Touge spoken storyboard

TouGe Spoken Storyboard

Task

Generate STORYBOARD.md from prepared root-level design assets and SCRIPT.srt.

Assume the user has already prepared the visual design assets, either manually or with any compatible design workflow. Do not require another Skill, redesign the look, create HTML compositions, render video, or run a long discovery interview. Convert the timed口播字幕 into a scene-by-scene storyboard that HyperFrames can implement.

Startup Gate

Use the current working directory as the project root. Scan only files directly in that root; do not recurse into subfolders.

Required root-level inputs:

• DESIGN.md
• design-tokens.css
• at least one *.html or *.HTML
• SCRIPT.srt

If any required input is missing, stop and ask the user to add it. Do not create STORYBOARD.md with placeholders for missing required assets.

If all inputs exist, read:

• DESIGN.md
• design-tokens.css
• every root-level *.html / *.HTML filename and enough content to identify which storyboard components or layout demos they represent
• SCRIPT.srt

One-Time Confirmation Card

After the startup gate passes, ask one compact confirmation message only. Include defaults and allow the user to override in one reply.

Ask for:

1. 是否真人出镜：default 真人出镜 + B-roll 信息层
2. 平台/画幅：default B站/知识区横屏，16:9，1920x1080
3. 节奏密度：default 教程型，中等密度
4. 输出质量：default high
5. 是否添加可见字幕：default 不添加字幕

If the user does not answer or says to use defaults, continue with:

• talking-head + B-roll
• 16:9
• 1920x1080
• 30fps
• tutorial / knowledge-video pacing
• high
• no visible subtitles

Do not continue asking follow-up questions unless a required asset is missing or the user explicitly asks to refine the choices.

Workflow

1. Validate required assets using the startup gate.
2. Run the one-time confirmation card.
3. Load references/storyboard-rhythm-rules.md before deciding scene density and duration.
4. Load references/srt-scene-breakdown.md before segmenting SCRIPT.srt and deciding generate vs skip-generation.
5. Load references/storyboard-component.md only after deciding a Scene needs generated visuals.
6. Load references/storyboard-output-rules.md and templates/storyboard-template.md before writing STORYBOARD.md.
7. Write STORYBOARD.md to the current project root.
8. Before finishing, self-check that generated Scenes have valid component IDs, composition file names, and visual_track 0-4; skipped Scenes must have none of those.

Storyboard Rules

• Preserve SRT timing. Every Scene must include start time, end time, and the exact subtitle lines covered.
• SCRIPT.srt is always the timing and narration source, but visible subtitles are optional.
• Default subtitle policy is 不添加字幕; if the user does not explicitly request visible subtitles, mark the storyboard as no subtitle overlay.
• If the user chooses no subtitles, do not plan subtitle tracks, caption components, karaoke text, or persistent on-screen captions.
• Do not map one SRT cue to one Scene by default. Merge or split by video form, information density, and semantics.
• Default tutorial pacing: combine related SRT cues into 2-6 second Scenes.
• Hook / short-form pacing can use 0.8-2.5 second Scenes when the user selects a short-video format.
• Documentary / slow pacing can use 3-8 second Scenes when the user selects a slower form.
• Decide whether each Scene should generate visual content before choosing components.
• Talking-head A-roll is reserved original video space only; do not generate cards, labels, numbers, waveforms, loading states, or demo placeholders inside that area.
• Generated visual content belongs in B-roll information layers and future HyperFrames sub-compositions.
• For generated Scenes, assign composition_src as compositions/scene-XX-kebab-name.html and visual_track from 0 to 4.
• The 5 visual tracks apply only to generated visual compositions; A-roll original video, audio, and subtitles are outside those tracks.
• Functional demos, live UI walkthroughs, code/terminal operations, and pure talking-only bridges should usually be skip-generation while preserving their SRT time range.
• Pure B-roll generation may use broll-*, icons.*, or illustrations.* components.
• If no visual task is justified, use skip-generation; use broll-abstract.placeholder only when a generated visual is needed but no component fits.
• On-screen text must come from the current Scene SRT, faithful extraction/summary/structure from it, or adjacent SRT for continuity. HTML demo sample text is never storyboard content.
• Use root-relative file paths in the storyboard.
• Do not invent missing assets. Mark optional non-blocking media as 待补充 only if STORYBOARD.md can still guide implementation.

References

• templates/storyboard-template.md: output skeleton for STORYBOARD.md.
• references/storyboard-output-rules.md: field constraints, consistency checks, and delivery checklist.
• references/storyboard-rhythm-rules.md: rhythm, duration, transition density, and scene count rules.
• references/srt-scene-breakdown.md: how to turn SCRIPT.srt into Scenes.
• references/storyboard-component.md: the only valid component/layout ID catalog.