Course Video Digital Human Instructor

Problem

Making instructor-led course videos is often blocked less by script quality than by workflow: a licensed instructor portrait, final lesson scripts, voice rights, and video models that accept specific image and audio formats must all line up. This skill targets that gap: it turns approved scripts and one authorized instructor portrait into digital-human talking-head course videos, delivered in lesson order.

How it works

• Input freeze: Check the portrait's MIME type, width, height, aspect ratio, byte size, and alpha channel, then create a lesson ledger. Scripts, likeness and voice rights, and pronunciation tables must be explicit; if repeated names, terms, or abbreviations appear without a pronunciation table, the workflow stops to collect readings.
• Voice selection: Use a catalog voice or, after consent and balance confirmation, clone a voice. Cloning goes through a confirmation card before beatra.voices.clone, then freezes the returned voice_id.
• TTS and video compatibility: Query TTS and image_to_video capabilities, compare the real audio mime_type, duration, and size against accepted constraints, and split the pilot lesson at sentence or paragraph boundaries into supported segments.
• Lesson delivery: Each audio segment must be confirmed before video generation. The video length is chosen as the smallest supported integer duration that is not shorter than the real audio, and final outputs are delivered by lesson and segment order with real size, duration, and usage.

Boundaries

It does not replace audio-only course packages, single talking-avatar clips, or slide parsing. Slide content only matters after the user extracts speakable text into the script. Likeness, voice, and billing authorization must be confirmed first; if formats are incompatible or balance is insufficient, the workflow stops before paid calls instead of silently changing formats or guessing outcomes.