OpenAI Sora 2: Deprecated Video API and Shutdown Guide

OpenAI Sora is no longer a normal AI video recommendation. As of August 16, 2026, the consumer Sora web and app experiences have already been discontinued, and the remaining developer API has a fixed shutdown date. OpenAI says the Sora 2 models and Videos API will stop working on September 24, 2026. Its official deprecations table does not list a replacement.

That changes the decision completely. The important question is not whether Sora 2 can make impressive video. It can. The question is whether anyone should begin depending on it now. For a new product, agency workflow, marketing system, or creator pipeline, the answer is generally no. For an existing API customer with unfinished renders, assets to export, or a migration to complete, there is still a narrow and useful final window.

### The Sora app is already closed

OpenAI's [Sora discontinuation notice](https://help.openai.com/en/articles/20001152-what-to-know-about-the-sora-discontinuation) says the Sora web and app experiences were discontinued on April 26, 2026. This means older descriptions that tell people to open Sora through ChatGPT, subscribe to ChatGPT Pro for Sora access, download the social app, use a Sora feed, or create through a consumer web interface are now misleading.

OpenAI provides a sunset page where former users can request an export of content they created. The company recommends exporting as soon as possible and says data associated with Sora use will be permanently deleted after discontinuation and any final export window. Anyone who used the app should treat export as a data-preservation task, not as an optional cleanup item. Save the finished videos, original prompts, source assets, thumbnails, project notes, consent records, and any metadata needed to understand or reuse the work later.

Purchased ChatGPT or Sora credits do not restore the discontinued app. OpenAI's help article says eligible purchased credits may still be usable for Codex, but that is a separate product and should not be confused with continued Sora access.

### The API remains available only temporarily

The remaining product is OpenAI's developer-facing Videos API. According to the [official video generation guide](https://developers.openai.com/api/docs/guides/video-generation), the affected models are `sora-2`, `sora-2-pro`, `sora-2-2025-10-06`, `sora-2-2025-12-08`, and `sora-2-pro-2025-10-06`. OpenAI's [API deprecations page](https://developers.openai.com/api/docs/deprecations) says all of them, along with the Videos API itself, will be removed on September 24, 2026.

A deprecation is not simply a warning that a newer model exists. OpenAI defines shutdown as the point when the model or endpoint is no longer accessible. Code that still calls `/v1/videos` after the deadline should therefore be expected to fail. There is no replacement model in the official table, so teams cannot solve the problem by changing one model name. They need a real migration plan that covers the provider, API design, prompts, reference assets, video and audio quality, job handling, safety rules, pricing, storage, and user expectations.

The remaining period should be used deliberately. Freeze nonessential feature development, inventory every Sora dependency, export required assets, identify the final renders that genuinely need Sora, and compare alternative systems using the same test set. Set an internal cutoff before September 24 so there is time to handle failed jobs and unexpected integration differences.

### What the remaining Sora 2 API can do

Sora 2 remains technically capable. It generates short video with synchronized audio from natural-language prompts or image references. OpenAI describes the standard `sora-2` model as the faster option for exploration, concepting, rough cuts, social clips, and rapid iteration. `sora-2-pro` targets higher-quality production output and supports 1080p landscape or portrait exports. Both can generate clips up to 20 seconds.

The API supports an image reference that becomes the first frame of a generation. This is useful when a team wants to anchor an original environment, object, visual language, or approved brand asset. The input image must match the target resolution and use JPEG, PNG, or WebP. The current guide also says images with human faces are rejected, so an otherwise suitable campaign image can still fail if a person appears in it.

Reusable character assets help maintain the core appearance of a non-human subject such as an animal, mascot, or object across shots. OpenAI recommends short source clips of roughly two to four seconds, matching the output aspect ratio. A generation can include up to two characters. The character's name should be written exactly in the prompt as well as passed through its ID. Human-likeness character uploads are blocked by default, with any possible eligibility handled separately through OpenAI.

Extensions can continue a completed video while using the source clip as context for motion, camera direction, and scene continuity. Each extension can add up to 20 seconds, and a clip can be extended up to six times for a maximum total length of 120 seconds. Extensions do not support character assets or image references, so teams must not assume every control is available in every operation.

Targeted edits can change an existing video while attempting to preserve its original composition and continuity. OpenAI recommends making one focused adjustment at a time. A request such as changing the palette or adding one clearly defined object is more suitable than a long list that changes lighting, subject, action, timing, camera, and audio together. Editing uploaded external videos is limited to eligible customers, while editing a Sora-generated video can use its video ID.

The system is asynchronous. A create request returns a job rather than an immediate MP4. Applications can poll job status or receive `video.completed` and `video.failed` webhook events. Completed jobs expose the MP4 plus a thumbnail and spritesheet. Ordinary download availability is short, with the documentation stating a maximum of one hour, so production systems should copy results into their own durable storage immediately.

The Batch API supports offline render queues for shot lists and studio workflows. Batch requests use JSON, require assets to be uploaded ahead of time, and support image references by file ID or URL. Completed Batch videos are available for up to 24 hours. Batch is useful for essential final workloads and currently carries lower listed prices, but it should not encourage teams to expand a dependency that is weeks from shutdown.

### Current API pricing

OpenAI's [API pricing page](https://developers.openai.com/api/docs/pricing) lists video costs per generated second. Standard Sora 2 at 720p is $0.10 per second. Sora 2 Pro is $0.30 per second at 720p, $0.50 per second at 1024p, and $0.70 per second at 1080p. Listed Batch prices are $0.05, $0.15, $0.25, and $0.35 per second respectively.

A 20-second 1080p Sora 2 Pro generation therefore has a listed standard model cost of $14 before retries, rejected ideas, alternate takes, storage, engineering, review, editing, and replacement-platform migration. That does not make it automatically expensive or cheap. It means teams should calculate cost using the full approval rate, not only the price of the final clip. Ten attempts to find one usable shot cost ten times as many generated seconds.

Use 720p and shorter durations for prompt testing. Move only an approved composition to Sora 2 Pro and higher resolution. Batch pricing can reduce the model component for offline work, but teams must allow for its operational constraints and the 24-hour download window. Most importantly, do not spend engineering effort optimizing a Sora-only cost advantage without accounting for the imminent rewrite.

### Restrictions matter before you submit

The current developer guide says API output must be suitable for audiences under 18. It rejects copyrighted characters and copyrighted music. Real people, including public figures, cannot be generated through the standard workflow. Input images with human faces are rejected, and human-likeness character uploads are blocked by default. These restrictions apply to prompts, reference images, character sources, and audio intent.

These are not edge details. They determine whether common use cases are viable. A marketer cannot assume they can animate a spokesperson photo. A fan creator cannot assume a famous fictional character or song is accepted. A product team cannot build a public-figure parody feature around the standard API. An agency should clear rights and policy fit before paying people to prepare a campaign around a workflow the API rejects.

OpenAI has also documented provenance measures for Sora, including C2PA metadata and visible or invisible signals in its first-party experiences. Provenance is useful, but it does not replace editorial responsibility. Generated video can still be misleading when removed from context, and metadata can be lost across some editing or distribution workflows. Teams should retain internal records connecting every published asset to its prompt, sources, approvals, generation details, edits, and disclosure decision.

### The migration test should evaluate more than visual quality

It is tempting to compare Sora alternatives with one impressive prompt and choose the prettiest output. That is not enough for a production migration. Build a representative test pack containing realistic and stylized scenes, portrait and landscape formats, camera motion, speech, ambience, sound effects, image references, character consistency, editing, and failure cases. Use the same creative brief and acceptance criteria for every platform.

Score prompt adherence, temporal consistency, subject identity, physics, text rendering, audio synchronization, speech quality, editability, latency, error handling, throughput, API clarity, webhook reliability, storage windows, provenance, moderation, commercial terms, data controls, support, and total cost per approved second. Document which features require a redesigned workflow instead of assuming direct equivalence.

For applications, place the new provider behind an internal abstraction rather than scattering provider-specific model names and request shapes across the codebase. Map job creation, status, cancellation, callbacks, result retrieval, errors, asset retention, and cost reporting. Run both systems during a short validation period if policy, budget, and time allow. Then switch production, monitor failures, communicate changes, and remove Sora credentials, webhooks, scheduled jobs, and documentation before the official endpoint disappears.

### The WhatAI verdict

Sora 2 remains a strong historical video model and a useful temporary benchmark. Its synced audio, image references, reusable non-human characters, extensions, focused edits, high-resolution Pro output, and Batch support show why developers may want to finish work already in progress. None of those capabilities overcome the shutdown decision.

Existing users should export first, finish only essential approved work, and migrate early. New users should not choose Sora as the foundation of a long-term product. OpenAI has already closed the consumer experience, has set September 24, 2026 as the API end date, and has not named a replacement. The honest recommendation is to treat Sora as a sunset record with a migration deadline, not as a subscription feature or a platform with a future roadmap.

WhatAI Decision Box

OpenAI Sora 2 can generate short videos with synchronized audio from text or image inputs, but it is no longer a normal product recommendation. The Sora app closed on April 26, 2026, and the remaining API shuts down on September 24, 2026.

The remaining Videos API supports Sora 2 and Sora 2 Pro, clips up to 20 seconds, image references, non-human characters, extensions, targeted edits, webhooks, downloads, and Batch jobs. Standard pricing runs from $0.10 to $0.70 per generated second, but every integration has a fixed end date and no official replacement is listed.

Share your Sora migration plan, output experience, pricing observations, and preferred replacement with the WhatAI community.

Use Cases

Pricing varies by plan and region — see current pricing.

Plan features change — last updated: 2026-08-16.

OpenAI Sora 2 (Deprecated) Community Discussions

Explore community discussions. Ask and answer questions on OpenAI Sora 2 (Deprecated) to grow and learn together.

OpenAI Sora 2 (Deprecated) Pros & Cons

How to Get Results with OpenAI Sora 2 (Deprecated): Step-by-Step Workflow

1. Confirm the shutdown exposure

   List every application, scheduled job, stored video ID, prompt template, webhook, SDK call, and downstream workflow that depends on the Videos API or any Sora 2 model before doing more generation work.

2. Set a migration deadline

   Plan to stop production use before September 24, 2026, leaving time for failed renders, asset export, replacement testing, security review, user communication, and rollback rather than treating the shutdown day as the project deadline.

3. Export existing assets

   Download every required MP4, thumbnail, spritesheet, prompt, metadata record, consent record, and source reference into durable storage. Do not depend on short-lived download windows or continued access after shutdown.

4. Choose only essential renders

   Limit new Sora jobs to work that must be completed during the final API period. Avoid expanding features, adding new customers, or creating fresh dependencies that increase the migration burden.

5. Use the lowest practical tier

   Prototype with Sora 2 at 720p and short durations. Move only approved shots to Sora 2 Pro or higher resolutions, then use Batch pricing for suitable offline queues while allowing enough time to download the results.

6. Write a precise shot prompt

   Define shot type, subject, action, setting, lighting, camera movement, visual style, and desired audio. Keep the scene achievable and compliant, then test one variable at a time instead of combining many changes.

7. Submit and monitor the job

   Create the asynchronous render, store its job ID, then poll at a reasonable interval with backoff or use signed webhook events. Handle queued, in-progress, completed, failed, and expired states explicitly.

8. Review and save immediately

   Inspect video, audio, continuity, prompt adherence, visual artefacts, rights, policy compliance, and accessibility. Download the MP4 and supporting assets promptly to your own durable system.

9. Test replacement systems

   Run a representative set of prompts, image references, edits, audio needs, portrait and landscape outputs, latency tests, failure cases, and cost measurements against alternative platforms using a consistent evaluation rubric.

10. Remove the Sora dependency

    Switch production traffic, delete obsolete secrets and webhooks, update documentation and customer messaging, retain required audit records, and verify that no scheduled process still calls a Sora model or Videos API endpoint.

OpenAI Sora 2 (Deprecated) Gotchas and Limits to Know Before You Start

• The consumer Sora web and app experiences ended on April 26, 2026.
• The Videos API and all listed Sora 2 models shut down on September 24, 2026.
• OpenAI currently lists no recommended replacement for the deprecated video models or endpoint.
• A single generation can be up to 20 seconds; extensions can reach 120 seconds total.
• Sora 2 is limited to 720p, while higher resolutions require Sora 2 Pro.
• Standard pricing reaches $0.70 per generated second for Sora 2 Pro at 1080p.
• API generations must be suitable for audiences under 18 under the current guide.
• Copyrighted characters and copyrighted music are rejected.
• Real people, including public figures, cannot be generated through the standard API workflow.
• Input images containing human faces are rejected under the current API restrictions.
• Character uploads are intended for non-human subjects, with human likeness blocked by default.
• Image references must match the target video resolution and use JPEG, PNG, or WebP.
• Character source clips work best at two to four seconds and matching aspect ratios.
• Extensions do not support characters or image references.
• Render jobs are asynchronous and may take several minutes, especially at longer durations or 1080p.
• Ordinary download URLs last up to one hour; Batch outputs remain downloadable for up to 24 hours.
• Generated video and audio still require human review for artefacts, meaning, rights, safety, and suitability.

Which OpenAI Sora 2 (Deprecated) Feature Fits Your Use Case

Starter Prompts for OpenAI Sora 2 (Deprecated)

Is OpenAI Sora still available?

Not as a consumer web or app product. OpenAI discontinued the Sora web and app experiences on April 26, 2026. The Sora 2 models and Videos API remain temporarily available to eligible API users until September 24, 2026.

When will the Sora API shut down?

OpenAI says the Videos API, sora-2, sora-2-pro, and the listed Sora 2 snapshots will shut down on September 24, 2026. After that date, those models and endpoints will no longer be accessible.

Does OpenAI recommend a replacement for Sora 2?

No. OpenAI's official API deprecations table currently shows no recommended replacement for the Videos API, Sora 2, Sora 2 Pro, or their listed snapshots. Teams need to evaluate alternatives and redesign their integrations independently.

How much does the remaining Sora API cost?

OpenAI lists standard pricing per generated second: Sora 2 at 720p costs $0.10; Sora 2 Pro costs $0.30 at 720p, $0.50 at 1024p, and $0.70 at 1080p. Listed Batch prices are half those standard rates. Check the official pricing page before use.

What can Sora 2 generate through the API?

The API can create video with synchronized audio from natural-language prompts or an image reference. It also supports reusable non-human character assets, extensions, targeted edits, MP4 downloads, thumbnails, spritesheets, polling, webhooks, and offline Batch queues.

How long can a Sora 2 video be?

A single generation can be up to 20 seconds. A completed clip can be extended up to six times, with each extension adding up to 20 seconds, for a maximum total length of 120 seconds. Extensions do not support characters or image references.

Can Sora 2 generate real people through the API?

OpenAI's current API guide says real people, including public figures, cannot be generated. Input images containing human faces are rejected, and human-likeness character uploads are blocked by default. Eligible customers must contact OpenAI about any special access.

What content restrictions apply to the API?

OpenAI says API generations must be suitable for audiences under 18. Copyrighted characters and copyrighted music are rejected, real people cannot be generated, human faces in input images are rejected, and prompts, references, and transcripts must comply with the applicable policies.

How long are completed Sora files available to download?

OpenAI's guide says ordinary completed-video download URLs are valid for a maximum of one hour after generation, so files should be copied to durable storage promptly. Batch-generated videos are available for download for up to 24 hours after the batch completes.

Should a new project start on Sora 2 now?

Usually no. With a fixed shutdown date and no official replacement, Sora 2 is a poor foundation for a new long-term product. It may still be useful for completing an existing workload, exporting assets, running a controlled final batch, or comparing outputs during migration.

Related AI Models: LLMs, Multimodal Systems, and More Tools

Explore the Network