dLazy Video Generation

What It Solves

When engineers use the dlazy CLI for video generation, the usual problem is not whether the API can be called, but choosing among too many models: veo-3.1, sora-2, kling-v3, seedance-1.5-pro, and wan2.6-r2v lean toward cinematic sequences, narrative clips, general short clips, image-to-video, digital humans, and action transfer. video generate reduces that selection burden by treating the prompt as the routing signal, choosing an appropriate dLazy video model, and then running the matching dlazy command.

How It Works

The skill is a lightweight client over the hosted dLazy API and does not add a local inference environment. On invocation, it performs three core steps:

• Model selection: It infers whether the request is better suited to cinematic output, narrative shorts, general video, or image-to-video based on the prompt, parameters, and input assets.
• Parameter checks: It uses dlazy -h to inspect command parameters and avoids missing required options such as --prompt.
• Generation execution: It sends text, image, or video inputs to the dLazy API; local media files are uploaded as part of the cloud generation flow, and output URLs are typically hosted on files.dlazy.com.

Authentication uses a dLazy API key. The key can be stored through the CLI in the user configuration directory, or supplied per invocation via the DLAZY_API_KEY environment variable. If the API returns unauthorized, the skill points to the dLazy dashboard for key retrieval; if it returns insufficient_balance, it explicitly reports the credit issue and directs the user to the recharge page.

Boundaries And Cautions

It is useful for engineering tasks that want to connect video generation to the dLazy ecosystem, especially the prompt-to-model routing and command execution path. It is not intended for offline inference or fully local generation, because media inputs and outputs go through dLazy cloud services. Errors related to safety policy, missing files, balance, or network failures should be handled using the documented 401, 501, 502, 503, and 504 codes rather than by blind retry.