Back to Blog
AI OrchestrationMulti-CloudCase Study
Jul 2, 2026 · 6 min read · Bonito Team

One Gateway, Every Creative Asset: Image and Video on Bonito

Most AI gateways stop at chat. You send messages, you get text back, and anything visual means a separate tool, a separate key, and a separate bill.

Bonito's gateway does not stop there. The same bn- API key that powers chat also generates images and video — with the same authentication, the same cost tracking, and the same audit log. One credential, every modality.

The endpoints

Alongside POST /v1/chat/completions, the gateway exposes:

  • POST /v1/images/generations — image generation and edits across dall-e-3, dall-e-2, gpt-image-1, and Google's nano-banana (Gemini 2.5 Flash Image).
  • POST /v1/videos, GET /v1/videos/{id}, and GET /v1/videos/{id}/content — submit, poll, and download generated video from OpenAI Sora-2 and Vertex AI Veo 2.0, 3.0, and 3.1.

Every one of them shares the same key prefix, the same policy enforcement, the same per-request cost logging, and the same immutable audit trail as your chat traffic. A creative team no longer juggles four dashboards, four keys, and four invoices to ship a campaign.

Your cloud, your spend

Image generation follows Bonito's bring-your-own-cloud priority. When you generate with nano-banana, the gateway uses *your* Google Vertex service account first, so the spend lands on *your* cloud and your data stays in your tenancy. The platform key is only a fallback for demo and unconnected orgs. Your creative spend shows up where your finance team already looks.

Redundancy that matters for creative

Creative pipelines run hot and cannot afford a single point of failure. Two things make the image path resilient:

  • A second image provider. Image generation and edits used to be pinned to one vendor, so a single billing cap could take down *all* image generation platform-wide. Now models matching nano-banana or Gemini route to Google's image model, so one provider's outage does not stop the line.
  • Transient-error retries with backoff. A momentary upstream 429 or 5xx no longer becomes a total outage — the gateway retries transient failures with short backoff, while real errors like billing or auth propagate immediately with no masking.

Real campaigns, one credential

This is not a lab demo. Creative studios like Memory Creative run campaign image and video generation through Bonito's gateway, and wineries like Peller Estates produce seasonal campaign assets on the same path — the identical bn- key their chat and agent workloads already use. Brief, ideate, produce, review, publish: all metered on one ledger, all in one audit trail.

Why it belongs in the gateway

An agent that can reason but cannot produce the asset is only half a worker. Putting image and video behind the same governed gateway as chat means an agent can draft the copy *and* generate the hero image *and* render the promo video, all under one set of rules, one budget, and one log. That is what it looks like when agents actually get work done — not just talk about it.

Connect a provider and try it on your own stack from Discover, or read the docs for the full request format.

Ready to manage your AI infrastructure?

Join teams using Bonito to connect, route, and optimize their AI stack.

Get started free
Bonito CLI

Bonito CLI

Deploy AI agents across any provider from one YAML file

Check it out on Product Hunt →

Related Articles