Sora API alternative: how to migrate after the Sora 2 shutdown

OpenAI shut it down. OpenAI's deprecations page says developers using the Videos API were notified on 24 March 2026 that the Videos API and every Sora 2 model (sora-2, sora-2-pro and their dated snapshots) would be removed from the API on 24 September 2026, and the replacement column for each of them is empty. The Videos API reference now opens by saying the Sora 2 models and Videos API were shut down on 24 September 2026 and that no one-to-one replacement API is available. We checked both pages on 17 October 2026.

The consumer app went first. OpenAI's help center article on the Sora discontinuation gives 26 April 2026 as the end of the Sora web and app experiences, with the API following in September.

Azure was the last route, and it has closed too. Microsoft's Foundry model retirement schedule lists sora-2 version 2025-10-06 (Preview) as retired on 15 July 2026, and sora-2 version 2025-12-08 (Preview) with a retirement date of 15 October 2026 and no replacement. Microsoft's lifecycle policy says a retired model answers every request with 410 Gone. As of October 2026, there is no first-party way to call Sora 2 from code.

So any production code that called Sora, on OpenAI or Azure, now needs another video model behind another API.

It should keep the shape your code already has: an asynchronous job, a status you can poll, a callback when it finishes and a price you know before you send. Then the migration is a field mapping, not a rewrite. Check each of these on any provider:

Almost one to one on the common path. The KreatorFlow developer API also creates video with POST /v1/videos and returns an asynchronous job, which you check with GET /v1/jobs/{job_id} or hear about through a signed webhook. The differences are field names and values: duration instead of seconds, aspect_ratio and resolution instead of a pixel size, and a model alias that names a Seedance model instead of sora-2.

Every response uses the same envelope: success, then data on success, or code and error on failure. A create call answers 202 with data.job, your balance_tokens and quoted_token_cost, so you know what a clip costs before it renders. On Seedance 2.0, video jobs accept 4, 5, 6, 8, 10, 12 and 15 seconds, 480p, 720p, 1080p and 4K, and the aspect ratios 16:9, 9:16, 1:1, 4:3, 3:4 and 21:9 or adaptive. GET /v1/models lists every alias with its limits and needs no key, so your code can read them instead of hard-coding them.

Swap the URL, the key and three fields. Authenticate with Authorization: Bearer and your KreatorFlow API key, post JSON to https://kreatorflow.ai/v1/videos and send an Idempotency-Key header, so a retried request returns the same job instead of creating a second one. The request below asks Seedance 2.0 for an 8-second 16:9 clip at 720p with generated audio and a webhook.

Prompts can be long: Seedance 2.0 takes up to 16,000 characters. If a field is out of range, the API answers 400 with a specific code such as duration_not_supported, aspect_ratio_not_supported or resolution_not_supported, and the message lists the allowed values, so mapping mistakes show up on the first call rather than after a render.

Split size into an aspect ratio and a resolution, and turn seconds into a number. Sora took size as a pixel string (720x1280, 1280x720, 1024x1792 or 1792x1024) and seconds as a string ("4", "8" or "12"). KreatorFlow takes aspect_ratio (or ratio) and resolution as two fields, and duration as a number from the model's list.

All three Sora lengths are valid on Seedance 2.0 (4, 5, 6, 8, 10, 12 and 15 seconds). Seedance 2.0 Fast stops at 10 seconds, so send 12-second jobs to Seedance 2.0 or shorten them. The helper below turns a stored Sora request body into a KreatorFlow one; the two larger Sora sizes map to 1080p, the closest tier above them.

Call GET /v1/jobs/{job_id} until the status is final. Sora reported queued, in_progress, completed and failed. A KreatorFlow job moves through created, queued and processing and ends as completed, failed, canceled or blocked, so treat those last four as terminal. When a job is completed, job.providerOutputUrl holds the video and job.providerOutputExpiresAt says when that link stops working; GET /v1/jobs/{job_id}/output streams the same file with your key whenever you need it again.

Polling is rate limited per key and per job. A 429 response carries a Retry-After header in seconds: wait that long and try again, as the loop below does. A failed job carries errorCode and errorMessage, and the tokens reserved for it go back to your balance.

Send webhook_url (or callback_url) with the create call, as an HTTPS address. KreatorFlow posts a JSON event to it when the job reaches a terminal state: completed, failed, canceled or blocked. The event has an id, a type such as job.completed and the job under data.job, and the request carries the headers X-KreatorFlow-Event-Id, X-KreatorFlow-Event-Type, X-KreatorFlow-Job-Id, X-KreatorFlow-Timestamp and X-KreatorFlow-Signature.

Verify every call before you trust it. The signature header reads t=<timestamp>,v1=<hex>, where the hex value is an HMAC-SHA256 of the timestamp, a dot and the raw request body, keyed with your workspace's webhook signing secret from the console. Compute it over the raw body before any JSON parsing, compare in constant time and reject old timestamps. Use the event id to ignore duplicates, and keep a slow poll running for jobs whose callback never arrives.

Image references become references or upload ids. Sora's input_reference took one image, by URL or file id, as the first frame. On KreatorFlow you can pass a public image as { "type": "image_url", "url": "...", "role": "first_frame" } in references, or upload the file and send first_frame_upload_id (plus last_frame_upload_id for an end frame). Seedance 2.0 takes up to 9 image references, 3 video references and 3 audio references per job; plain reference images guide appearance and style. Frame images and reference images cannot be mixed in one job, and the API says so with a 400 before anything is billed.

Files go through uploads. Where Sora used file ids, KreatorFlow uses a three-step upload: POST /v1/uploads returns a direct_upload_url and its headers, you PUT the bytes there, and POST /v1/uploads/complete gives you the upload_id to reference.

Remix, edit and extend become /v1/video-edits. Sora had remix, edits and extensions endpoints. KreatorFlow has POST /v1/video-edits with edit_mode set to transform, restyle or extend, and the source clip as source_video_upload_id or a public HTTPS URL in source_video_url. Edits need a Seedance 2 model alias.

Characters become registered characters. Register a reusable character with POST /v1/characters (mode register and an image upload_id), store the character_id it returns and pass it in character_ids on /v1/videos. GET and DELETE /v1/characters/{character_id} manage them.

Every video job is priced by the second, at a rate set by the model and resolution. You prepay tokens (1 token = $0.05); a job costs the rate times its length, rounded up to the cent and then to the next whole token, and the create response returns that quote. Seedance 2.0 at 720p is $0.110 per second, so a 5-second clip is $0.55 (11 tokens) and a 12-second clip, Sora's longest, is $1.35. At 1080p the rate is $0.240 per second, so 10 seconds cost $2.40. Seedance 2.0 Fast at 720p is $0.084 per second, for drafts and batches.

Generated audio has no separate charge on the rate card. Video you send in does: a job with a reference video or an edit's source clip is billed on input plus output seconds at 60% of the normal rate, and never costs less than the same job without video. Its create response sets quoted_token_cost_is_maximum, because the input length is only measured once the job runs; the final charge uses the measured length and the rest is released.

For how these rates compare with other Seedance providers, setting by setting and dated, see the per-second comparison and the Seedance API pricing guide.

Mostly naming, a few endpoints with no equivalent, and the model itself. Plan for these before you switch:

Seedance and Seedream are ByteDance's video and image models. KreatorFlow is an official partner of BytePlus, ByteDance's enterprise AI platform, and the developer API runs on BytePlus, ByteDance's official AI platform. The same key also covers image jobs on POST /v1/images and speech and sound on POST /v1/audio, so moving off Sora can be the start of one integration rather than three.