跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
浏览器与自动化

storyboard-lineart-generator

Generate and register per-shot storyboard line-art from an approved storyboard session. Use after script package approval and text-only prompt preflight. Supports parallel/multi-agent generation and updates session JSON/HTML with saved asset paths.

kevinchin12 · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add kevinchin12/storyboard-director --skill storyboard-lineart-generator

技能说明


name: storyboard-lineart-generator description: "Generate and register per-shot storyboard line-art from an approved storyboard session. Use after script package approval and text-only prompt preflight. Supports parallel/multi-agent generation and updates session JSON/HTML with saved asset paths."

Storyboard Line-Art Generator

Use this sub-skill when the user only wants line-art generation from an existing approved storyboard session, or when the main pipeline delegates the line-art stage.

This sub-skill can run independently from a session.json.

Runtime Root

Resolve STORYBOARD_DIRECTOR_ROOT before running commands:

  • installed bundle: ${CODEX_HOME:-$HOME/.codex}/skills/storyboard-director
  • source checkout: the nearest ancestor containing both the main SKILL.md and scripts/

Use absolute script paths. Do not assume the current workspace contains the skill source folder.

Session Resolution

Use the explicit session path when provided. Without one, inspect configured sessions/ and examples/ and proceed only when exactly one session matches the request. If several sessions could match, ask for the exact path; never choose the newest session by guesswork.

Scope

Do:

  • read the approved session.json
  • confirm aspect_ratio exists; route a missing ratio back to script planning instead of mutating approved inputs
  • confirm text_review.script_logic_check and text_review.line_art_prompt_preflight have no blocking issues
  • create a worker queue before image generation
  • generate one real raster line-art image per shot
  • save or register each result under the session assets/ folder
  • update only line_art.gallery and compact status fields
  • render session.html
  • run smoke validation

Do not:

  • rewrite the script
  • rewrite production anchors unless generation is blocked by a concrete prompt error
  • generate final colored frames
  • generate aggregate video prompts
  • upload generated line art back to AI for visual critique
  • ask for API keys or switch to CLI/API image fallback

Multi-Agent Rule

If there are 2 or more independent shots, build a worker queue before the first image call.

Worker input should be short:

  • session_path
  • assets_dir
  • project_mode
  • aspect_ratio
  • compact production_reference
  • shot_id
  • line-art prompt
  • needed character/scene reference paths
  • target output filename

Worker output must be path-based:

  • shot_id
  • asset_type: line-art
  • status
  • output_path
  • optional retry_prompt
  • short note

Only the orchestrator edits session.json and renders HTML.

Create the full queue before any image call. Default to four concurrent line-art workers. If delegation is unavailable, record serial_unavailable and execute the same queue in Shot order.

Parallel workers must return explicit local source paths. Do not use one shared capture_generated_asset.py marker across concurrent jobs.

Line-Art Rules

Line art is a contour and structure map:

  • sparse outer contours
  • key silhouette lines
  • major object boundaries
  • horizon, vanishing, or perspective guide lines only when useful
  • white or off-white background
  • thin clean lines

Avoid:

  • shadows
  • gray fills
  • hatching
  • cross-hatching
  • stippling
  • texture
  • material rendering
  • lighting effects
  • decorative detail
  • dense sketch buildup

Each prompt must separately state:

  1. shot structure and depth layers
  2. camera, eye level, framing, and perspective
  3. one frozen action moment
  4. contour and avoid constraints

Use this pattern:

Use case: illustration-story
Asset type: storyboard line-art draft for <Shot ID>.
Primary request: real model-created raster line-art image; not SVG, typography, or polished illustration.

SHOT STRUCTURE:
<subject placement, major objects, foreground / midground / background>

CAMERA AND PERSPECTIVE:
<shot size, viewpoint, framing, vanishing point, required alignments>

ACTION LAYER:
<one frozen action moment>

DETAIL LIMIT:
Only outer contours, key silhouettes, major boundaries, minimal face/hand cues, and necessary perspective guides.

STYLE CONSTRAINTS:
Sparse thin contour drawing, locked <aspect_ratio>, plain white or off-white background, long edge around 1536px. No color, shadows, gray fills, hatching, stippling, gradients, texture, material rendering, lighting effects, decorative detail, text, labels, borders, UI, watermark, logos, or extra characters.

Use a two-pass blocking-layout then clean-contour process only when preflight identifies concrete perspective or shared-space risk. After one repeated perspective failure, simplify the task to an animation layout or blocking thumbnail instead of repeating the same atmospheric prompt.

Line-art JPGs should normally stay below about 1.5MB. If one is heavier, retry once with stricter contour-only constraints.

Target new assets as .jpg, usually:

assets/shot01-line.jpg
assets/shot02-line.jpg

Registration

After every successful output, register it:

python3 "$STORYBOARD_DIRECTOR_ROOT/scripts/register_story_asset.py" SESSION.json --asset-type line-art --shot-id "Shot 01" --source SOURCE_IMAGE --output-name shot01-line.jpg --validate-profile smoke --workspace-root /path/to/workspace

Verify:

  • the registered path is under assets/
  • session.json has the matching line_art.gallery item
  • session.html contains the same path

If the generated source file cannot be located, mark the shot as blocked and keep the prompt for retry. Do not claim delivery from a chat-only image.

Preferred built-in image_gen capture:

python3 "$STORYBOARD_DIRECTOR_ROOT/scripts/capture_generated_asset.py" SESSION.json --mark --workspace-root /path/to/workspace

Then call image_gen, and register the new local file:

python3 "$STORYBOARD_DIRECTOR_ROOT/scripts/capture_generated_asset.py" SESSION.json --asset-type line-art --shot-id "Shot 01" --output-name shot01-line.jpg --workspace-root /path/to/workspace