PLATFORMS Signal 94
AI Gateway adds asynchronous video generation with webhook, poll, and start/get options
↗
Developers can launch video generation without blocking a request, choosing a webhook, polling interval, or start/get pattern to retrieve the result later.
Video generation can take seconds or minutes, which often exceeds HTTP request timeouts in serverless environments. The asynchronous options let applications avoid those timeouts by decoupling the request from the rendering process and delivering results via webhook, poll, or later retrieval.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Asynchronous video generation can be used via a webhook, polling with generateVideo, or the startVideo/getVideoStatus pair.
Existing generateVideo calls remain unchanged and continue to block until the video is ready.
The operation returned by startVideo is JSON-serializable, allowing it to be stored in a database or passed through a queue for later status checks.
THE READ
What the cluster adds up to.
AI Gateway now provides asynchronous video generation alongside its original synchronous generateVideo call. Developers can select a webhook, polling, or startVideo/getVideoStatus pattern to avoid keeping a single HTTP request open for the full rendering time. This change directly addresses the problem that video generation may take seconds or minutes and can exceed request timeouts. The synchronous generateVideo option remains available for workloads that can keep a request open.
Adopting the webhook option requires the application to generate a shared token, store it, and expose an endpoint that can verify the incoming webhook as described in the documentation. The polling option needs the calling process to stay alive while the SDK sends status requests at the configured interval, adding responsibility for managing timeouts and retries. The startVideo/getVideoStatus approach requires persisting the operation identifier, which is JSON-serializable, so a later process or queue can check completion without the original caller staying active.
Environments that cannot keep a process running, such as short-lived serverless functions with no external storage, may find polling or startVideo/getVideoStatus impractical because they rely on the process being alive to poll or to later retrieve the operation. Systems that cannot receive inbound HTTP requests cannot use the webhook option without first exposing a public endpoint. When immediate video data is required and any extra latency is unacceptable, the original blocking generateVideo remains the only choice, though it is still subject to request timeout limits.
All asynchronous options support the same input modalities, text-to-video, image-to-video, reference-to-video, and the SDK returns GeneratedFile objects that provide uint8Array, base64, and mediaType representations. Existing generateVideo calls continue to work unchanged, preserving backward compatibility for scripts that can keep a request open. The documentation includes upgrade instructions for the AI SDK and AI Gateway provider to access these new features.
Written by elseif from the cluster below · checked for specifics the sources never containedTHE CLUSTER