Generative Media Skills - Setup Guide¶
Source: calesthio/generative-media-skills
GitHub: calesthio/generative-media-skills (137⭐, MIT, published July 11, 2026)
Skills: 153 packages across 25 categories (45 indexed on skills.sh at 43-77 installs each, 2,143 combined indexed installs)
Category: Media Production / Generative AI Operations
First Seen: August 26, 2026 sweep
Quality Tier: 🔵 Community - high craft quality (per-skill EVAL.md, dated fact verification, progressive disclosure) but low install counts and no extracted marketplace audit verdicts; inspect SKILL.md packages before production use
The Generative Media Skills suite is a complete media production brain for AI agents: 153 hand-researched skill packages spanning provider intelligence (image, video, TTS, voice, music, avatar, 3D), production craft (cinematography, editing, sound design, post-production), content formats (ads, shorts, podcasts, demos, trailers), and deterministic QA tooling. It is one of the few skills.sh suites that names Hermes explicitly - the repo ships a HERMES.md entrypoint and the README lists "OpenClaw/Hermes-style agents" as a first-class target.
This maps directly onto CorpusIQ's daily media pipeline: UGC video generation, GPT-Image-2 image production, HyperFrames and Remotion composition, and the pending ElevenLabs voiceover migration.
Installation¶
The maintained-checkout pattern (recommended - keeps all 153 skills updateable with git pull):
git clone https://github.com/calesthio/generative-media-skills.git
# ~/.hermes/config.yaml
skills:
external_dirs:
- /absolute/path/to/generative-media-skills/skills
Or via the skills.sh marketplace CLI (verified working Aug 26, 2026 - discovers 151 of 153 packages):
# Install the whole suite
npx skills add calesthio/generative-media-skills
# Install a single skill
npx skills add calesthio/generative-media-skills -s openai-gpt-image
# List available skills without installing
npx skills add calesthio/generative-media-skills --list
Skills live under skills/providers/<category>/<skill>/ and skills/production/<category>/<skill>/. Each package is plain Markdown (SKILL.md) plus optional scripts/, references/, and assets/ - copy any individual package into ~/.hermes/profiles/<profile>/skills/ and Hermes loads it natively.
Known issue: the CLI skips 2 packages with a YAML parse error (amazon-rekognition, google-cloud-vision - multi-line frontmatter descriptions that trip the strict parser). Install those two by copying their directories manually.
Prerequisites¶
| Requirement | Details |
|---|---|
| Provider API keys (per skill) | Provider skills cite APIs but never bundle keys. Image skills need OpenAI/Stability/Flux keys; TTS skills need ElevenLabs/Cartesia/OpenAI keys; video skills need Runway/Kling/Veo keys - only for the providers you actually use |
| Runtime tooling (skill-dependent) | ffmpeg-media-finishing needs ffmpeg; comfyui-media-workflows needs a ComfyUI instance; hyperframes-video-composition needs the HyperFrames runtime; remotion-video-composition needs Node + Remotion |
| Hermes Agent | Any version that loads SKILL.md packages; no minimum version constraint |
No account registration is required for the suite itself - it is MIT-licensed Markdown + scripts.
Core Skills¶
| Skill | Installs | Use For |
|---|---|---|
| generative-media-skills (meta) | 77 | Suite entrypoint - routes a brief to the right provider and production skills |
| editing-montage | 63 | Montage/editing craft: rhythm, match cuts, structure |
| cinematic-shot-direction | 62 | Shot plans, camera language, lighting intent for video generation prompts |
| audio-mixing-mastering | 59 | Loudness, balance, and delivery mastering for voiceover and music tracks |
| color-grading-finishing | 58 | Look consistency and finishing passes across a shot series |
| generated-media-qa | 57 | Deterministic QA checks on generated media output |
| character-design-continuity | 54 | Keeping a character/avatar visually consistent across shots |
| storyboard-previsualization | 54 | Previsualization boards before committing to renders |
| ecommerce-product-imagery | 53 | Product image production for commerce pages |
| ffmpeg-media-finishing | 52 | FFmpeg-driven finishing: containers, codecs, captions, checksums |
| media-qc-delivery | 51 | Delivery-time QC: specs, manifests, platform requirements |
| social-short-production | 49 | Vertical social short format: hooks, pacing, platform specs |
| hyperframes-video-composition | - | HyperFrames timeline composition (matches CorpusIQ's HyperFrames stack) |
| heygen-avatar-video | - | HeyGen avatar video production |
| openai-gpt-image | - | GPT Image 2/1.5 workflows - matches CorpusIQ's gpt-image-2-medium backend |
| elevenlabs-tts | - | ElevenLabs voiceover production - the pending fix for the rejected macOS say voice |
| ugc-ad-production | - | UGC-style ad production playbook |
| remotion-video-composition | - | Remotion React video composition |
| media-provenance-rights | 47 | Provenance, consent, and rights records for published media |
Full catalog: SKILL_INDEX.md in the repo (153 packages: 25 categories split into Providers and Production families).
CorpusIQ Use Cases¶
| Use Case | How |
|---|---|
| Daily UGC video pipeline | ugc-ad-production + social-short-production replace ad-hoc scripting with researched format specs; media-qc-delivery adds a deterministic QA pass before Postiz posting |
| Image production consistency | openai-gpt-image encodes the current GPT Image 2 lifecycle (aliases, deprecations, cost limits) - exactly the backend image_generate uses; pairs with the GPT-Image-2 Style Library already catalogued |
| Voiceover migration | elevenlabs-tts + elevenlabs-dubbing-voice-conversion provide the researched playbook for moving off macOS say -v Samantha (rejected for robotic quality) to ElevenLabs voices |
| HyperFrames compositions | hyperframes-video-composition and comfyui-media-workflows (which validates HyperFrames timelines) harden the existing HyperFrames UGC pipeline |
| Social QA + rights | generated-media-qa + media-provenance-rights gate every outgoing asset - scriptable checks instead of eyeballing, plus provenance records for platform disputes |
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.
Limitations / Verification¶
- Install counts are modest (43-77 per indexed skill). The value is the research craft: every SKILL.md carries dated fact verification and per-package
EVAL.mdscoring. - Marketplace security audit verdicts (Gen Agent Trust Hub / Socket / Snyk) were not extracted this sweep - inspect
scripts/folders before running anything in production; most packages are Markdown-only. - Provider facts (prices, limits, deprecations) are marked volatile and dated (e.g., gpt-image-1 deprecation verified 2026-07-09) - re-verify against the linked official docs before quoting to users.
- The repo was last pushed July 14, 2026 - treat provider sections as snapshots of that date.
Related¶
- GPT-Image-2 Style Library Setup - style templates for the same image backend
- Skills Marketplace - marketplace index for more discovery batches
- New Skills - August 26, 2026 - discovery page for this sweep
Powered by CorpusIQ