Add graduation-memory-video skill — cinematic graduation tribute video creator

[arlenyang413-creator]

Summary

New Agent Skill for creating cinematic graduation tribute videos from a single portrait photo. Covers the complete workflow from image generation through video assembly to piano accompaniment.

What it does

Given one portrait photo, the skill guides an agent through:

1. Generate 6 graduation scene images (9:16 vertical) — high school, bachelor's, master's, doctoral, diploma, memorial book cover
2. Create AI transition videos (Kling or moviepy fallback) with cinematic variable-duration pacing
3. Assemble 15-18 second video with crossfade transitions and fade-out ending
4. Add warm piano solo accompaniment (Suno/Mureka or Python wave fallback)

Key features

• Variable-duration cinematic pacing — not uniform 3s/clip, but emotional rhythm: longer opening/closing, shorter transitions, peak moments get more screen time. Based on analysis of real Lovart成品 videos.
• Chinese academic gown color standards — correct 学士/硕士/博士 gown colors per Chinese university conventions
• Emotional color tone progression — 6 images follow deliberate warm-to-deep-to-soft color arc
• Person consistency strategy — detailed guidance on maintaining face consistency across text-to-image generation (the skill's greatest challenge)
• Complete prompt templates — full English prompts for all 6 images and 6 transition videos in references/prompt_templates.md
• Fallback workflows — moviepy crossfade montage + Python wave module for environments without Kling/Suno

Skill structure

graduation-memory-video/
├── SKILL.md              # Main instructions (179 lines)
├── LICENSE.txt           # Apache-2.0
└── references/
    └── prompt_templates.md  # Complete image & video prompts

Specification compliance

• ✅ name field: graduation-memory-video (lowercase, hyphens, ≤64 chars, matches directory)
• ✅ description field: 536 chars (≤1024), includes trigger keywords in English + Chinese
• ✅ license field: Apache-2.0
• ✅ compatibility field: documents environment requirements
• ✅ metadata field: author, version, category
• ✅ SKILL.md < 500 lines (179)
• ✅ References in separate file for progressive disclosure
• ✅ File references use relative paths from skill root, one level deep

Testing

Tested with actual image generation (ImageGen) and moviepy video assembly on WorkBuddy platform. The cinematic pacing strategy is derived from pixel-difference analysis of a real Lovart成品 video (720×1280 @ 30fps, 18.1s), which revealed 4 main scene blocks (5s→6s→~6.6s) rather than uniform 6×3 seconds.