crazyooo Talking Head Video Producer
Create a finished video in which a real presenter carries the argument and supporting visuals clarify, prove, compare, or pace the spoken content. The skill is portable: do not assume a particular workbench, editor, project API, or directory layout already exists.
Establish the production
Accept a script, raw presenter footage, or both. Determine the target platform, aspect ratio, desired length, language, source assets, brand materials, and delivery format from the request and available files. Infer ordinary defaults when safe: use 1920x1080, 30 fps, and the source audio language when the platform is unspecified.
A finished real-person video requires readable presenter footage. If it is missing, create a shoot-ready script, framing plan, performance notes, and production plan, then request the footage. Do not silently replace a real presenter with an avatar. Generate or clone a voice or likeness only when the user explicitly requests it and the required authorization and cost approval are present.
If the user supplies a brand system, follow it. Otherwise offer or select one of the visual presets in references/presets.md. Treat a preset as a starting system, not a rigid template.
Production order
For end-to-end work, read references/workflow.md before editing. The final speech timeline is the anchor: finish semantic A-roll decisions before committing graphics, B-roll, music, or captions.
Use complete semantic units when removing retakes, false starts, fillers, and pauses. Preserve useful setup, connective words, tone, and natural breathing. Prefer a conservative local cut over an edit that changes meaning or sounds stitched together.
Plan every auxiliary visual against a concrete purpose:
- Keep the presenter when expression, trust, judgment, or emotion carries the moment.
- Use an overlay when the presenter and the information should be understood together.
- Use a full-frame explanation when the viewer must inspect a process, comparison, interface, quotation, diagram, or evidence.
- Use B-roll to provide evidence, establish context, or cover a necessary cut—not as unrelated decoration.
- Avoid repeating the full spoken sentence as on-screen text. Use short labels, values, relationships, and emphasis.
Inspect actual frames before placing overlays. Protect the face, hands, demonstrated objects, microphone, burned-in text, captions, and platform interface zones. Check entry, middle, and exit frames; a clear first frame does not prove the subject remains clear.
Default to review checkpoints after the speech edit and after a representative visual sample. If the user explicitly requests uninterrupted end-to-end execution, continue without routine pauses, but still stop for missing source media, external spending, rights or identity questions, and destructive actions.
Truthful completion
Use the available transcription, media inspection, graphics, image, audio, editing, and rendering capabilities. Follow the instructions of any specialized tool or skill actually used. Never claim to have watched, rendered, synchronized, or verified media unless that operation was performed.
Before delivery, read and apply references/quality-gates.md. Deliver the playable final file when rendering is available, plus the production artifacts needed to revise it. If a capability is unavailable, preserve completed artifacts, state the exact blocker, and do not label a plan or mockup as the final video.