What is Sora?

Sora is a video generation model developed by OpenAI and first previewed in February 2024; Sora 2 and Sora 2 Pro launched on Sep 30, 2025 with an invite-based app, and OpenAI discontinued the Sora app on Apr 26, 2026, with the API ending on Sep 24, 2026. Outputs carried C2PA metadata and a visible watermark; weights were never released.

OpenAI's help center documents the discontinuation timeline; the API date is a planned sunset and may still change before Sep 24, 2026.

Which Sora versions existed?

OpenAI previewed the first Sora in February 2024 as a research model shown to red teamers and artists, then released Sora Turbo to ChatGPT Plus and Pro subscribers later that year. Sora 2 and Sora 2 Pro followed on Sep 30, 2025 with synchronized audio, a consumer iOS app and web access at sora.com.

According to OpenAI's help center, Sora 1 was sunset for US users on Mar 13, 2026, the Sora app was discontinued on Apr 26, 2026, and the Sora API is scheduled to end on Sep 24, 2026. No Sora weights were ever published, so the model line ends with the service.

Did Sora allow NSFW output?

No. OpenAI's usage policies prohibit sexual content involving minors, non-consensual intimate imagery and impersonation, and Sora applied prompt and output classifiers on top of them. The Sora 2 system card describes additional restrictions on uploads containing real people and on generating public figures without consent.

The provenance layer was the distinctive part: OpenAI stated that every Sora video included both visible and invisible signals, meaning a visible moving watermark plus C2PA content credentials in the file. That combination let platforms and readers verify Sora origin, which unwatermarked open models never offered.

“Every video generated with Sora includes both visible and invisible provenance signals.”

— OpenAI, Creating with Sora safely, accessed Sep 3, 2026

How is Sora used in tools we review?

Not at all: OpenAI's policies and classifiers made Sora unusable for the outputs the tools in our rankings produce, and the service is now discontinued. We keep the entry because Sora remains the reference case for provenance and because readers still search for it when comparing video tools.

For the video-capable services we did test, see the Pornworks review and the video hub. Under how we rate, the presence of any provenance marking is one of the points we record; most tested services show none.

What do C2PA credentials prove?

C2PA content credentials are signed metadata attached to a file that records which tool created it and what edits followed. Sora 2 outputs carried them, so an untouched file could be checked with a credentials reader. Most social platforms strip metadata on upload, which is why OpenAI added a visible watermark too.

For readers assessing a clip, present credentials are strong evidence; absent credentials prove little, because they may have been removed. The C2PA and watermarking entries explain the checks, and deepfake detection covers what to do when neither signal remains.

Sources

1. OpenAI — Sora 2 is here — launch Sep 30, 2025; February 2024 preview, accessed Sep 3, 2026.
2. OpenAI — Creating with Sora safely — visible and invisible provenance signals, accessed Sep 3, 2026.
3. OpenAI — Sora 2 system card (PDF) — safety mitigations, accessed Sep 3, 2026.
4. OpenAI — Usage policies — prohibited content, accessed Sep 3, 2026.
5. OpenAI help center — What to know about the Sora discontinuation — Apr 26, 2026; API Sep 24, 2026, accessed Sep 3, 2026.
6. OpenAI help center — Sora 1 sunset FAQ — Mar 13, 2026 for US users, accessed Sep 3, 2026.

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