Not tested by us yet. We generate no video in our tests, and Baidu has published no per-second price or clip-length limit. Everything below comes from Baidu's announcement on PR Newswire. Every quality and cost claim is Baidu's own. We'll update this post when independent clips or a published price appear.
ERNIE Wizshot Micro-Expression (Wizshot-ME for short) is Baidu's application model for expressive video. It takes text, images and speech together and accepts long text input of up to 4,000 Chinese characters. Baidu says it leads the industry for podcasts and short animated dramas "at half the typical cost", with no baseline or price attached. It is available through the API on Baidu AI Cloud's Qianfan platform. If you produce talking-head or narrated animated content in Chinese, it is worth a trial. If you need a published price to budget against, wait.
| ERNIE Wizshot Micro-Expression | |
|---|---|
| Maker | Baidu |
| Announced | 2026-11-12 |
| Type | Expressive video from text, image and speech |
| Text input limit | Up to 4,000 Chinese characters |
| Access | API on Baidu AI Cloud Qianfan |
| Price | Not published ("half the typical cost", vendor claim) |
| Clip length, resolution | Not published |
What it is
Baidu describes it as a model that "specializes in highly expressive video creation". The name points at the focus: faces and small expressions, which is where talking-head video tends to fall apart. The announcement does not explain the architecture, and it does not say whether speech is an input that drives lip movement or a track the model generates. Treat those as open questions.
It launched alongside two other models. ERNIE 5.5 Flash Preview is a text model with 280 billion total parameters and 14 billion active per token, a 1M context window, and API pricing of RMB 1 per million input tokens and RMB 4 per million output tokens; we cover it in our ERNIE 5.5 Flash Preview review. ERNIE Light Deep Thinking targets multimedia search. Wizshot-ME is the only video model of the three.
The claims, and what is missing
Baidu makes one quantitative claim: half the typical cost for podcasts and short animated dramas. That is the weakest kind of number, because "typical" is undefined. It could mean human production, another API or Baidu's own earlier model. The announcement names no competitor and no price, so there is nothing to check it against.
Also missing: clip length, output resolution, frame rate, supported languages beyond Chinese, and any benchmark. The 4,000-character cap is the only hard limit. Baidu does not say how long the resulting video runs.
The name is easy to confuse. Despite the "Wizshot" label, we found nothing in the announcement linking it to Wizshot 1.0 from Wizstar, a separate company's reference-to-video model. We treat them as unrelated.
How it compares on paper
Only two nearby models publish a per-second price we can link. ByteDance's Seedance 2.5 Flash is reported at about 0.99 yuan per second for 720p, with clips up to 30 seconds. Google's Gemini Omni 1.2 Flash lists $0.0695 per second at 720p with audio on OpenRouter. Wizshot-ME has no equivalent figure, so a cost comparison is impossible today.
The difference in focus is clearer. Seedance and Omni are general text-to-video and image-to-video tools. Wizshot-ME is pitched at one narrow job: expressive, speech-driven video for podcasts and short dramas.
Who should care
- Chinese-language podcast and short-drama producers: the speech-plus-text input fits your workflow. Request Qianfan access and run a test script.
- Teams comparing video APIs on price: wait for a published rate, then run the same script through Seedance 2.5 Flash and Wizshot-ME.
- Anyone outside Chinese-language content: the announcement mentions no other languages, so there is little to evaluate yet.
We will publish a tested update if clips become available through an API we can call.