@tanstack/ai-gemini 0.16.0 → 0.17.0

package/src/model-meta.ts

@@ -631,7 +631,11 @@ const IMAGEN_3 = {
     GeminiCommonConfigOptions &
     GeminiCachedContentOptions
 >
-/** 
+/**
+ * Veo video generation models. Pricing is per second of generated video
+ * (audio+video rate where the model supports audio).
+ * @experimental Veo video generation is an experimental feature and may change.
+ */
 const VEO_3_1_PREVIEW = {
   name: 'veo-3.1-generate-preview',
   max_input_tokens: 1024,
@@ -650,9 +654,9 @@ const VEO_3_1_PREVIEW = {
   },
 } as const satisfies ModelMeta<
   GeminiToolConfigOptions &
-  GeminiSafetyOptions &
-  GeminiGenerationConfigOptions &
-  GeminiCachedContentOptions
+    GeminiSafetyOptions &
+    GeminiCommonConfigOptions &
+    GeminiCachedContentOptions
 >
 
 const VEO_3_1_FAST_PREVIEW = {
@@ -673,9 +677,9 @@ const VEO_3_1_FAST_PREVIEW = {
   },
 } as const satisfies ModelMeta<
   GeminiToolConfigOptions &
-  GeminiSafetyOptions &
-  GeminiGenerationConfigOptions &
-  GeminiCachedContentOptions
+    GeminiSafetyOptions &
+    GeminiCommonConfigOptions &
+    GeminiCachedContentOptions
 >
 
 const VEO_3 = {
@@ -696,9 +700,9 @@ const VEO_3 = {
   },
 } as const satisfies ModelMeta<
   GeminiToolConfigOptions &
-  GeminiSafetyOptions &
-  GeminiGenerationConfigOptions &
-  GeminiCachedContentOptions
+    GeminiSafetyOptions &
+    GeminiCommonConfigOptions &
+    GeminiCachedContentOptions
 >
 
 const VEO_3_FAST = {
@@ -719,9 +723,9 @@ const VEO_3_FAST = {
   },
 } as const satisfies ModelMeta<
   GeminiToolConfigOptions &
-  GeminiSafetyOptions &
-  GeminiGenerationConfigOptions &
-  GeminiCachedContentOptions
+    GeminiSafetyOptions &
+    GeminiCommonConfigOptions &
+    GeminiCachedContentOptions
 >
 
 const VEO_2 = {
@@ -741,10 +745,10 @@ const VEO_2 = {
   },
 } as const satisfies ModelMeta<
   GeminiToolConfigOptions &
-  GeminiSafetyOptions &
-  GeminiGenerationConfigOptions &
-  GeminiCachedContentOptions
-> */
+    GeminiSafetyOptions &
+    GeminiCommonConfigOptions &
+    GeminiCachedContentOptions
+>
 
 const GEMINI_3_5_FLASH = {
   name: 'gemini-3.5-flash',
@@ -878,20 +882,17 @@ export const GEMINI_TTS_VOICES = [
 
 export type GeminiTTSVoice = (typeof GEMINI_TTS_VOICES)[number]
 
-/*   const GEMINI_AUDIO_MODELS = [
-  GEMINI_2_5_PRO_TTS.name,
-  GEMINI_2_5_FLASH_TTS.name,
-  GEMINI_2_5_FLASH_LIVE.name,
-  GEMINI_2_FLASH_LIVE.name,
-] as const
-
-  const GEMINI_VIDEO_MODELS = [
+/**
+ * Veo video generation models.
+ * @experimental Veo video generation is an experimental feature and may change.
+ */
+export const GEMINI_VIDEO_MODELS = [
   VEO_3_1_PREVIEW.name,
   VEO_3_1_FAST_PREVIEW.name,
   VEO_3.name,
   VEO_3_FAST.name,
   VEO_2.name,
-] as const */
+] as const
 
 // Manual type map for per-model provider options
 export type GeminiChatModelProviderOptionsByName = {

package/src/video/video-provider-options.ts

@@ -0,0 +1,126 @@
+/**
+ * Gemini Veo Video Generation Provider Options
+ *
+ * Based on https://ai.google.dev/gemini-api/docs/video
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+import type { DurationOptions } from '@tanstack/ai/adapters'
+import type { GenerateVideosConfig } from '@google/genai'
+import type { GEMINI_VIDEO_MODELS } from '../model-meta'
+
+/**
+ * Model type for Gemini Veo video generation.
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export type GeminiVideoModel = (typeof GEMINI_VIDEO_MODELS)[number]
+
+/**
+ * Supported aspect ratios for Veo video generation. This is the `size` value
+ * for the Gemini video adapter — Veo expresses output shape as an aspect
+ * ratio (plus an optional `resolution` in `modelOptions`), not pixel
+ * dimensions.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export type GeminiVideoSize = '16:9' | '9:16'
+
+/**
+ * Provider-specific options for Gemini Veo video generation.
+ *
+ * Derived from the SDK's `GenerateVideosConfig`, minus the fields the
+ * adapter manages itself:
+ * - `durationSeconds` — set via the typed top-level `duration` option
+ *   (use `adapter.snapDuration(seconds)` to coerce raw seconds)
+ * - `aspectRatio` — set via the top-level `size` option
+ * - `lastFrame` / `referenceImages` — set via image parts in the `prompt`
+ *   with `metadata.role: 'end_frame'` / `'reference'`
+ * - `httpOptions` / `abortSignal` — client-level transport concerns
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export type GeminiVideoProviderOptions = Omit<
+  GenerateVideosConfig,
+  | 'durationSeconds'
+  | 'aspectRatio'
+  | 'lastFrame'
+  | 'referenceImages'
+  | 'httpOptions'
+  | 'abortSignal'
+>
+
+/**
+ * Model-specific provider options mapping.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export type GeminiVideoModelProviderOptionsByName = {
+  [TModel in GeminiVideoModel]: GeminiVideoProviderOptions
+}
+
+/**
+ * Model-specific size (aspect ratio) mapping.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export type GeminiVideoModelSizeByName = {
+  [TModel in GeminiVideoModel]: GeminiVideoSize
+}
+
+/**
+ * Per-model prompt input modalities. Every Veo model accepts image
+ * conditioning inputs (first frame, last frame, reference images) alongside
+ * the text prompt.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export type GeminiVideoModelInputModalitiesByName = {
+  [TModel in GeminiVideoModel]: readonly ['image']
+}
+
+/**
+ * Per-model duration unions (seconds, as numbers — the API's
+ * `parameters.durationSeconds` field is numeric).
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export type GeminiVideoModelDurationByName = {
+  'veo-3.1-generate-preview': 4 | 6 | 8
+  'veo-3.1-fast-generate-preview': 4 | 6 | 8
+  'veo-3.0-generate-001': 4 | 6 | 8
+  'veo-3.0-fast-generate-001': 4 | 6 | 8
+  'veo-2.0-generate-001': 5 | 6 | 8
+}
+
+/**
+ * Runtime duration table backing `availableDurations()` / `snapDuration()`.
+ *
+ * Curated from the official Veo docs
+ * (https://ai.google.dev/gemini-api/docs/video) — the Gemini OpenAPI spec
+ * types the `:predictLongRunning` request's `parameters` as unconstrained,
+ * so it carries no per-model duration information to derive these from.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export const GEMINI_VIDEO_DURATIONS: {
+  readonly [TModel in GeminiVideoModel]: DurationOptions<
+    GeminiVideoModelDurationByName[TModel]
+  >
+} = {
+  'veo-3.1-generate-preview': { kind: 'discrete', values: [4, 6, 8] },
+  'veo-3.1-fast-generate-preview': { kind: 'discrete', values: [4, 6, 8] },
+  'veo-3.0-generate-001': { kind: 'discrete', values: [4, 6, 8] },
+  'veo-3.0-fast-generate-001': { kind: 'discrete', values: [4, 6, 8] },
+  'veo-2.0-generate-001': { kind: 'discrete', values: [5, 6, 8] },
+}
+
+/**
+ * Look up the duration options for a Veo model.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export function getGeminiVideoDurationOptions<TModel extends GeminiVideoModel>(
+  model: TModel,
+): DurationOptions<GeminiVideoModelDurationByName[TModel]> {
+  return GEMINI_VIDEO_DURATIONS[model]
+}