fal

fal is a media provider (image + video), so it's wired differently from the text providers above. Instead of building an AI SDK model with createOpenAICompatible, you hand its key to createFalMediaAdapter and drop it into createMediaLCR.

import { createFalMediaAdapter } from "ai-lcr";

const fal = createFalMediaAdapter({
  apiKey: process.env.FAL_KEY!,
  // optional: async-job poll knobs
  pollIntervalMs: 3000,   // default
  pollTimeoutMs: 300_000, // default; a timeout fails over to the next route
});

fal serves every model — image and video alike — through one async queue API (submit → poll → fetch). The adapter runs that loop for you, and a poll timeout fails over to the next provider in the route.

Model ids are fal's path-style ids, used as the route's externalId: fal-ai/nano-banana-2 (image), fal-ai/veo3.1 (video). fal's queue result does not carry a per-call price, so the cost comes from the pricing you declare on each route in the registry (often unit: "second" for video, unit: "image" or unit: "megapixel" for image).

registry: {
  "google/veo-3": {
    id: "google/veo-3",
    modality: "video",
    routes: [
      { provider: "fal", externalId: "fal-ai/veo3.1", pricing: { unit: "second", cents: 40 } },
    ],
  },
}

See Image & video for the full media flow and the bundled model catalog.