On November 12, 2026, Baidu introduced ERNIE-Wizshot-ME, a dedicated micro-expression video generation model listed on the Baidu AI Cloud Qianfan product portal and API reference documentation (with API docs dated November 11, 2026).

Rather than acting as a general text-to-video generator, ERNIE-Wizshot-ME combines a single character portrait image, text guidance, and an audio track to animate a still face into a talking-head clip. Baidu states the model focuses on close-up facial micro-expressions—including subtle eyebrow movements, lip-corner tension, and gaze shifts—for use cases such as podcast clips, short-drama close-ups, and narrated monologues.

How the API Works: Two-Step Asynchronous Task Flow

Per the Qianfan API reference documentation, ERNIE-Wizshot-ME operates via a standard two-step asynchronous job pipeline rather than returning video synchronously:

Step 1 — Submit Job (POST Create Video Generation Task): Send a POST request containing the reference portrait image, text prompt instructions, and driving audio track. Once accepted, the endpoint returns a unique task_id.

Step 2 — Poll Status (GET Query Video Generation Task): Poll the GET task-query endpoint with the returned task_id until processing completes, then retrieve the rendered 1080p video output. Companion endpoints for listing tasks (GET) and canceling queued jobs (DEL) are also documented.

图片说明

Pricing, Concurrency Caps & Practical Boundaries

Resolution, Duration & Per-Second Billing: The Qianfan product listing specifies 1080p output resolution and clip durations from 4 to 15 seconds (4s–15s). The published list price is RMB 0.70 per second of generated video—putting a 5-second talking-head clip at RMB 3.50 and a full 15-second clip at RMB 10.50.

Strict Default Concurrency & Queue Limit (1 Each): According to Qianfan's model billing table, the default concurrency limit and queue limit for ERNIE-Wizshot-ME are both set to 1. Standard accounts can only run one active generation task and hold one task in queue at a time unless they request a quota increase.

What This Is Not & Unverified Claims: This is a specialized portrait-animation video endpoint, not a general chat LLM, and it does not support clips longer than 15 seconds. Baidu has not published parameter counts or separate HKD billing terms for Hong Kong accounts, and we have not independently tested lip-sync accuracy or queue latency.