v4.0.1 — Gemini Omni

🆕 Google Gemini Omni (May 19, 2026)

Added 3 new endpoints for Google's brand-new "anything from anything" multimodal video model:

• `gemini-omni/video` — Video generation from text + up to 7 images + 3 audio refs + 1 video + 3 character IDs. 4K, 16:9/9:16, 4-10s
• `create_omni_voice` — Build a reusable voice character (preset voice + custom description) → kieAudioId
• `create_omni_character` — Build a reusable visual character (image + voice) → character_id

Suggested workflow

```

1. create_omni_voice (define voice: "warm gravelly British male")
2. create_omni_character (image + description + voice_id)
3. generate_video model="gemini-omni/video" (uses character_ids)
   ```

Other improvements

• Added `Dockerfile.local` for faster local development builds
• `docker-compose.yml` now builds from local source by default
• Version bumped to 4.0.1 in server.mjs, package.json, and server.json

Total tools: 38 (was 36)

See CHANGELOG.md for full history.