dLazy Multimodal Generation

Problem Solved

Multimodal generation tools often split image, video, and audio into separate commands, parameter sets, and service entry points. A caller must decide the task type, model capability, and CLI flags before running anything. dLazy Multimodal Generation wraps this layer into one skill: after a user asks for a poster, narrative clip, voiceover, music, or sound effect, the skill routes the request to a suitable dlazy CLI model and executes the generation command.

How It Works

The skill is a thin client over the hosted dLazy API. It routes by intent across model categories: image options include realism, fast sketch, text-to-image, SVG/vector, and Midjourney style; video covers cinematic clips, narrative shorts, image-to-video, and action transfer; audio covers TTS, music, and sound effects. The usual execution path is:
- determine the media type;
- choose the closest model;
- inspect that model's required parameters with dlazy -h;
- run a command such as dlazy seedream-4.5 --prompt ....

Local file paths are uploaded to dLazy media storage, and returned output URLs are hosted by dLazy.

Boundaries

It fits engineering and content workflows that already have a dLazy API key and accept cloud generation. It is not suitable for offline, private, or strictly local-file-only pipelines. insufficient_balance means credits need recharge; unauthorized means the API key is missing or invalid; 502 usually indicates a local file read failure; and 504 can reflect an asynchronous backend failure or safety-policy rejection.