BlaineGale/silver-design-skills

Social Media Video Production

Turn a topic into a directly publishable multi-platform video. Covers four content types: travel culture, knowledge/educational, product promotion, and lifestyle/story. Supports three production lines: output prompt packages, generate visuals, or produce final MP4s.

Core Principles

First set specifications, then write content, then produce visuals. Getting the order wrong leads to rework: aspect ratio determines the script, the script determines visuals, and subtitle specifications follow the aspect ratio.

Consume on demand. Image and video generation consumes credits; before any visual generation, report the estimated cost and let the user confirm before proceeding.

Do not overstep. Do not automatically publish to any platform, do not read or write user account credentials, and do not delete user files.

Execution Flow

Step 1: Confirm Specifications

Extract the following information from user input. Fill in missing items using the table below. Only ask clarifying questions when a missing item would significantly affect the result, and ask at most one critical question at a time.

Item	Default	Options
Aspect Ratio	9:16 portrait	9:16 portrait, 16:9 landscape, 1:1 square
Resolution	1080P	720P lightweight, 1080P HD
Duration	45 seconds	15s, 30s, 45s, 60s, 90s
Content Type	Infer from user description	Travel culture, knowledge/educational, product promotion, lifestyle/story, scenic compilation
Production Line	Infer from user phrasing	Prompt package, visuals, final MP4

Production line inference rules:

• User says "give me prompts", "I want to copy to Doubao/Jimeng/Kling", "just the copy" → Route A
• User says "generate visuals", "make a video", "I want images", "add visuals" → Route B
• User says "compose", "produce final", "I want an MP4", "final file" → Route C
• If nothing is explicit: start with Route A's prompt package, and at the end mention the option to upgrade to B or C

Resolution is fixed based on aspect ratio: 9:16 → 1080×1920, 16:9 → 1920×1080, 1:1 → 1080×1080.

Step 2: Write Script and Storyboard

Select the script structure based on the content type, and load references/script-templates.md.

Output as a Markdown table with columns: Shot Number, Time Range, Visual Content, Voiceover Copy, Subtitle Text, Asset Requirements.

Hard requirements:

• The first 3 seconds must contain a hook, type depending on content: question, contrast, suspense, benefit point, visual spectacle
• Each shot's voiceover should be 15 to 25 characters, spoken in no more than 6 seconds
• Total duration must exactly match the target duration; shot time intervals must be continuous with no gaps
• Subtitle text must match the voiceover copy exactly; do not write a separate version
• The last 3 seconds must have a clear call to action or a suspenseful closure

Step 3: Define Visual Style

Load references/visual-styles.md, and select a style based on content type and emotion.

For each shot, output a visual prompt in the format:

Subject + Action + Environment + Lighting + Camera Language + Style + Aspect Ratio

The prompt must include an aspect ratio keyword so that image and video generation models can recognize it.

Step 4: Output Subtitle Specifications

Load references/spec-sheet.md and output a subtitle parameters table.

Required fields: Font, Font Size, Color, Stroke Width, Stroke Color, Position, Background Style, Display Mode, Animation.

Auto-calculate font size based on aspect ratio: portrait: 60, 1:1 square: 54, landscape: 48.

Subtitle text must be broken into lines sentence by sentence; no single line exceeds 14 Chinese characters. Split if longer.

Step 5: Design Voiceover Plan

Load references/tts-domestic.md and output a TTS plan.

Recommend in order of priority: Free tier (Edge TTS), Advanced tier (SiliconFlow / Fish Audio), Cloning tier (VoxCPM).

Must provide: Voice name, Speed, Volume, Emotion annotation, Channels for audition.

Step 6: Produce the Output

Execute according to the production line.

Route A: Prompt Package – Output a complete Markdown document containing the script table, visual prompts, subtitle specifications, TTS parameters, and editing parameters. Can be directly copied to Doubao, Jimeng, Kling, or Jianying.

Route B: Visual Generation – First report the estimated credit cost. After user confirmation, call image and video generation capabilities shot by shot. Use the prompts from Step 3 when calling.

Route C: Final MP4 – Call the local toolchain to synthesize an MP4.

python scripts/make_video.py --spec spec_file.json --out output_directory

Step 7: Deliver and Self-Check

Check each item in the table below. Deliver only when all pass.

Check Item	Pass Criteria
Duration	Actual duration within 1 second of target
Aspect Ratio	Resolution matches aspect ratio
Subtitles	No typos, no obscuring key visuals
Audio-Visual Sync	Subtitle timing aligned with voiceover
Assets	No obvious repeated clips
Copyright	Background music and asset sources are credited
Platform Compliance	No absolute claims, no false promises

If any item fails, fix and re-render.

Differences Between the Three Production Lines

Route A: Prompt Package

The deliverable is a Markdown document. The user takes it and generates on their own platforms. No credits consumed, no local environment required.

Overseas asset libraries (Pexels, Pixabay, Coverr) are marked in the document as "proxy required"; domestic platforms are marked as no proxy needed.

Route B: Visual Generation

The deliverable is local image and video clip files. Credits consumed: ~5-10 credits per image, ~50-100 credits per 5-second video.

Before execution, report the estimated cost for that shot. The visual description for each shot must be specific enough to be used directly for generation; abstract descriptions lead to empty visuals.

Route C: Final MP4

The deliverable is a final MP4 file. Requires FFmpeg on the local machine.

Visual sources can be files from Route B's output or user-provided materials. When using local materials, first confirm the file path and duration.

Domestic Accessible Channels

Only use domestically accessible channels as the main line:

• TTS: Edge TTS (free, no key required), SiliconFlow, Fish Audio, MiniMax
• Copywriting: Doubao, Tongyi Qianwen, DeepSeek, Zhipu Qingyan, Kimi
• Visuals: Doubao, Jimeng, Kling, Tongyi Wanxiang
• Editing: Jianying

Overseas sources (Pexels, Pixabay, Coverr, ElevenLabs beyond Edge TTS) are only optional enhancements and must be marked as requiring a proxy when used.

Stop Conditions

Stop and explain if any of the following occur; do not force through:

• The combination of duration and aspect ratio requested by the user is not achievable with current video generation capabilities
• The visual prompt cannot support the narrative intent of that shot
• In the final MP4 route, FFmpeg is unavailable and the user does not accept degradation
• Involving unauthorized assets, portraits of others, or copyrighted content
• The user requests automatic publishing to a platform

Dependency Check

Run before execution:

python scripts/check_environment.py

If it returns ready, the environment is complete; proceed to the normal flow. If it returns partial, continue with available features and explain the limitations. If it returns needs_setup, show only the missing items for the current task. If it returns unavailable, provide recovery steps and a safe degradation plan.

Degradation Strategy

Missing Item	Impact	Degradation
Video generation capability	Route B cannot output dynamic visuals	Output static images with Ken Burns easing, or switch to Route A
Image generation capability	Cannot generate first frame	Use text-to-video only, or switch to Route A
FFmpeg	Route C cannot synthesize	Switch to Route A, and output a subtitle file that Jianying can import directly
Font files	Subtitle rendering fails	Fall back to system black font; indicate that style may deviate
Background music	Final video has no music	Output a silent video and attach a music recommendation list

After degradation, always mark the limitations and verification status. Stop if core correctness cannot be guaranteed; do not guess to fill in results.

Reference Files

• references/script-templates.md – Script structures and hook patterns for the four content types
• references/visual-styles.md – Style library, color schemes, camera movements, and visual prompt templates
• references/spec-sheet.md – Full parameters for aspect ratio, resolution, subtitles, TTS, and editing
• references/tts-domestic.md – Domestic TTS channels, voice selection, and parameter configuration
• references/setup-guide.md – Dependency installation and account configuration
• assets/subtitle-presets.json – Subtitle style presets