The short version: the shutdown OpenAI announced when this article first ran in March is now complete. OpenAI announced on March 24, 2026 that it was discontinuing Sora. The Sora app and website closed on April 26, 2026, and the Sora 2 models and OpenAI's Videos API were switched off on September 24, 2026. OpenAI's developer documentation now says there is no one-to-one replacement API. If you made videos with Sora, you need a new tool — and for most Sora workflows, Google's Veo 3.1 is the closest match: it also generates video and sound together from a prompt or an image, and it adds frame and reference controls Sora never had.
This article explains the timeline, what you lose, where Veo 3.1 is a like-for-like swap and where it is not, and how to adapt your prompts.
What happened to Sora 2
OpenAI released Sora 2 on September 30, 2025, together with a Sora social app. OpenAI pitched it on physical realism (missed basketball shots that rebound off the rim instead of teleporting into the hoop) and on synchronized dialogue and sound effects. The app's Cameo feature let people record themselves once and appear, with their voice, in generated scenes. In October 2025, OpenAI raised clip length to 15 seconds for all users and 25 seconds for Pro users on the web, and added Storyboards for Pro users.
The shutdown happened in stages:
| Date | What changed |
|---|---|
| March 24, 2026 | OpenAI announces it is discontinuing Sora in the app and the API, and notifies API developers |
| April 26, 2026 | Sora web and app experiences end |
| September 24, 2026 | sora-2, sora-2-pro and the Videos API are shut down |
Sources: OpenAI's API deprecations page and Sora 2 model pages, and OpenAI's Help Center article on the Sora discontinuation (checked September 28, 2026).
What Sora users lose
It helps to be precise about what made Sora useful, because that determines what to look for next:
- Text and image to video with sound. Sora 2 generated dialogue, sound effects and ambience with the picture.
- Longer single clips. Up to 15 seconds in the app for everyone, 25 seconds for Pro users on the web.
- Storyboards for planning a clip shot by shot (Pro).
- Cameos — putting your own face and voice into a scene.
- Physical realism that OpenAI made a headline feature.
- API access through
sora-2(720p) andsora-2-pro(up to 1080p), in portrait or landscape.
No single replacement covers all of it. Cameo-style self-insertion in particular has no direct equivalent in Veo 3.1, and Google's safety filters restrict generating real people from photos.
Why Veo 3.1 is the closest replacement
Veo 3.1 was released by Google DeepMind on October 15, 2025 and updated in January 2026. Here is how it lines up against what Sora 2 offered, based on Google's Gemini API documentation and OpenAI's model pages:
| Veo 3.1 (Google) | Sora 2 (before shutdown) | |
|---|---|---|
| Status | Available in the Gemini app, Flow, the Gemini API and Vertex AI | App closed April 26, 2026; API closed September 24, 2026 |
| Native audio | Yes, always on | Yes |
| Clip length | 4, 6 or 8 seconds | Up to 15 s (25 s for Pro on the web) |
| Longer sequences | Extension: +7 s per step, up to about 148 s total at 720p | — (Storyboards were for shot planning) |
| Resolution | 720p, 1080p, 4K | 720p (sora-2), up to 1080p (sora-2-pro) |
| Aspect ratios | 16:9, 9:16 | Landscape, portrait |
| Start from an image | Yes | Yes |
| First and last frame | Yes | No |
| Reference images | Up to 3 (Veo 3.1 and Veo 3.1 Fast) | Cameos (a recorded likeness) |
Where Veo 3.1 matches or goes further: sound generated with the picture, a higher resolution ceiling (Google describes its 1080p and 4K output as upscaled), vertical and horizontal formats, and more ways to pin a shot down — a start image, an optional end image, or up to three reference images for a consistent character or product.
Where you will need to adjust: single clips are shorter. Sora let you generate 15 to 25 seconds in one go; Veo 3.1 gives you up to 8 seconds per clip; to go longer, you extend a clip in Google's own tools or edit several clips together. If most of your Sora videos were single long takes, plan them as a sequence of short shots instead.
Moving your Sora prompts to Veo 3.1
If your Sora prompts were loose and mood-driven, expect to tighten them: Veo 3.1 rewards specific direction — Google highlights stronger prompt adherence as one of 3.1's main improvements. When you port a Sora prompt:
- Split long ideas into shots. A 20-second Sora concept becomes two or three 8-second shots, each with one main action.
- Name the camera. "Slow push-in," "handheld," "low-angle tracking shot" — say it instead of hoping for it.
- Write the sound. List the ambience, the key effect, and any spoken line in quotes. An 8-second clip fits a line or two of dialogue, not a speech.
- Fix the look with an image when it matters. If a character or product must stay the same across shots, start each shot from the same still, or use reference images.
- Write in English for the most predictable results; Google says other languages have not been fully evaluated.
For developers: the Sora API is gone, so pipelines need a new provider. Veo 3.1 is available through the Gemini API and Vertex AI as a paid preview, with Veo 3.1, Veo 3.1 Fast and Veo 3.1 Lite tiers. Veo 3 (the original 3.0 model) is marked deprecated in Google's API, so build on 3.1.
Which Veo 3.1 tier to start with
Google offers three Veo 3.1 tiers, and the right one depends on how final the clip needs to be:
- Veo 3.1 — the highest-fidelity tier, for shots that go into the final cut.
- Veo 3.1 Fast — a lower-cost everyday tier; on this site it is also the one that accepts up to three reference images.
- Veo 3.1 Lite — Google's most cost-effective tier, good for drafting prompts and testing ideas before spending more.
A practical routine for former Sora users: draft the prompt on Lite, then regenerate the version you like on Veo 3.1 at the resolution you need.
Using Veo 3.1 on Veo Video
Veo Video runs all three Veo 3.1 tiers. Every generation is an 8-second clip with sound, in 16:9 or 9:16, at 720p, 1080p or 4K, from text or from an image with an optional end frame. Clip extension and 4- or 6-second lengths are not available here. New accounts get 30 credits, enough for one Veo 3.1 Lite clip at 720p — a good way to test one of your old Sora prompts before committing. If you relied on Sora's longer single takes, the AI video generator also runs MiniMax H3 Max for longer clips in one pass.
For a feature-by-feature look at the two models, read Veo 3.1 vs Sora 2. If you are weighing other replacements, see Veo 3.1 vs Kling AI and Veo 3.1 vs Grok Imagine.