Vidu S1 Dify Plugin

This plugin calls the Vidu S1 API to create and inspect live digital-human
sessions. It is designed primarily for Workflows used by teams that already
have, or plan to build, an RTC client; its tools remain compatible with Dify
Agents for non-sensitive session operations.

Product Scope

The plugin runs short-lived server-side tool calls inside Dify. It can create a
Vidu session and return the structured session and RTC data needed by a later
Workflow step or by your application backend. Use a Workflow for complete RTC
integration so sensitive RTC Credentials can be passed deliberately between
nodes instead of being handled conversationally.

It does not embed a digital-human call in Dify. A complete Vidu S1
experience also requires:

• An application that integrates the Aliyun RTC SDK, joins the returned RTC
  channel, publishes microphone media, and renders the digital-human stream.
• A server-side Vidu control WebSocket that authenticates with the Vidu API key,
  completes the handshake, handles retries and heartbeats, and ends
  the session.

The Vidu API key must stay in Dify or another trusted backend. The returned
is a short-lived, single-session credential. The plugin exposes it
only as structured tool output so an RTC client can join the channel; it does
not include the token in its human-readable text or links. Dify Workflow run
history may retain structured outputs, so configure retention and access as you
would for other sensitive credentials.

What Users Get

Integration Flow

Tools

returns the following stable Workflow contract:

The plugin does not pass through the complete upstream HTTP response. Session
fields stay at the top level for convenient Workflow references, while all RTC
Credential fields stay under . If any field in this contract is missing
from the Vidu response, the tool fails immediately instead of returning an
empty value that would fail later in the RTC Client.

Configuration

When credentials are saved, the provider makes a read-only
request with a 10-second timeout. This verifies both the
API key and selected region without creating or billing a live session. A
temporary Vidu API outage can therefore prevent credential setup; validation
errors are sanitized and never include the API key or upstream response body.

Verification

1. Run the automated tests and package the plugin with the current Dify CLI.
2. Install the package in a test Dify workspace and verify that every declared
   output can be referenced by a later Workflow node.
3. With a real Vidu API key, create a session and confirm that it starts in
   ; query it and check the billing fields; list voices and compare the
   result with the Vidu account.
4. Pass the create response to an RTC test application, join the channel,
   complete the control WebSocket handshake, confirm audio/video interaction,
   end the call, and verify the final bill.

Local Development

Install runtime and test dependencies in a Python 3.12 environment:

Run tests from the repository root:

Package the plugin from the directory above this plugin project:

Distribution

For early users, publish the generated as a GitHub Release asset and
ask users to install it in Dify with Install Plugin > From GitHub.

For Marketplace distribution, submit the packaged file to
and keep , , , and
in the plugin root.

Version 0.0.3

Adds the source repository and maintainer contact metadata required by current
Marketplace validation. Includes the 0.0.2 fixes below, with no breaking changes.

Version 0.0.2

Credential validation and API failures now suppress sensitive upstream exception
chains in Dify logs, including timeout errors. Malformed API keys are rejected
before an HTTP request. The package excludes maintainer notes and local artifacts.
Existing S1 tools, parameters, and Workflow outputs remain compatible.

Source and Support

• Source: https://github.com/shengshu-ai/vidu-s-api
• Bug reports and support: https://github.com/shengshu-ai/vidu-s-api/issues