Gemini 2.5 Multilingual Text To Speech

Problem

For multilingual products, automation pipelines, and prototypes, teams often need to turn text into usable audio quickly. Local TTS stacks can be heavy to maintain, while cloud models need a consistent calling interface. gemini 2.5 tts wraps Gemini 2.5 text-to-speech as a command-style skill: provide a prompt, parameters, or local file paths, and receive audio generated by the dLazy API instead of operating your own inference service.

How It Works

The skill is a thin client over @dlazy/cli, using dLazy's hosted API for inference and media handling. The main workflow includes:
- Authentication: store a dLazy API key via the CLI, or pass DLAZY_API_KEY for each invocation; keys can be rotated or revoked from the dLazy dashboard.
- Invocation: run dlazy gemini-2.5-tts with a prompt and relevant options; parameters can also reference upstream outputs using pipe references such as @N or @stdin.
- File handling: local image, video, or audio paths are processed through dLazy media storage, and generated output URLs are typically hosted on files.dlazy.com.
- Async tasks: with --no-wait, the response includes a task field, and you should poll the result with dlazy status.

Boundaries And Notes

It fits scenarios that need a stable cloud inference result, not fully offline or strict data-residency workflows, because prompts, parameters, and local file paths may be sent to the dLazy API. When you see unauthorized, insufficient_balance, or backend safety-policy errors, follow the message to add a key, top up credits, or adjust the input. The skill focuses on invoking speech generation; it does not handle post-production, subtitle alignment, or advanced voice design.