Video Script Generation Skill#

This skill provides instructions on how to transform a user's storytelling prompt into a valid input-scripts.json entry for the video generator.

1. JSON Structure#

Every script must be an object within the main array in input/input-scripts.json.

json

{
  "id": "unique-id",
  "title": "Display Title",
  "orientation": "landscape" | "portrait",
  "voice": "en-US-JennyNeural" | "en-US-GuyNeural",
  "script": "The actual narrative content..."
}

2. Director Mode (Manual Visual Cues)#

To ensure high-quality, relevant stock footage, use "Director Mode" tags. Place them at the start of the sentence or block they describe.

• Syntax: [Visual: Descriptive Query]
• Best Practice: Be specific. Instead of [Visual: nature], use [Visual: green forest sunlight rays].
• Placement: The visual will stay active until the next [Visual: ] tag appears.

Example:

"[Visual: futuristic city neon night] The city never sleeps. [Visual: robotic arm assembly] High-tech manufacturing is the backbone of the economy."

3. Audio & Voice Settings#

You can choose from several high-quality neural voices. Specify these in the voice field of your JSON job.

Available Voices

GenderVoice IDStyle/Region👨 Maleen-US-GuyNeuralDeep, Authoritative (Recommended)👨 Maleen-US-ChristopherNeuralCalm, Steady👨 Maleen-GB-RyanNeuralBritish Accent👨 Maleen-IN-PrabhatNeuralIndian Accent👩 Femaleen-US-JennyNeuralWarm, Professional (Recommended)👩 Femaleen-US-AriaNeuralFriendly, Helpful👩 Femaleen-US-SaraNeuralCheerful, Bright👩 Femaleen-GB-SoniaNeuralBritish Accent

4. Job Settings Keys#

KeyTypeDescriptionidStringUnique slug for the video (used for the folder name).titleStringThe main title displayed in the video.orientationStringlandscape (16:9) or portrait (9:16).voiceStringUse one of the Voice IDs from the table above.showTextBoolean(Optional) Set to false to hide captions.defaultVideoString(Optional) Local filename for fallback (in input-assests/).scriptStringThe content to be spoken, including [Visual: ...] tags.

5. Script Writing Rules#

1. Scene Breaks: Aim for scene changes every 5-10 seconds. Longer scenes can get repetitive.

2. Voiceovers: The script text is exactly what will be spoken. Do NOT include instructions like (Scene 1) in the script text, as the TTS will read it.

3. Troubleshooting Visuals: If a search is too specific and returns 0 results, the system will automatically try fallbacks (Pixabay -> Generic).

   • Tip: Use keywords that describe the action or vibe (e.g., "morning sunrise", "fast car driving").