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 conn_init handshake, handles retries and heartbeats, and ends the session.

The Vidu API key must stay in Dify or another trusted backend. The returned rtc.token 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

Tool	Result	Standalone value
create_live_session	Live session metadata and RTC join credentials.	Requires an external RTC client and control WebSocket to start a call.
get_live_session	Current status and billing data.	Can be used directly in a Workflow.
list_voices	Custom cloned voices available to the Vidu account.	Can be used directly in a Workflow.

Integration Flow

Dify Workflow
  -> create_live_session
  -> pass live_id and RTC credentials to a trusted application backend
  -> RTC client joins Aliyun RTC and publishes local media
  -> backend opens the authenticated Vidu control WebSocket
  -> backend completes conn_init and keeps the connection alive
  -> get_live_session reads final status and billing data

Tools

Tool	Purpose
create_live_session	Create a Vidu S1 live session from persona and avatar image URI.
get_live_session	Query session status and billing fields.
list_voices	List custom cloned voices available to the Vidu account.

create_live_session returns the following stable Workflow contract:

live_id
status
live_duration
call_mode
rtc.app_id
rtc.channel_id
rtc.user_id
rtc.token
rtc.token_expire_at

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 rtc. 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

Field	Type	Required	Notes
vidu_api_key	secret	Yes	Accepts vda_xxx or Token vda_xxx.
region	select	Yes	cn uses https://api.vidu.cn; global uses https://api.vidu.com.

When credentials are saved, the provider makes a read-only GET /live/v1/voices 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 waiting; 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:

uv sync --project plugins/dify --dev

Run tests from the repository root:

PYTHONPATH=plugins/dify uv run --project plugins/dify pytest plugins/dify/tests

Package the plugin from the directory above this plugin project:

dify plugin package ./plugins/dify

Distribution

For early users, publish the generated .difypkg 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 langgenius/dify-plugins and keep README.md, PRIVACY.md, _assets/, and manifest.yaml in the plugin root.

Source and Support

• Source: https://github.com/shengshu-ai/vidu-s1-api
• Bug reports and support: https://github.com/shengshu-ai/vidu-s1-api/issues