitles, and per-segment audio for further editing. It also includes fallbacks: presenter timeouts can fall back to image video, and voice failures can fall back to edge-tts.

Boundaries

Best for product demos, feature explainers, legal, or corporate short videos. If the user only asks to “generate a video,” the skill still needs a script. Presenter mode requires a host photo, and image count mismatches should be treated as errors or fixed before running.