Many Providers
Beyond language models, Rook’s generation features (image, video, 3D) run through a single typed seam that can carry any model from any vendor — with per-model pricing and per-route capability flags. A new model is a registration, not a release.
Providers
Section titled “Providers”| Provider | Lifecycle | Coverage | Billing |
|---|---|---|---|
| Gemini (direct) | sync | image · text | per-token |
| Veo (direct) | async | video | per-second |
| Google Enterprise (Vertex AI) | sync · async | image · video | billed to your organization’s Google Cloud project |
| fal.ai (aggregator) | sync · queue | image · video · 3D | per-MP · per-call |
| Replicate (aggregator) | prediction · poll | image · video · 3D | per-compute-sec |
| Tencent Hunyuan (planned) | async | 3D generation | per-call · tiered |
A curated roster
Section titled “A curated roster”You don’t have to wade through hundreds of models — Rook curates a working set:
- Image — Nano Banana (Gemini) · Flux 2 Pro · GPT-Image-2 Edit
- Video — Veo · Seedance 2.0 · Kling v3
- 3D — Hunyuan 3D 3.1 and mesh cleanup are planned roadmap lanes
Built to absorb new models
Section titled “Built to absorb new models”The seam was designed around six realities of working with many vendors — sync vs. queued lifecycles, per-route capabilities, mixed pricing models, per-provider options, URL-or-inline results, and different secret schemes — so adding a model is routine.
Bring your own key
Section titled “Bring your own key”BYOK throughout — one keyring, separate vaults. You use your own provider accounts; Rook just routes to them. Google Enterprise is the exception to keys: you sign in to your organization’s Google Cloud project instead.
When Google is the primary provider for an image or video model, a stored fal.ai key acts as a fallback if the Google key is missing or rejected. See RookVision for the exact rules.