Principles
- One video covers only one core judgment.
- Give conflict, result, benefit, or a specific scenario in the first 3 seconds; prioritize second person (“you...”) or a repeatable motif sentence to pull viewers in.
- Subtitles must stand on their own without sound: with sound off and only subtitles visible, it must still be understandable.
- Spoken narration must be sayable; do not cut a long article into subtitles.
- Visuals carry the tasks of proof, demonstration, comparison, or emotion, and do not repeat the subtitles.
- Prioritize real human voices, product screen recordings, and real evidence; do not default to digital humans or animation.
- A digital human is a production mode after the user actively chooses it, and there must be rights to use the avatar and voice; the user is responsible for manually uploading, creating, and selecting sensitive materials in the chosen digital human platform, and retaining human preview and publish confirmation.
Confirm Inputs
Check: target platform, audience, content goal, duration, on-camera mode, available materials, core judgment, action, and content that must not be shown. When information that would change the plan is missing, ask at most 3 questions.
Video Production Modes
方案模式: When there is no original footage, no confirmation that digital human platform materials are ready, or no clear authorization for a finished cut, deliver a filmable plan and do not generate a finished video without authorization.真人剪辑模式: When the user provides original footage and explicitly requests editing, use the video, subtitle, and audio capabilities available in the current environment.数字人制片模式: When the user explicitly chooses a digital human, confirms portrait and voice rights, and has already manually uploaded, created, and selected the digital human and voice on the chosen platform, follow digital-human-production.md to generate digital human explainer segments, then complete visual parsing, subtitles, the person window, and the vertical finished video.动画解说制片模式: When the user explicitly wants an animated finished video without an on-camera person, and the runtime discovers voiceover, image, and video composition capabilities, follow animated-explainer-production.md and explainer-narrative.md to complete the narrative skeleton, synthesized voiceover, visual assets, shot composition, music licensing, and mixing, and produce the main version first.
The animated explainer mode is likewise discovered by the runtime: it can be an installed animation production Skill, an editing software project template, or a user-specified video tool; the core workflow does not hard-code any particular image or voiceover service. Before paid generation, explain the estimated cost to the user.
The generation tool for digital human mode is discovered by the runtime. If the current environment provides Topview MCP, after the user confirms that the digital humans and voices in Topview are ready, it can be used as an adapter for generating digital human segments; this Skill does not call upload credentials or cloning interfaces on behalf of the user to upload the original avatar or recordings. The core workflow does not hard-code any vendor API. When generation or synthesis capabilities are unavailable, downgrade to a digital human storyboard, visual mapping, subtitle script, asset list, and a manually completable publishing package.
Multi-platform Versions
When a finished video will be published to multiple platforms, first determine the main version (YouTube is usually 16:9, vertical platforms are usually 9:16), then according to video-platform-versions.md Derive other versions: landscape-to-portrait (native re-layout or landscape-to-portrait packaging), burned-in subtitles or SRT, covers for each platform, AI labeling, and music attribution. Use video-release-matrix-template.md to record each version and release field.
Structure
Choose one main structure:
- Problem → Method → Example.
- Result → Process → Lesson.
- Conflict → Choice → Consequence.
- Demonstration → Explanation → Applicable boundaries.
End with one memorable point and one main action.
Storyboard
Use storyboard-template.md. For each shot, clearly write:
- Time and duration.
- Shot size, visuals, character actions, or screen-recording operations.
- Voiceover, subtitles, and on-screen text.
- Evidence, B-roll, sound effects, BGM, and transitions.
The sum of all shot durations must equal the target duration. Subtitles and voiceover facts must be consistent.
Output package
Deliverables:
- 3 hooks and the recommended version.
- Full voiceover script.
- Storyboard table or screen-recording checklist.
- Subtitle script.
- Cover title and visual description.
- Publishing title, body, topics, and BGM suggestions.
- Shooting or material gaps.
Additional deliverables for digital human production mode: portrait and voice rights confirmation, platform manual upload and material readiness status, digital human generated clips, image parsing and character window specifications, subtitle files or subtitle text, pronunciation glossary, final video technical specifications, AI/digital human disclosure requirements, and acceptance results. The delivery package does not include original avatars, recordings, or platform private identifiers.
Additional deliverables for animated explainer production mode: the main version and derived versions for each platform, subtitle files, covers for each platform, music licensing and attribution text, pronunciation glossary, fact source list, technical specifications such as loudness and aspect ratio, production project directory, and rebuild commands.
For multi-platform publishing, deliver a publishing matrix; uploading and publishing are handed to self-media-video-publisher, and per-platform authorization must be followed.
When the user has not provided original footage, default to stopping at a shootable plan; however, when the user explicitly selects digital human production, completes rights confirmation, and confirms that the digital human and voice have been manually uploaded and selected on the chosen platform, you may continue to discover and invoke the digital human, video, subtitle, and audio tools in the current environment.