Video Script
This Skill only owns the video production definitions within video/script/. Upstream novels, world settings, chapters, full character profiles, client materials, or other sources remain under their original authoritative sources. Only the conversions and choices made for the performance, scenes, and shot production of the current video are stored here.
1. Read Upstream Sources and VIDEO.md
First, read VIDEO.md and the user-specified upstream sources. When the source is a novel, chapter, or other long-form narrative, first read references/ADAPTATION.md. Only extract the facts, character relationships, plots, scenes, products, or limitations truly needed for the current video. Keep pointers; do not copy the complete upstream content into the video script layer.
If there are conflicts or missing key facts between sources, first identify the gaps. Do not fabricate upstream facts just to make the video script complete.
2. Create a Minimal Video Script Layer
The responsibilities of the four common script files remain separated according to references/SCRIPT-FILES.md. Avoid copying visual design, cinematography, and generation parameters back into the script layer. When actually writing the files, you can follow the minimal templates in references/SCRIPT-TEMPLATES.md, creating only the files needed for the current project.
Create only the files needed for the current project:
SCRIPT.md: The story, structure, scenes, actions, and dialogue of the current video.CHARACTERS.md: The performance, appearance, dialogue, and voice requirements for characters in the current video.SCENES.md: The scenes, spaces, world representation, and continuity focus used in the current video.SHOTS.md: An overview of all shots in the entire video.
Simple product shots, videos without characters, or projects with a complete script may omit any of these files. Do not create empty files just for template completeness.
3. Story and Characters
When the current video includes a complete plot, character arcs, dialogue, or comedic beats, first read references/NARRATIVE.md. When character emotional changes, important reactions, or relationship conflicts need to actually translate into shot performances, then read references/PERFORMANCE.md to convert emotional words into observable body, gaze, movement, and behavior beats. Only the minimal rules required for all story projects are kept below.
When there is a story, prioritize ensuring the causal chain and character actions are valid before pursuing visual spectacle:
- Events are driven by existing conditions, character choices, or previous events.
- Characters act and speak only based on currently known information.
- Important turning points come from character choices or already established rules.
- Actions, information, or visual elements that need to be resolved at the end should have been truly established earlier in the narrative.
- Products or props cannot solve a problem that has not appeared before out of nowhere.
Character definitions go beyond just personality tags. Record their goals, concerns, typical action patterns, reactions under stress, dialogue rhythm, vocal characteristics, allowed emotional range, and changes in this video when truly needed by the current video. When a character needs to maintain a consistent voice identity across shots, hand over detailed timbre and voice references to video-audio, and do not duplicate a second set of voice specifications in the script layer.
4. Dialogue
When a shot includes visible speaking characters, long dialogue, multi-character conversation, or synchronized lip movements, read references/DIALOGUE-TIMING.md. First confirm that the lines can be naturally delivered within the target duration before proceeding to shot generation.
Dialogue serves the character's current actions; it is not responsible for explaining the world setting on behalf of the author. Narration / voice-over also belongs to script content, but should only be used when there is a genuine need for a narrative perspective, time compression, or information that is difficult to express naturally through visuals. Hand over sound production to video-audio as needed; do not treat narration as a universal fix-all layer for unclear scenes. Prioritize letting the previous line, previous action, or new information naturally lead to the next line; information that can be expressed through action should not be redundantly explained with dialogue.
After completing the dialogue, quickly review the entire conversation in chronological order to check for issues like sudden topic changes, repetitive explanations, lack of emotional transitions, or all characters using the same sentence structure. Natural dialogue does not mean forced inclusion of internet slang.
When VIDEO.md has a clear total duration / strict upper limit, read references/DURATION-BUDGET.md. First do a rough duration budget based on story segments and SHOTS.md to avoid discovering that the entire video length is out of control only after generation is complete.
5. Scene and Shot Overview
SCENES.md only records the spaces, visuals, and world rules needed for the current video. For complex machinery, unfamiliar devices, or critical spatial relationships, put detailed explanations in an accompanying document next to the corresponding reusable assets, rather than stacking all low-frequency rules into SCENES.md.
SHOTS.md is used to quickly read the entire film's shot arrangement from start to finish. Each entry must at least contain a shot ID, target duration, visual/action description, and shot purpose. Detailed production requirements are handed over to SHOT.md in video-shot.
For long projects, only add chapters or higher-level organization when the existing shot list has become difficult to read. Do not pre-assume cinematic-level hierarchy.
Completion Criteria
The video script layer is sufficient to let the next step know clearly: what this video is about, how the current characters perform, how the scenes are presented, and what shots are needed. At the same time, it has not duplicated the upstream source into a second fact database.