OpenAI's Sora API Shuts Down September 24 - What Creators Should Do Now

OpenAI's Sora video generation API shuts down on September 24, 2026. OpenAI announced the sunset date back in March, and the Sora website and mobile apps were already retired on April 26, 2026 - so this second cutoff is the part that actually breaks anything still running. If a script, an automation, or a third-party tool you use pulls video generations from Sora's API, it stops working the moment the API goes dark, with no announced grace period after that date.

What is actually ending on September 24

• The Sora-2 API - the developer endpoint apps and scripts call to generate video programmatically - is discontinued entirely.
• Any third-party tool, bot, or internal pipeline built on top of that API stops producing new generations the moment it goes offline.
• OpenAI has not promised a recovery window for account data after the cutoff, the same as when the consumer app shut down in April.

Who this actually affects

If you only ever used the Sora app or website, this deadline does not touch you - that side shut down months ago. The September 24 cutoff is for anyone who kept a workflow alive through the API: a content pipeline that calls Sora programmatically, a third-party app or plugin still built on it, or a batch-generation script scheduled to run on a cadence. If that describes any part of your setup, this is the actual deadline to plan around.

Before September 24

• Export or download anything you still need from a Sora-connected tool or account now, rather than waiting for the cutoff week.
• Audit any app, plugin or automation you rely on for a hidden Sora API dependency - a tool does not need the word "Sora" in its name to be calling it under the hood.
• Pick a replacement model early and re-test your best-performing prompts against it. Prompts do not transfer one to one between video models - each has its own preferred prompt shape, reference format and length limits.

Where creators are moving instead

None of these are drop-in replacements for Sora's exact behavior, but they cover the same core job - text-to-video and image-to-video generation - and are actively maintained: Google's Veo 3.1, Kuaishou's Kling 3.0, and ByteDance's Seedance 2.0 are the three most commonly cited alternatives right now. Each has different strengths in clip length, motion realism and reference handling, so the right pick depends on the kind of content you make - test more than one before standardizing a workflow on it.

Using it in Karya

Karya's Video Studio already runs Seedance 2.0 (Standard, Fast and Mini tiers) and Kling 3.0 for text-to-video and image-to-video, plus Google's Gemini Omni for video generation with first and last-frame control - all live today, with the credit cost shown before you generate. If Sora's API was part of your pipeline, this is a reasonable moment to test the same prompts against one of these instead of waiting for the cutoff to force the switch.