Local AI asset generation MCP server: end-to-end text-to-image/audio/speech and image/text-to-3D with Qwen3-TTS, Stable Audio Open, SSD-1B, and TripoSR.