Wan 2.6 R2V Reference Image Video Generation

Problem

When engineers need to turn a single reference image into a continuous video, calling the underlying video model directly often spreads the work across API authentication, file upload, async polling, and error recovery. wan2.6 r2v collapses those steps into one Wan 2.6 R2V call: provide a reference image, prompt, and parameters, then receive a downloadable video URL. It fits workflows such as design review, asset preview, or front-end content demos.

How It Works

• Core capability: generate continuous videos from reference images, triggered by intents like wan 2.6, reference image to video, and generate video.
• Authentication: requests require a dLazy API key; the key can be saved by the CLI in the user config directory or supplied per invocation via DLAZY_API_KEY.
• Execution path: the skill is a thin client over the hosted dLazy API. prompt and parameters are sent to api.dlazy.com for inference, local image/video/audio files are uploaded to files.dlazy.com for the model to read, and returned video URLs are hosted on that domain.
• Async tasks: with --no-wait, the response may return only task.generateId and status; poll the task until generation completes.
• Error handling: common issues include unauthorized access, missing --prompt, local file not found, insufficient balance, server errors, and async task failures caused by safety policy.

Boundaries and Caveats

This skill does not replace local video rendering or manage long-term video storage; output URL lifetime depends on the dLazy service policy. If the response contains insufficient_balance, explicitly tell the user that credits are insufficient and guide them to the dLazy dashboard. If it returns unauthorized, configure the API key before resuming. Local file paths must be readable, and files enter the cloud generation pipeline, so verify data boundaries for sensitive assets.