onimusya/media-gen / skills/media-generation
media-generation
UnverifiedGenerate and edit images, create videos and music, synthesize speech, transcribe and translate audio using multiple AI providers (OpenAI, Google, ElevenLabs, Deepgram, Fal, Luma, Replicate, Stability, Runway, OpenRouter, Edge TTS). Use when the user asks to create media assets, generate pictures, make videos or music, produce voiceovers, transcribe recordings, or work with any visual/audio content.
| File | Kind | Size |
|---|---|---|
| LICENSE | other | 1.0 KB |
| RELEASE_NOTES.md | markdown | 8.6 KB |
| SKILL.md | skill md | 9.2 KB |
| scripts/install.cmd | script | 1023 B |
| scripts/install.ps1 | script | 1.7 KB |
| scripts/install.sh | script | 1.1 KB |
| scripts/media-gen.cmd | script | 78 B |
| scripts/media-gen.mjs | script | 564.7 KB |
| scripts/media-gen.ps1 | script | 145 B |
| scripts/media-gen.sh | script | 148 B |
RELEASE_NOTES.md
View on GitHub# Release Notes
## v1.7.0 (2026-11-21)
### Music generation
- Add `music generate` for Google Lyria 3.5 and ElevenLabs Music v2.5, with MP3 output, duration, and instrumental options.
- List music models under the `music-generate` capability and document the new command with runnable examples.
### Provider and CLI updates
- Default Google video to Gemini Omni 1.2 Flash and use its duration parameter.
- Add OpenAI GPT Image 2.6 models and refresh the Runway and OpenRouter catalogs.
- Protect existing video outputs, honor provider defaults in video commands, and route Luma requests with the credential for each API.
- Fix the npm package entry point, upgrade vulnerable dependencies, and use Node.js 24 LTS for development. The standalone CLI targets Node.js 18 and later.
### Validation
- 79 tests pass, along with type checking, lint, and the production build.
- Live API tests produced audio from both music providers and video from Gemini Omni 1.2 Flash.
## v1.6.1 (2026-10-23)
### Fixes
- Validate Azure GPT Image 2/2.5 custom dimensions locally before making an API request.
- Require width and height to be positive integers divisible by 16 for Azure image deployments and OpenAI GPT Image 2 models.
- Suggest the nearest aligned dimensions when validation fails; for example, `1920x1080` now suggests `1920x1088`, with `1536x864` documented as an exact 16:9 option.
- Apply the same validation to generation, editing, batch requests, and dry runs.
### Validation
- 67 tests covering provider API contracts, image-size validation, model catalog consistency, and command model resolution.
- TypeScript checks, production build, and valid/invalid Azure CLI dry runs pass. No paid generation requests were used.
## v1.6.0 (2026-10-23)
### Models and capabilities
- Add GPT Image 2.5 Flare and Sunburst for OpenAI and Azure. Azure image models are `gpt-image-2.5-flare`, `gpt-image-2.5-sunburst`, and `gpt-image-2`.
- Add Google Nano Banana 2.1 image generation/editing, Gemini Omni 1.1 Flash video generation/image-to-video, and Gemini 3.8 Flash TTS.
- Add Eleven v4/v4 Turbo, Luma Ray 3.2, and Deepgram Nova-3 Pharma.
- Refresh OpenRouter models with FLUX 3, Qwen Image, Seedream 5, Hailuo 3, Seedance 2.5, Wan 3, and other verified catalog entries.
### API fixes
- Parse Google Interactions media from REST output steps and preserve WAV audio; wrap legacy PCM as WAV.
- Route Stability SD3.5/Core/Ultra through the multipart Stable Image API.
- Read MiniMax TTS audio from `data.audio` and preserve fal queue status/result URLs.
- Use required Runway pixel ratios and durations, handle terminal OpenRouter video states, and pass configured video models correctly.
- Route Ray 3.2 through Luma Agents; support `LUMA_AGENTS_API_KEY` while retaining legacy job routing.
### Migration notes
- Imagen and direct Gemini API Veo 3.1 previews are retired. Use Nano Banana and `gemini-omni-1.1-flash`; Omni returns a completed video synchronously.
- GPT Image models reject the legacy `style` parameter; describe style in the prompt.
- Remove advertised fal/Replicate edit and image-to-video capabilities without adapter implementations.
- Document capability-specific model IDs and provider documentation for all 13 providers in the README.
### Validation
- 50 tests covering API contracts, model catalog consistency, and command model resolution.
- TypeScript checks, production build, and CLI dry runs pass. No paid generation requests were used.
## v1.5.0 (2026-07-09)
### New Capabilities
- **OpenRouter video generation** — 12 models via `POST /api/v1/videos` (async with polling)
- google/veo-3.1, google/veo-3.1-fast, google/veo-3.1-lite
- alibaba/happy-horse-1.1, alibaba/happy-horse-1.0, alibaba/wan-2.7
- x-ai/grok-imagine-video
- kuaishou/kling-video-3.0-pro, kuaishou/kling-video-3.0-standard
- minimax/hailuo-2.3
- bytedance-seed/seedance-2.0-fast, bytedance-seed/seedance-2.0
- **OpenRouter image** — added microsoft/mai-image-2.5
- Auto-generate `~/.media-gen/.env` template on first run (all keys commented out)
- `--version` now reflects package.json (injected at build time)
- `.plugin/plugin.json` for Open Plugins standard (cursor.directory compatibility)
- `npx skills add onimusya/media-gen` installation support
### Fixes
- Fixed OpenRouter video endpoint (`POST /api/v1/videos`, not `/video/generations`)
- Fixed OpenRouter video download (uses `unsigned_urls[]` with auth header)
- Fixed `~/.media-gen/logs/` not created on `--help` or `--version`
- Fixed `--version` showing hardcoded `1.0.0`
## v1.3.0 (2026-07-08)
### New Capabilities
- **Runway image generation** — gen4_image, gen4_image_turbo, gemini_image3_pro, gemini_image3.1_flash, gemini_2.5_flash, gpt_image_2
- **Runway multi-provider video** — veo3, veo3.1, seedance2, seedance2_fast, seedance2_mini, happyhorse_1_0, gemini_omni_flash (text-to-video + image-to-video)
- **Deepgram TTS** — Aura-2 and Aura models with 19 featured voices (English, German, Italian, Japanese)
### Updated Models
- **Runway**: expanded from 5 to 18 models (image + video + image-to-video)
- **Deepgram**: added `aura-2` and `aura` TTS models with voice IDs (e.g., `aura-2-thalia-en`)
- **Fal.ai**: added ByteDance Seedream v5/v4, Dreamina v3.1, all Seedance 2.0 variants (standard/fast/mini)
### Fixes
- Fixed Fal.ai queue status/download URL construction (strips endpoint suffix for queue paths)
- Fixed Azure provider to use Azure AI Services OpenAI-compatible endpoint format
## v1.2.0 (2026-07-08)
### Changes
- **Azure provider rewritten** for Azure AI Services OpenAI-compatible endpoint (`https://{resource}.services.ai.azure.com/openai/v1`)
- Azure models: gpt-image-2, gpt-image-1.5, gpt-4o-mini-tts with 13 voice IDs
- Added `LICENSE` and `RELEASE_NOTES.md` to skills folder
- Added prerequisites section to README
- Added ByteDance Seedream v5/v4 and Dreamina v3.1 for Fal.ai
- Added all Seedance 2.0 variants for Fal.ai (text-to-video, fast, mini)
## v1.1.0 (2026-07-07)
### New Providers
- **MiniMax (Hailuo)** — video generation (Hailuo 2.3, 2.3Fast, 02) + TTS (speech-2.8-hd/turbo)
- **Google Gemini TTS** — 30 voices, controllable style, 3 models (gemini-3.1-flash-tts-preview, gemini-2.5-flash-preview-tts, gemini-2.5-pro-preview-tts)
### New Features
- `--instructions` option for OpenAI/Azure TTS (gpt-4o-mini-tts) to control voice style, tone, accent
- `MEDIA_GEN_VOICE_ID` env var — set default voice-id so `--voice-id` is optional
- `MEDIA_GEN_LOG_LEVEL` env var — configurable log level (silent/error/warn/info/debug/trace)
- Structured logging to `~/.media-gen/logs/media-gen-YYYY-MM-DD.log`
- Minified + obfuscated production build (519KB)
- Release scripts (`scripts/release.sh`, `scripts/release.ps1`)
- `config init --global` for user-level setup at `~/.media-gen/`
### Fixes
- Fixed Fal.ai queue status/download URL construction (model path stripping for Seedance 2.0)
- Fixed home `~/.media-gen/.env` loading — now properly overrides stale system env vars
- Fixed Google Veo API: correct model names (`veo-3.1-generate-preview`), auth headers (`x-goog-api-key`), job polling URLs
- Azure provider rewritten to use Azure AI Services OpenAI-compatible endpoint
### Updated Models
- **OpenAI**: gpt-image-2, 13 voice IDs (alloy, ash, ballad, coral, echo, fable, nova, onyx, sage, shimmer, verse, marin, cedar)
- **Azure**: gpt-image-2, gpt-image-1.5, gpt-4o-mini-tts, 13 voice IDs
- **ElevenLabs**: 21 premade voice IDs (fetched from live API, free plan)
- **Google**: imagen-4.0, veo-3.1 variants, 30 TTS voices (Kore, Puck, Zephyr, etc.)
- **Fal.ai**: Seedance 2.0 (standard/fast/mini), Seedream v5/v4, Dreamina v3.1
- **Deepgram**: nova-3, nova-3-medical
- **Luma**: ray3.2, ray3.14
- **Runway**: gen4.5, aleph-2.0
- **Replicate**: flux-2-pro/max, flux-kontext-pro, gen-4.5, happy-horse-1.0
- **OpenRouter**: gpt-image-2, gemini-3.1-flash-image, grok-imagine, flux.2-klein-4b
## v1.0.0 (2026-07-06)
### Initial Release
- 13 provider adapters: OpenAI, Google, Azure, ElevenLabs, Deepgram, Fal.ai, Luma AI, Replicate, Stability AI, Runway, OpenRouter, MiniMax, Edge TTS (free)
- CLI commands: image generate/edit, video generate/image-to-video/extend, voice tts/clone/isolate, audio transcribe/translate
- AI agent SKILL.md following Agent Skills open standard
- Default provider/model/voice-id via env vars
- Multi-level config: `~/.media-gen/.env` → project `.env` → env vars
- `models.json` config for model/voice registry (bundled + user override at `.media-gen/models.json`)
- Cross-platform scripts (bash, cmd, PowerShell)
- Single bundled JS file via esbuild
- Async video job polling with `--wait`, `--poll-interval`, `--timeout`
- Structured JSON output (`--json`) for agent integration
- Dry-run mode (`--dry-run`) for cost-safe validation
- 25 tests passing
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |