@effect-uai/fal 0.13.0

package/src/models.ts

@@ -0,0 +1,49 @@
+/**
+ * fal endpoint identifiers (as of September 2026). Unlike every other
+ * provider here, the "model" is a *path*: it selects the endpoint, and
+ * text-to-image and edit are separate endpoints of the same model
+ * family. Pass a `FalImageModel` to `generate` and a `FalImageEditModel`
+ * to `edit`.
+ *
+ * The `(string & {})` tail keeps autocomplete on the literals while
+ * accepting any string, so fal's whole catalogue works without an SDK
+ * update: the id is the model page's URL after `fal.ai/models/`.
+ *
+ * Read that literally. Whether an id carries the `fal-ai/` prefix varies
+ * even between generations of one model (`fal-ai/bytedance/seedream/v4.5/edit`
+ * has it, `bytedance/seedream/v5/pro/text-to-image` does not), and a
+ * wrong guess is fal's "Application not found", not a 404 you can read.
+ *
+ * Reference: https://fal.ai/models
+ */
+export type FalImageModel =
+  | "fal-ai/flux-2-pro"
+  | "fal-ai/flux/schnell"
+  | "fal-ai/flux/dev"
+  | "bytedance/seedream/v5/pro/text-to-image"
+  | "alibaba/qwen-image-3/text-to-image"
+  | "meta/muse-image/text-to-image"
+  | "openai/gpt-image-2"
+  | "fal-ai/nano-banana-2"
+  // eslint-disable-next-line @typescript-eslint/ban-types
+  | (string & {})
+
+/**
+ * Edit endpoints. A generate id sent to `edit` gets fal's 422 for an
+ * unknown field.
+ *
+ * Which wire field the references ride in differs per endpoint and the
+ * adapter works that out on its own, so any edit endpoint fal hosts
+ * works here, single-image ones included.
+ */
+export type FalImageEditModel =
+  | "fal-ai/flux-2-pro/edit"
+  | "bytedance/seedream/v5/pro/edit"
+  | "fal-ai/bytedance/seedream/v4.5/edit"
+  | "alibaba/qwen-image-3/edit"
+  | "fal-ai/nano-banana-2/edit"
+  | "openai/gpt-image-2/edit"
+  | "fal-ai/qwen-image-edit"
+  | "fal-ai/flux/dev/image-to-image"
+  // eslint-disable-next-line @typescript-eslint/ban-types
+  | (string & {})