Write visual facts as reusable, modifiable, directly copyable character board, scene board, prop board, or state image prompts, and save them all to
剧集/<EP>/图片提示词.md. Use IMG-... headings for each item, with the body placed in a ### 可复制提示词 blockquote.
IMG-... is only the stable ID of that prompt entry; it does not declare that an image has been generated or is available. Actual images can only come from inputs already provided by the creator, or from confirmed production results.
Quick Start
用 $short-drama-image-prompts 为 EP001 已确定的人物、地点和道具写可直接复制的图片提示词
Entry
If there is a current visual setting, you can start directly; read the script only when the prompt needs to confirm a story state. Look Development is an optional branch.
First confirm the purpose: identity board, styling/state variant, location board, prop board, combined production sheet, or comparison style frame.
When storyboarding needs to produce an end frame, the accepted "ending keyframe prompt" can be turned into an independent IMG-... state image entry:
state the source SHOT and endpoint, follow the visual facts and reference scope, and do not redesign the action. Actual generation uses that IMG entry as the source.
Creator-readable explanations follow the project language; copyable body text follows short-drama.json#/format/prompt_language. When there is no
short-drama.json , the body defaults
to using en, and stays consistent within the same task; the prompt language cannot be inferred from the creator explanation language.
Workflow
- Lock down the identity, state, space, or comparison variable that this image must fix.
- Read only the required asset facts, visual direction, negative constraints, and provided reference images.
- Write the body in the order of "subject and identity anchors → current variant → composition/view → light color/material → background boundaries → prohibitions."
- Each reference controls only a clear part of identity, styling, geography, composition, or style, and in the
REF-...slot write the closed-vocabulary用途。 - Check whether identity and variant, text policy, viewpoint, and lighting conflict.
- If the user wants all assets, complete all of them; asset groups are only internal batches.
Prompt Requirements
- Start by writing the object and purpose; do not drown the identity in style words.
- Include only requirements that the current image can satisfy at the same time; for multi-view/state comparisons, state the layout relationship clearly.
- Keep stable recognition anchors; variants change only the parts allowed to vary.
- For entries named by continuity locks in "Visual Setting.md," the locked aspects appear as-is in the copyable body, so that this reference image itself fixes the cross-shot facts.
- Avoid piling up quality words that have no verification function.
- Visible text, logos, watermarks, interfaces, and subtitles are explicitly allowed or prohibited.
- The body can be copied directly and contains no placeholders, process explanations, file paths, or QA conclusions.
On-Demand Knowledge
By default, read only this SKILL and the current visual setting. When encountering the corresponding issue, open only one:
- Stage boundaries and rule levels:Stage Contract
- Minimal recipe for an ordinary single image:General Recipe
- Character identity boards and styling consistency:Characters and Styling
- Location geography, viewpoints, and lighting:Location Board
- Props, scale, materials, and text:Prop Board
- Shape and state variants:Shape and State Variants
- Multi-object combination board:Production Sheet Recipe
- Representative frames for comparing visual directions:Lookdev Style Frames
- Local modifications and preserve set:Spot Modifications
- Generatability check before completion:Review and Examples
Completion and Production
Completion is achieved when every named object has a clear purpose, copyable body text, reference boundaries, and prohibitions, and they do not contradict one another. When this request simultaneously
modifies image prompts and storyboards, refresh the "image prompt items" of the affected shots in that round, without creating a separate reconciliation artifact. Actual generation must
be transferred to $short-drama-produce, present the precise task and obtain explicit confirmation; this skill does not call external services.
Once the five creation documents are complete, you may transfer to $short-drama to perform a mechanical check of the cross-document structure once; content quality is still reviewed by the creator.
Installation and Maintenance
Run only during installation, upgrade, or troubleshooting python3 scripts/selftest.py。