Lyric Video Director

A production-grade Hermes Agent skill for directing stylized lyric videos from a song, exact lyrics, and rough line timings.

It treats the lyric as the protagonist: GPT Image 2 designs typography-led compositions, Gemini Omni Flash animates them through reference-to-video, and deterministic QC preserves exact text and the untouched master audio.

What it covers

• Audio inspection, structure analysis, and coverage-complete timelines
• Visual bibles and tracer sequences before expensive bulk generation
• Kinetic typography and semantic word-object choreography
• Wordless instrumental narratives with causal transitions
• Exact-text verification and localized fallback repairs
• FFmpeg assembly against the original master song
• Technical, textual, musical, and creative-parity QC

Install

Copy or clone this repository into your Hermes profile's skills directory:

mkdir -p ~/.hermes/profiles/<profile>/skills/creative
git clone https://github.com/ShadoW-Shinigami/lyric-video-director.git \
  ~/.hermes/profiles/<profile>/skills/creative/lyric-video-director

Replace <profile> with your active profile, such as telegram.

The main skill definition is SKILL.md. Hermes discovers its linked references, scripts, and templates from the repository layout.

Repository layout

.
├── SKILL.md
├── references/
│   ├── art-direction-and-prompts.md
│   ├── audio-and-timeline.md
│   ├── full-song-production-qc.md
│   ├── instrumental-narrative-direction.md
│   ├── kinetic-typography-reference-video.md
│   ├── non-lyric-retrofit-workflow.md
│   ├── typography-and-qc.md
│   └── worked-examples.md
├── scripts/
│   ├── inspect_audio.py
│   └── validate_timeline.py
└── templates/
    ├── project.yaml
    └── timeline.example.json

Requirements

The skill expects access to:

• Python 3
• ffmpeg and ffprobe
• GPT Image 2 image generation
• FAL Gemini Omni Flash reference-to-video tools
• Optional: songsee for visual audio analysis

Run the helper checks:

python3 scripts/inspect_audio.py --help
python3 scripts/validate_timeline.py --help

Core rule

The supplied song is sacred. Generated video audio is discarded, and the original master remains the sole final soundtrack.

License

MIT — see LICENSE.