Baiyin Digital Human Video Generation

Problem

When a user wants to generate an AI digital human lip-sync video through the Baiyin Open Platform, the flow can involve selecting a model code, uploading local assets to public URLs, creating an asynchronous task, and polling task results. This skill turns that process into a conversation: it recognizes create, status, and result intents, and asks for missing required fields when needed.

How It Works

The skill expects BAIYIN_API_KEY and calls endpoints under https://ai.hikoon.com. Before creating a task, it queries the available digital human models, displays only modelName, and maps the user’s selection or keyword to modelCode. For local files, chat attachments, or otherwise inaccessible URLs, it directs the user to upload through the Baiyin file upload endpoint, then uses data.url for params.audio and params.first_frame. Optional fields such as params.resolution, count, and prompt are included only when provided, without inventing defaults. After task creation, it returns taskId, requestId, and current status; queries return queued, processing, succeeded, or failed, and a successful result prioritizes the video URL.

Boundaries

This skill covers creating and querying Baiyin digital human video tasks only; it does not replace general-purpose AI video generation. The model list comes from live API responses, not fixed data. It should not claim the video is complete before the task is succeeded, and garbled output such as ?? should be flagged as abnormal rather than delivered as a normal result.