Hermes Agent keeps proving it’s not just a coding assistant. A community member just built a full lyric video pipeline as a Hermes skill, stringing together GPT Image 2 and Google’s Gemini Omni Flash to turn Suno songs into typography-driven music videos. The result is a production tool, not a demo.
The lyric-video-director skill by Harsha Subramanyam (@Shad0WShinigami on X) takes a song file, exact lyrics, and rough line timings, then runs them through a multi-stage pipeline. GPT Image 2 designs the typography compositions. Gemini Omni Flash animates them into video segments with reference-to-video. The generated audio gets thrown out. The original master song is the only soundtrack.
The result is a lyric video where words actually move. They animate, transform, and become objects on screen instead of static captions over a wallpaper.
The Hermes Agent Pipeline Goes Deeper Than You’d Expect
The skill has 10 reference documents covering audio analysis, art direction, typography QC, and full-song production verification. It inspects the song structure, classifies every instrumental gap, builds a visual bible before generating anything, runs a tracer sequence on one verse before bulk generation, and verifies every rendered frame before assembly.
This isn’t a five-line prompt. It uses the same SKILL.md format that ships with every Hermes install, complete with scripts, templates, and references. Install it by cloning the repo into your skills directory and Hermes discovers it automatically on the next load:
git clone https://github.com/ShadoW-Shinigami/lyric-video-director.git \
~/.hermes/skills/creative/lyric-video-directorThe tech stack alone is worth noting. GPT Image 2 handles the image generation. Gemini Omni Flash (Google’s I/O 2026 video model) powers the animation. The skill ships with its own audio analysis scripts, project templates, and a timeline validation tool.
I’ve written about the Hermes Agent /learn command that lets you create skills from any source material, and the Creative Ideation skill that ships with the agent. This lyric video skill shows what’s possible when someone goes all-in on the format.
What Makes This Different From a Simple Prompt
The skill enforces non-negotiable rules. The master audio is sacred. It never time-stretches, remixes, or regenerates the supplied song. Every millisecond of the track gets covered, including intro breaths, solos, drum fills, and outro decay. A visual bible gets locked before any bulk generation starts, so you don’t end up with 50 unrelated images.
It also runs a tracer sequence before full production: one verse line, one high-energy line, and one instrumental gap get validated before the rest of the song gets processed. If the creative direction doesn’t work, you find out after 2 minutes of testing, not after 20 minutes of rendering.
The same Awesome Hermes Agent curated list tracks the growing ecosystem around it, and the official Hermes Agent Skills Hub gives you a catalog of what’s available across registries.
How the Multi-Model Stack Works
GPT Image 2 designs each typography composition as a standalone image. The lyric text gets rendered as creative typography with the visual bible’s color palette, materials, and motion grammar baked in. Gemini Omni Flash then takes those reference images and animates them into video clips through reference-to-video, preserving the exact text while making the words move, transform, and interact.
The skill’s core philosophy: the lyric is the protagonist. It defaults to rapid composition resets, word-object metaphors, mixed-media graphic systems, and spatial typography. It does not default to portrait-led cinematic scenes with captions laid beside them.
The Hermes Agent MoA 2.0 feature already lets you combine any provider’s models in one session. This lyric video skill takes that same multi-model concept and applies it to creative production instead of chat.
Bottom Line
The lyric-video-director is a genuinely impressive demonstration of what a well-built Hermes Agent skill can do. It bridges GPT Image 2, Gemini Omni Flash, and Suno into a single pipeline with production-level QC. That alone makes it worth trying.
I installed it on my machine today. If you use Hermes Agent and make music with Suno, you’ll want to do the same.
