Sora Is Gone: What Creators Are Using Instead in Late 2026

Short answer

Sora shutdown what creators use instead: the app on April 26, 2026 and the API on September 24, 2026. Creators moved mostly to Google Veo, Runway, Kling and Luma. For short-form specifically, generation was never the bottleneck — distribution was.

Key takeaways

What actually happened and when

OpenAI's own help center is the authoritative source on these dates. Exports are available from the Sora library until the deadlines pass, after which user data is permanently deleted.

Sora launched as a TikTok-style social app and closed roughly six months later. Reported causes: about $1 million a day in operating cost against roughly $2.1 million in total revenue, sharply falling active users, persistent copyright and deepfake problems, and the collapse of a Disney partnership reported at around $1 billion that had licensed characters for the app. The Sora research team moved to world-simulation research tied to robotics.

Where creators went

There is no single successor, and anyone telling you otherwise is selling something. Pick by job.

Veo 3.1 is the one worth understanding in detail if you work in short-form, because it is the only replacement built for the shape of the output. It generates natively in 9:16 rather than producing 16:9 and cropping, at 1080p with 4K upscaling, and it is rolling into YouTube Shorts, YouTube Create, Flow, the Gemini API, Vertex AI and Google Vids. Google reports more than 275 million videos generated in Flow in its first five months — a self-reported number, so weigh it accordingly.

The trade-off is provenance: every Google-generated video carries a SynthID watermark, and the Gemini app can verify whether a given video came from a Google model. That is a feature if you want to prove what is real and a constraint if you would rather your output not be identifiable.

Licensing also differs materially: Runway and Pika generally grant output ownership on paid plans, while Kling grants a usage right. For anything commercial, that distinction matters more than a quality benchmark.

Why this barely touched short-form clipping

It is worth being blunt about scope. Sora generated video. Clipping tools cut existing video. Almost no short-form workflow depended on Sora, and the shutdown changed nothing structural for creators repurposing podcasts, streams or interviews.

Where it did bite: B-roll. A number of creators had started generating filler footage rather than licensing stock, and that pipeline broke. The replacements are straightforward — Veo, Kling and Pika all do short generative B-roll, and most clipping tools ship stock B-roll integrations that never depended on a generative model in the first place.

The deeper point the shutdown made is one the numbers had already been making: generative video was never the clipper's bottleneck. Moment selection is. Reframing is. Captions are. Those three decide whether a clip works, and not one of them is a generation problem — no model, however good, tells you which 40 seconds of a two-hour podcast are worth posting.

A channel that publishes 40 clips a month with stock B-roll outperforms one that publishes four with spectacular generated B-roll, and no model release changes that arithmetic.

The vendor risk lesson

Sora went from launch to shutdown in about six months, with five months of notice on the API. That is the shortest lifecycle a major AI product has had, and it is a planning input now, not a curiosity.

What to do about it:

• Do not build a workflow on a single model you cannot swap. Aggregators such as Krea, Leonardo, fal and Higgsfield give pay-as-you-go access to several models under one interface, which is a reasonable hedge.
• Export your assets on a schedule, not at a deadline. Every vendor here can delete your library after a notice period. Sora's export window is closing on September 24, 2026.
• Prefer tools where output ownership is explicit. A usage right is not the same thing as ownership if a clip becomes commercially significant.
• Keep the source material. Generated footage is replaceable. The interviews, streams and recordings you own are not, and they are what a clipping pipeline actually runs on.

That last point is the one that survives whichever model wins next. Tools change every few months; a library of long-form you own does not.

What to do this month

Concretely, before September 24, 2026:

1. Export everything from Sora. Videos and images can be downloaded from the library. After the deadline, data is permanently deleted with no recovery path.
2. Check whether any production workflow calls the Videos API. The Sora 2 model aliases and snapshots are removed, so calls will fail rather than degrade.
3. Pick a replacement by job, not by benchmark. Generative B-roll, image-to-video and full shot direction are different problems with different winners.
4. Re-check that generation is a constraint at all. For most short-form operations it is not, and the budget is better spent on publishing cadence.

If your last month of output would have been identical without any generated footage — which is true for most podcast and stream clipping — the correct response to the Sora shutdown is to do nothing and move on.