ZSDL Digital Human Student Edition

Problem: Many Moving Parts, Strict POST Constraints

Digital human video generation involves avatars, voices, TTS, video jobs, and downloads. Manual API use is fragile: POST cannot use curl, and fields like video_model_id, text, and content are easy to mix up.

How It Works: From Asset Check to Automated Download

• Verify GET /api/digital.base_image/list and GET /api/digital.CloneVoice/list with Authorization: Bearer, then create the 我的内容库（勿动）/13数字人作品/... folder structure.
• Number the copy first; do not run TTS yet. Then choose the user's own avatar and voice.
• Use Python to call POST /api/digital.v2.Voice/synthesis with text, then poll GET /api/digital.v2.Voice/result?id={id} to store audio_url.
• Submit video jobs with POST /api/digital.video/create, passing video_model_id, audio_url, version, name, and priority; do not pass raw text again.
• Normal and off-peak queues get delayed status checks. Successful jobs download output_video_url; failed jobs record the reason.

Boundaries and Caveats

• GET may use curl; POST must use Python to avoid Schannel SSL compatibility issues affecting request-body parsing.
• It suits users who already have a ZSDL key and manage personal avatars, voices, and video tasks; it does not cover public assets, billing, or permission approval.
• Sensitive-word blocks mark the affected item and show the original text. The user can revise, regenerate TTS, and resubmit without affecting other jobs.