laitszkin/openai-text-to-image-storyboard

OpenAI Text to Image Storyboard

Standards

• Evidence: Generate prompts from the supplied content and require only the minimal mandatory inputs such as project path and content name.
• Execution: Convert scenes into valid JSON prompts, load the skill-local .env, and call the storyboard generator in the same turn whenever possible.
• Quality: Keep prompts and outputs in narrative order, use only /images/generations, and leave role-schema management to the video orchestration skills.
• Output: Write images plus storyboard.json under pictures/<content_name>/ without overwriting existing filenames.

Workflow

1. Read user text and decide the target scenes in the agent.
2. Convert scene prompts into a JSON prompt file (prompts.json).
3. As soon as article/chapter content is available, directly run the script (do not stop at suggestion-only mode).
4. Use this skill folder's .env first, then call /images/generations to render images.
5. Save files in narrative order and write storyboard.json.

Agent Execution Requirement

• After receiving article/chapter/script content, immediately enter generation flow.
• Convert content into scene prompts and execute the Python script in the same turn whenever possible.
• Only ask follow-up questions when mandatory inputs are missing (for example: no output project path or no content name).
• Always provide prompts through --prompts-file with valid JSON.
• Do not define roles.json schema in this skill; role schema details belong to video-production or novel-to-short-video.

Environment Configuration

Create .env in this skill folder (default path used by script):

• ~/.codex/skills/openai-text-to-image-storyboard/.env

You can still override via --env-file when needed. All CLI parameters take priority over environment variables.

• OPENAI_API_URL (required)
• OPENAI_API_KEY (required)
• OPENAI_IMAGE_MODEL (optional, default gpt-image-1)
• OPENAI_IMAGE_RATIO (optional, e.g. 16:9 / 4:3; recommended)
• OPENAI_IMAGE_ASPECT_RATIO (optional fallback alias)
• OPENAI_IMAGE_SIZE (optional, e.g. 1024x768, for providers that expect size)
• OPENAI_IMAGE_QUALITY (optional)
• OPENAI_IMAGE_STYLE (optional)

A template is provided at:

• ~/.codex/skills/openai-text-to-image-storyboard/.env.example

Command

Use JSON prompt file:

apltk generate-storyboard-images \
  --project-dir /path/to/project \
  --env-file ~/.codex/skills/openai-text-to-image-storyboard/.env \
  --content-name "1_chapter_title" \
  --prompts-file /path/to/prompts.json

prompts.json format:

[
  {
    "title": "Rain Chase",
    "prompt": "cinematic rain-soaked alley, tense running pose, blue neon reflections, dramatic rim light"
  },
  {
    "title": "Underground Archive",
    "prompt": "ancient underground library, floating dust in warm volumetric light, mysterious atmosphere"
  }
]

Notes:

• prompts.json must be valid JSON.
• For role-related schema (roles.json and recurring-role definitions), follow video-production or novel-to-short-video.
• If the provider ignores aspect_ratio, pass --image-size 1024x768 or set OPENAI_IMAGE_SIZE=1024x768.
• You can pass --api-url and --api-key to override OPENAI_API_URL and OPENAI_API_KEY.
• When an aspect ratio is set, the script applies center-crop post-processing so outputs still match the target ratio.

參考資料

• references/generate-storyboard-images.md — apltk generate-storyboard-images 工具的完整參數說明。在執行 CLI 生成圖片前閱讀。

Output Convention

Generated files:

• pictures/1_chapter_title/01_rain_chase.png
• pictures/1_chapter_title/02_underground_archive.png
• pictures/1_chapter_title/storyboard.json

If a filename already exists, the script appends _2, _3, etc. to avoid overwriting.