Vidu S1¶

Development line: project:vidu-s1 · thread vidu-s1
Last event: - · 0 dated since - · Researched: 2026-09-06 · confidence: medium

What it is¶

Vidu S1 is ShengShu’s hosted real-time video model for creators and product teams building spoken, interactive digital characters. Vidu lists it separately from Q1/Q2/Q3 clip-generation models.

• Character creation: builds a character from a persona, an initial human/anime/pet image, and a selected or cloned voice.
• Dynamic control: changes character actions from spoken input during an audio or video session.
• Client access: offers Vidu Stream and a currently documented Vidu S1 API (Beta) using a control WebSocket plus AliRTC media.

The product page lists 540p at 25 fps, up to 42 fps. Session duration and billing must be read from the live-session response because vendor documentation conflicts. Use it for responsive avatar interactions, not for a local-model install or a conventional finished clip.

Development line¶

• The dated line is not written up yet; what is known stands in the sections below.

What changed¶

2026-07-10 — Vidu S1’s public GitHub presentation received three experience-GIF edits. The formal release was already dated 2026-07-03, so this was a demo and documentation refinement rather than a new model release.

How to use this¶

1. For a no-code trial, open Vidu Stream, sign in and verify the account, then select a character or create one. — https://www.vidu.com/vidu-stream
2. Set the persona, upload an initial image, then select, upload, or clone a voice before starting the conversation. — https://www.vidu.com/vidu-stream
3. Use spoken instructions during the conversation to direct the character’s actions. — https://www.vidu.com/vidu-stream
4. For an application, obtain an API key and create a Live session with POST /live/v1/lives; retain live.id and the returned RTC credentials. — https://github.com/shengshu-ai/vidu-s1-api/blob/main/skills/vidu-s1-api/SKILL.md
5. Join Aliyun RTC and establish the authenticated control WebSocket, wait for conn_init_ack.success=true, then hang up, close the WebSocket, and leave the RTC channel. — https://github.com/shengshu-ai/vidu-s1-api/blob/main/skills/vidu-s1-api/SKILL.md

Best practices¶

• For browser clients, keep the raw Vidu API key on a trusted backend and proxy the authenticated WebSocket. Native browser WebSockets cannot set the required authorization header. — https://github.com/shengshu-ai/vidu-s1-api/blob/main/skills/vidu-s1-api/SKILL.md
• Treat the returned live_duration and final billed usage as authoritative. Do not hard-code an “unlimited” or fixed session cap. — https://github.com/shengshu-ai/vidu-s1-api/blob/main/skills/vidu-s1-api/references/api-reference.md
• In video mode, implement close–backoff–reconnect handling for NOT_READY. Create a fresh Live session for LIVE_CONN_INIT_FAILED. — https://github.com/shengshu-ai/vidu-s1-api/blob/main/skills/vidu-s1-api/SKILL.md
• Integrate Aliyun RTC as a separate media layer and subscribe once to the active remote stream. Ignore stale RTC or WebSocket events from a previous session. — https://github.com/shengshu-ai/vidu-s1-api/blob/main/skills/vidu-s1-api/SKILL.md
• Handle both server type:6 hangups and abnormal WebSocket closure. After an active hangup, close the WebSocket before calling leaveChannel(). — https://github.com/shengshu-ai/vidu-s1-api/blob/main/skills/vidu-s1-api/references/websocket-protocol.md

Superseded by this¶

• 2026-07-03 — Treating Vidu S1 as a standard prompt-submit-then-wait video generator or a Vidu Q-series variant is obsolete. It launched as a separate real-time interactive digital-character line.
• 2026-07-03 — Calling 2026-07-10 the Vidu S1 launch is obsolete. ShengShu formally released it on 2026-07-03.

Still unknown¶

• The public platform API says live_duration has a 600-second maximum, while the vendor-owned integration guide says to use a policy-specific returned value and shows 7200 seconds. No live account response was available to reconcile them.
• Current vendor sources also conflict on credit-unit and extra voice-clone pricing. Use an authenticated console or final billing response before approving a budget.
• The linked Vidu-S1 repository provides documentation and images, not public model weights or a runnable inference package; local deployment and model-license rights remain unverified.
• The 42-fps and long-duration figures are vendor and preprint claims, not an independent latency or quality guarantee for every deployment.

Sources¶

Agent brief¶

• Subject: project:vidu-s1, thread vidu-s1, 0 dated events - → -.
• Practical note: See the sourced usage and practice sections above, including their limits.
• Confidence: medium. Dated supersedes above are the authority for what is obsolete.