dLazy Audio Generation

Problem

When a prompt asks for speech, music, or sound effects, the correct dLazy CLI model is not always obvious: TTS, music, SFX, and voice cloning use different model names and parameter shapes. dlazy-audio-generate is meant to reduce that routing burden by having the agent classify the request, choose the matching model, and run the appropriate command.

How It Works

The skill is a thin client for the dLazy hosted API and relies on the dLazy CLI plus an API key. Its main workflow is:
- detect intent from cues such as generate audio, text to speech, generate music, or sound effect;
- choose among available models: gemini-2.5-tts, doubao-tts, and keling-tts for TTS; suno.music for music; keling-sfx for sound effects; and vidu-audio-clone or kling-audio-clone for voice cloning;
- inspect command parameters with dlazy -h, then execute the selected command against api.dlazy.com;
- if local file paths are supplied, the CLI uploads them to dLazy media storage, and generated output URLs are hosted on files.dlazy.com.

Boundaries

The skill does not replace the dLazy account, billing, or compliance layer. Missing credentials can produce unauthorized, insufficient credits can produce insufficient_balance, unreadable local files can fail before inference, and asynchronous generation can return 504 with a backend reason. Because output depends on a third-party API and safety policies, prompts, cloned-voice inputs, and distribution of generated audio should be handled with appropriate permissions and review.