Video¶
Video generation is the most expensive call most teams make. If a provider you use still offers it in the OpenAI format, the gateway checks each call against your policies, charges it to the right team and records it, with the code you already have.
What's supported¶
OpenAI retired its own Videos API and the Sora 2 models on 24 September 2026, so an OpenAI provider no longer makes videos, through the gateway or directly. Google's Veo preview models were retired on 22 October 2026.
The gateway still accepts the OpenAI-format video paths and passes them on unchanged to a provider added with the OpenAI kind of API that serves them:
| Path | What it is |
|---|---|
POST /v1/videos, GET /v1/videos/{id}, GET /v1/videos/{id}/content, POST /v1/videos/{id}/remix |
Create a video job, check it, download the result, remix it |
POST /v1/video/generations |
The other OpenAI-format video path |
They aren't translated: a call naming a model on a provider that doesn't speak the OpenAI API gets a 400
bad_request. Whether a given provider serves either path is that provider's choice: check its own documentation.
How a call looks¶
The base address and key are the same as for the OpenAI SDK. The create
call names a model, so the gateway knows where to send it. Checking a job, listing jobs and downloading the result name
no model: send the x-provider header with the name of the provider that made the video, as listed under
Providers in the Control room. Without it those calls get a 404.
curl https://<your-gateway>/v1/videos \
-H "Authorization: Bearer $OPENAI_API_KEY" -H "Content-Type: application/json" \
-d '{"model": "<a video model your provider serves>", "prompt": "A paper boat on a canal, slow pan"}'
curl https://<your-gateway>/v1/videos/<video id> \
-H "Authorization: Bearer $OPENAI_API_KEY" -H "x-provider: <provider name>"
The SDK helpers that poll for you (create_and_poll, poll in the OpenAI Python SDK) poll without the header: poll
yourself and pass it with extra_headers.
Errors¶
The codes are those of Chat Completions. Specific to video:
| Status | Code | When |
|---|---|---|
| 400 | bad_request |
The model is on a provider that doesn't speak the OpenAI API |
| 400 | provider_override_not_allowed |
x-provider names a provider that doesn't serve the model |
| 404 | A check, list or download without x-provider |
|
| 404 | endpoint_not_found |
A near miss: /v1/video/generation (it's generations) |
The cost of each call is in Request logs.
Good to know¶
- Guardrails never scan the video itself.