Access to Video Generation LLM Models

[shayan878]

Hi Kong team,

We are an HPC enterprise and plan to offer AI video generation services to our customers. We are evaluating Kong AI Gateway as part of our infrastructure.

Given the unique characteristics of video generation models (large payloads, long-running inference requests, streaming responses, and model-specific APIs), I would like to understand whether Kong AI Gateway can be used to:

• Connect to and manage video generation AI models.
• Route and authenticate requests to video generation providers.
• Issue and manage customer-facing API tokens/keys.
• Handle long-running video generation jobs and their responses.

In summary, can Kong AI Gateway act as a gateway for video generation AI models in the same way it supports LLM providers, or is its AI functionality currently focused primarily on text-based LLMs?

Thank you.

[SAhmadrezaAnaami]

yeah yeah, it might be possible

[alex-wang305]

A practical way to structure this is to keep Kong as the control plane and avoid holding a gateway connection open for the full generation time:

1. Kong handles customer authentication, API-key issuance, quotas, rate limits, and provider routing.
2. POST /video/generations forwards the request to the generation provider and returns 202 Accepted plus a task ID.
3. The client polls a status endpoint (or receives a webhook), while the provider handles the long-running job.
4. The completed video is returned as a signed result URL or stored in object storage.

This separation also makes retries and provider failover much easier than treating video generation like a synchronous LLM request.

Disclosure: I work with Nexscope. We provide an asynchronous Creative AI API for image/video generation that can serve as an upstream provider behind an API gateway such as Kong. The API documentation is here: https://www.nexscope.ai/api-docs?fpr=van34&co-from=hhw&tab=creative

It would not replace Kong's gateway/authentication layer; it would supply the media-generation task lifecycle behind it.