Updated: September 3, 2026 · Docs · Clothoff AI Editorial Team

What is Vidu?

ShengShu's closed video model, best known for holding a subject's identity across a clip from reference images — and, in 2026, for real-time generation.

Vidu is a closed video generation model developed by ShengShu Technology, a Beijing company built on Tsinghua University research, and served through the vidu.com app and API; Vidu Q3 launched in early 2026, Q3 Reference-to-Video on Apr 13, 2026 and the real-time Vidu S1 on Jul 3, 2026. Its signature feature is reference-to-video, which keeps a subject consistent.

Vidu versions, access and filters — verified against developer pages on September 3, 2026
VersionDateDeveloperLicense / AccessFilters
Vidu Q2 and earlier2025 (versions listed on vidu.com; days not restated)ShengShu TechnologyClosed; vidu.com app and APITerms of Use apply; no separate NSFW clause on the terms page read
Vidu Q3Jan 30, 2026 (per ShengShu press coverage; product page unreachable at test time)ShengShu TechnologyClosed; app and APIPlatform moderation; rules not stated
Vidu Q3 Reference-to-VideoApr 13, 2026ShengShu TechnologyClosed; app and APIPlatform moderation; rules not stated
Vidu S1Jul 3, 2026ShengShu TechnologyClosed; real-time interactive generation; technical report on Hugging FaceNot stated in the report

ShengShu publishes launches as press releases; the vidu.com product pages did not load during verification, so dates rely on those releases.

What makes Vidu different from other video models?

Vidu's defining mode is reference-to-video: the model takes one or several reference images of a subject, object or scene and generates a clip in which that subject stays consistent in appearance and motion. ShengShu extended this in Vidu Q3 Reference-to-Video, released Apr 13, 2026, with wider visual and audio control.

Vidu S1, announced Jul 3, 2026, moves in a second direction: real-time interactive generation, where frames are produced as the user steers the scene. Its technical report is published on Hugging Face as a paper, but no weights accompany it; every Vidu version remains closed.

Does Vidu allow NSFW output?

Vidu runs only on ShengShu servers, so platform moderation applies to every prompt and reference upload. The developer Terms of Use, effective Jan 23, 2025 and issued by ShengShu AI HK Limited, bind users to lawful use, but the page we read on Sep 3, 2026 has no clause that names sexual content.

Absence of a named clause does not create permission. Reference-to-video is precisely the capability that turns a real person's photograph into moving intimate imagery, and doing so without documented consent is unlawful in the United States, the EU and elsewhere, whatever the platform's terms state.

“ShengShu Launches Vidu Q3 Reference-to-Video with Expanded Visual and Audio Capabilities.”

ShengShu Technology press release via PR Newswire, accessed Sep 3, 2026

How is Vidu used in tools we review?

No tool in our rankings names Vidu as a backbone. The relevance is conceptual: the reference-to-video mode that Vidu popularized is the same mechanism some third-party services describe when they promise consistent identity across a clip, so readers should apply the same consent test to both.

For the video-capable services we did test, see the Pornworks review and the Undresswith AI review. Neither discloses its model family; our scores under how we rate rest on tested upload handling, retention and deletion.

What does reference-to-video mean for consent?

A reference image is an identity token: the model is instructed to preserve the face and body it is shown. That makes reference-to-video more sensitive than text-to-video and comparable to face swap in legal exposure.

The reference-to-video and digital consent entries explain what documented permission looks like. If a clip of you was generated without it, the NCII report page lists the removal routes we track.

Sources

  1. Vidu — official site — product entry point, accessed Sep 3, 2026.
  2. Vidu developer platform — Terms of Use — effective Jan 23, 2025; ShengShu AI HK Limited, accessed Sep 3, 2026.
  3. ShengShu — Vidu Q3 Reference-to-Video release — Apr 13, 2026, accessed Sep 3, 2026.
  4. ShengShu — Vidu S1 release — Jul 3, 2026, accessed Sep 3, 2026.
  5. Hugging Face — Vidu S1 technical report — real-time interactive generation paper, accessed Sep 3, 2026.

Reference entry by the Clothoff AI Editorial Team; not legal advice. The access date marks the last verification.

FAQ

Frequently asked questions about Vidu

Is Vidu open source?

No. Every Vidu version, from the early releases through Vidu Q3, Q3 Reference-to-Video and the real-time Vidu S1, is a closed model served through the vidu.com app and API. ShengShu published a technical report for S1 on Hugging Face, but no weights, so there is no local inference path.

When were Vidu Q3 and Vidu S1 released?

ShengShu's press coverage places Vidu Q3 in late January 2026 and the Q3 Reference-to-Video upgrade on Apr 13, 2026; the Vidu S1 real-time model was announced on Jul 3, 2026. The vidu.com product pages did not load during verification, so the Q3 day rests on press releases rather than the product site.

What is reference-to-video?

Reference-to-video is a generation mode in which one or more reference images define a subject, object or scene, and the model produces a clip that keeps that subject consistent throughout. Vidu made the mode a headline feature; it is also the mode with the highest consent risk when the reference is a real person.

Does Vidu block sexual content?

The developer Terms of Use we read on Sep 3, 2026 require lawful use but do not name sexual content specifically, and ShengShu has not published a standalone NSFW policy on the pages we could reach. Server-side moderation still applies, and non-consensual intimate imagery is unlawful regardless of the platform's wording.

What is Vidu S1?

Vidu S1 is ShengShu's real-time interactive video model, announced on Jul 3, 2026, that generates frames as the user directs the scene instead of rendering a fixed clip after the prompt. A technical report appears on Hugging Face as a paper, and the model itself remains closed and cloud-hosted.

Do the tools we review run on Vidu?

No reviewed tool names Vidu publicly. Our rankings score tested behavior rather than model claims: how uploads are handled, how long files are retained, whether deletion works and how moderation responds. Reference-based identity preservation is assessed for consent risk whatever model performs it.