feat(ai-gemini): add Gemini Omni 1.1 Flash (gemini-omni-1.1-flash)

.changeset/gemini-omni-1-1-flash.md

@@ -0,0 +1,7 @@
+---
+'@tanstack/ai-gemini': patch
+---
+
+Add Gemini Omni 1.1 Flash (`gemini-omni-1.1-flash`) as the GA Interactions video model.
+
+`geminiVideo('gemini-omni-1.1-flash')` uses the existing Interactions video path. `gemini-omni-flash-preview` stays as a deprecated alias until it shuts down on 2026-09-30. `modelOptions.resolution` maps onto `response_format.resolution` (`360p` | `720p` | `1080p` | `4k`, default 720p). Duration stays 3–10 seconds per call.

docs/config.json

@@ -520,7 +520,7 @@
           "label": "Video Generation",
           "to": "media/video-generation",
           "addedAt": "2026-04-15",
-          "updatedAt": "2026-08-20"
+          "updatedAt": "2026-08-31"
         },
         {
           "label": "Generation Hooks",

docs/media/video-generation.md

@@ -609,7 +609,7 @@ fal is the exception: `duration` is typed from `@fal-ai/client`'s
 
 #### Gemini Omni Flash (Interactions API) Model Options
 
-Gemini Omni Flash (`gemini-omni-flash-preview`) is Google's multimodal
+Gemini Omni Flash (`gemini-omni-1.1-flash`) is Google's multimodal
 video-generation model with conversational editing. It only serves the
 [Interactions API](https://ai.google.dev/gemini-api/docs/omni), and the same
 `geminiVideo()` adapter routes it automatically:
@@ -620,34 +620,45 @@ video-generation model with conversational editing. It only serves the
   When Google delivers by reference instead, the Files API URI passes through
   and needs your API key to download, like Veo.
 
-Clips are 720p at 24 FPS. `duration` accepts any value in the **3 to 10 second**
-range (fractional seconds included), defaulting to 10 seconds when omitted:
+`duration` accepts any value in the **3 to 10 second** range (fractional
+seconds included), defaulting to 10 seconds when omitted:
 
 - `availableDurations()` reports
   `{ kind: 'range', min: 3, max: 10, unit: 'seconds' }`.
 - Out-of-range `duration` values are rejected at job creation.
 - `snapDuration(n)` snaps raw seconds into the range, clamping to its bounds and
   rounding to whole seconds.
 
-The `size` option maps onto the interaction's output aspect ratio:
+The `size` option maps onto the interaction's output aspect ratio
+(`'16:9'` default, or `'9:16'`). Pass `modelOptions.resolution` to set
+`response_format.resolution`:
+
+- `'360p'`
+- `'720p'` (default)
+- `'1080p'` (upscaled)
+- `'4k'` (upscaled)
 
 ```typescript ignore
 import { generateVideo, getVideoJobStatus } from "@tanstack/ai";
 import { geminiVideo } from "@tanstack/ai-gemini";
 
-const adapter = geminiVideo("gemini-omni-flash-preview");
+const adapter = geminiVideo("gemini-omni-1.1-flash");
 
 const { jobId } = await generateVideo({
   adapter,
   prompt: "A woman playing violin outdoors at golden hour",
   size: "9:16", // aspect ratio: '16:9' (default) or '9:16'
   duration: 6, // 3-10 seconds; omit for the 10s default
+  modelOptions: { resolution: "1080p" }, // '360p' | '720p' | '1080p' | '4k'
 });
 
 const status = await getVideoJobStatus({ adapter, jobId });
 // status.url → 'data:video/mp4;base64,…' once completed
 ```
 
+`gemini-omni-flash-preview` still type-checks as a deprecated alias until
+it shuts down on 2026-09-30. Use `gemini-omni-1.1-flash` in new code.
+
 Image and video prompt parts are sent to the interaction as content blocks,
 grouped as images, then videos, then the text prompt (Omni doesn't use Veo's
 `metadata.role` routing), so you can condition the generation on stills or short
@@ -668,7 +679,7 @@ edits the video while preserving everything you didn't mention:
 import { generateVideo } from "@tanstack/ai";
 import { geminiVideo } from "@tanstack/ai-gemini";
 
-const adapter = geminiVideo("gemini-omni-flash-preview");
+const adapter = geminiVideo("gemini-omni-1.1-flash");
 
 // Turn 1: generate
 const first = await generateVideo({

examples/ts-react-media/src/components/OmniStudio.tsx

@@ -17,7 +17,7 @@ import type { MediaPromptPart } from '@tanstack/ai/client'
 import { generateVideoFn } from '@/lib/server-functions'
 import { readMediaFile, toImagePart, toVideoPart } from '@/lib/media'
 
-const OMNI_MODEL = 'gemini-omni-flash-preview'
+const OMNI_MODEL = 'gemini-omni-1.1-flash'
 /** Omni bills per second of generated video. */
 const PRICE_PER_SECOND = 0.1
 

examples/ts-react-media/src/components/VideoGenerator.tsx

@@ -115,7 +115,7 @@ export default function VideoGenerator({
   const omniInRun =
     selectedModel === 'all'
       ? geminiModels.length > 0
-      : selectedModel.startsWith('gemini-omni-flash-preview')
+      : selectedModel.startsWith('gemini-omni')
 
   // Each card runs its own generation, so the form is busy while any of them
   // reports that it is.

examples/ts-react-media/src/lib/models.ts

@@ -187,18 +187,18 @@ export const VIDEO_MODELS = [
     provider: 'fal' as const,
   },
   {
-    id: 'gemini-omni-flash-preview',
-    name: 'Gemini Omni Flash (Text-to-Video)',
+    id: 'gemini-omni-1.1-flash',
+    name: 'Gemini Omni 1.1 Flash (Text-to-Video)',
     description:
-      'Google multimodal video generation with conversational editing, via the Interactions API (3-10s, 720p)',
+      'Google multimodal video generation with conversational editing, via the Interactions API (3-10s, 360p/720p/1080p/4k)',
     mode: 'text-to-video' as const,
     provider: 'gemini' as const,
   },
   {
-    id: 'gemini-omni-flash-preview/image-to-video',
-    name: 'Gemini Omni Flash (Image-to-Video)',
+    id: 'gemini-omni-1.1-flash/image-to-video',
+    name: 'Gemini Omni 1.1 Flash (Image-to-Video)',
     description:
-      'Animate an image with Gemini Omni Flash via the Interactions API',
+      'Animate an image with Gemini Omni 1.1 Flash via the Interactions API',
     mode: 'image-to-video' as const,
     provider: 'gemini' as const,
   },

examples/ts-react-media/src/lib/server-functions.ts

@@ -447,12 +447,12 @@ function videoStreamForModel(data: VideoRequest): AsyncIterable<StreamChunk> {
     // Gemini Omni Flash (Interactions API, GEMINI_API_KEY). One model
     // serves both UI entries; it accepts text, image, AND video prompt
     // parts (sent as interaction content blocks: images, then videos,
-    // then text). Clips are 3–10s at 720p (default 10s when `duration`
-    // is omitted); `size` is the output aspect ratio. Passing
+    // then text). Clips are 3–10s (default 10s when `duration` is
+    // omitted); `size` is the output aspect ratio. Passing
     // `previous_interaction_id` chains a prompt onto a prior generation
     // for conversational editing.
-    case 'gemini-omni-flash-preview':
-    case 'gemini-omni-flash-preview/image-to-video': {
+    case 'gemini-omni-1.1-flash':
+    case 'gemini-omni-1.1-flash/image-to-video': {
       const prompt = asOmniPrompt(data.prompt)
       if (
         data.model.endsWith('/image-to-video') &&
@@ -466,7 +466,7 @@ function videoStreamForModel(data: VideoRequest): AsyncIterable<StreamChunk> {
       return generateVideo({
         stream: true,
         pollingInterval: VIDEO_POLL_INTERVAL_MS,
-        adapter: geminiVideo('gemini-omni-flash-preview'),
+        adapter: geminiVideo('gemini-omni-1.1-flash'),
         prompt,
         size: aspectRatio ?? '16:9',
         ...(duration !== undefined ? { duration } : {}),

packages/ai-gemini/src/adapters/video.ts

@@ -228,15 +228,18 @@ function interactionUsageToTokenUsage(
  * requires the API key (`x-goog-api-key` header or `?key=` query
  * parameter) to download.
  *
- * **Gemini Omni Flash** (`gemini-omni-flash-preview`) only serves the
- * Interactions API: `createVideoJob` creates a background interaction with
+ * **Gemini Omni Flash** (`gemini-omni-1.1-flash`, plus the deprecated
+ * `gemini-omni-flash-preview` alias) only serves the Interactions API:
+ * `createVideoJob` creates a background interaction with
  * `response_modalities: ['video']`, `getVideoStatus` polls it by id, and
  * `getVideoUrl` returns the inline base64 MP4 as a `data:` URL (or the
  * Files API URI when the server delivers by reference). Image and video
  * prompt parts are sent as interaction content blocks, grouped as images,
  * then videos, then the text prompt (interleaving is not preserved); pass
  * `modelOptions.previous_interaction_id` to conversationally edit a prior
- * Omni generation.
+ * Omni generation. `modelOptions.resolution` maps onto
+ * `response_format.resolution` (`'360p' | '720p' | '1080p' | '4k'`,
+ * default 720p).
  *
  * @experimental Video generation is an experimental feature and may change.
  */
@@ -344,9 +347,8 @@ export class GeminiVideoAdapter<
     >,
   ): Promise<VideoJobResult> {
     const { prompt, size, duration, logger } = options
-    const modelOptions = options.modelOptions as
-      | GeminiOmniVideoProviderOptions
-      | undefined
+    const { resolution, ...passthroughModelOptions } =
+      (options.modelOptions as GeminiOmniVideoProviderOptions | undefined) ?? {}
 
     try {
       const resolved = resolveMediaPrompt(prompt)
@@ -384,23 +386,25 @@ export class GeminiVideoAdapter<
         )
       }
 
-      // Aspect ratio and clip length ride on `response_format`. Duration is
-      // a `"<seconds>s"` string, accepted anywhere in the 3–10s range
-      // (fractional included) and defaulting to 10s when omitted — verified
-      // against the live API; the docs don't publish the range constraints.
+      // Aspect ratio, clip length, and resolution ride on `response_format`.
+      // Duration is a `"<seconds>s"` string, accepted anywhere in the 3–10s
+      // range (fractional included) and defaulting to 10s when omitted.
+      // Resolution is `'360p' | '720p' | '1080p' | '4k'` and defaults to
+      // 720p when omitted — https://ai.google.dev/gemini-api/docs/omni
       const responseFormat =
-        size !== undefined || duration !== undefined
+        size !== undefined || duration !== undefined || resolution !== undefined
           ? {
               response_format: {
                 type: 'video' as const,
                 ...(size !== undefined && { aspect_ratio: size }),
                 ...(duration !== undefined && { duration: `${duration}s` }),
+                ...(resolution !== undefined && { resolution }),
               },
             }
           : {}
 
       const interaction = await this.client.interactions.create({
-        ...modelOptions,
+        ...passthroughModelOptions,
         model: this.model,
         input: [{ type: 'user_input', content }],
         response_modalities: ['video'],
@@ -659,6 +663,12 @@ export class GeminiVideoAdapter<
   }
 }
 
+/** @deprecated Shuts down 2026-09-30. Use `gemini-omni-1.1-flash`. */
+export function createGeminiVideo(
+  model: 'gemini-omni-flash-preview',
+  apiKey: string,
+  config?: Omit<GeminiVideoConfig, 'apiKey'>,
+): GeminiVideoAdapter<'gemini-omni-flash-preview'>
 /**
  * Creates a Gemini video adapter with an explicit API key.
  * Type resolution happens here at the call site.
@@ -681,6 +691,11 @@ export class GeminiVideoAdapter<
  * });
  * ```
  */
+export function createGeminiVideo<TModel extends GeminiVideoModel>(
+  model: TModel,
+  apiKey: string,
+  config?: Omit<GeminiVideoConfig, 'apiKey'>,
+): GeminiVideoAdapter<TModel>
 export function createGeminiVideo<TModel extends GeminiVideoModel>(
   model: TModel,
   apiKey: string,
@@ -689,6 +704,11 @@ export function createGeminiVideo<TModel extends GeminiVideoModel>(
   return new GeminiVideoAdapter({ apiKey, ...config }, model)
 }
 
+/** @deprecated Shuts down 2026-09-30. Use `gemini-omni-1.1-flash`. */
+export function geminiVideo(
+  model: 'gemini-omni-flash-preview',
+  config?: Omit<GeminiVideoConfig, 'apiKey'>,
+): GeminiVideoAdapter<'gemini-omni-flash-preview'>
 /**
  * Creates a Gemini video adapter with automatic API key detection from environment variables.
  * Type resolution happens here at the call site.
@@ -719,6 +739,10 @@ export function createGeminiVideo<TModel extends GeminiVideoModel>(
  * const status = await getVideoJobStatus({ adapter, jobId });
  * ```
  */
+export function geminiVideo<TModel extends GeminiVideoModel>(
+  model: TModel,
+  config?: Omit<GeminiVideoConfig, 'apiKey'>,
+): GeminiVideoAdapter<TModel>
 export function geminiVideo<TModel extends GeminiVideoModel>(
   model: TModel,
   config?: Omit<GeminiVideoConfig, 'apiKey'>,

packages/ai-gemini/src/index.ts

@@ -112,6 +112,7 @@ export {
 export type {
   GeminiInteractionsVideoModel,
   GeminiOmniVideoProviderOptions,
+  GeminiOmniVideoResolution,
   GeminiVideoModel,
   GeminiVideoModelDurationByName,
   GeminiVideoModelInputModalitiesByName,

packages/ai-gemini/src/model-meta.ts

@@ -807,11 +807,44 @@ const VEO_3_1_LITE_PREVIEW = {
 >
 
 /**
- * Gemini Omni Flash — multimodal video generation with conversational
- * editing. Serves only the Interactions API (`generateContent` rejects it),
- * so it routes through the interactions-based path of the video adapter,
- * not Veo's `:predictLongRunning` flow. Pricing is per second of generated
- * video ($0.10/sec). 720p / 24 FPS, 3–10 second clips (default 10s).
+ * Gemini Omni 1.1 Flash — GA. Multimodal video generation with
+ * conversational editing. Serves only the Interactions API
+ * (`generateContent` rejects it), so it routes through the
+ * interactions-based path of the video adapter, not Veo's
+ * `:predictLongRunning` flow. Pricing is per second of generated video
+ * ($0.10/sec). 360p / 720p (default) / 1080p / 4k at 24 FPS, 3–10 second
+ * clips (default 10s).
+ * @see https://ai.google.dev/gemini-api/docs/models/gemini-omni-flash
+ * @experimental Omni video generation is an experimental feature and may change.
+ */
+const GEMINI_OMNI_1_1_FLASH = {
+  name: 'gemini-omni-1.1-flash',
+  max_input_tokens: 1_048_576,
+  max_output_tokens: 1,
+  supports: {
+    input: ['text', 'image', 'video'],
+    output: ['video', 'audio'],
+  },
+  pricing: {
+    input: {
+      normal: 0,
+    },
+    output: {
+      normal: 0.1,
+    },
+  },
+} as const satisfies ModelMeta<
+  GeminiToolConfigOptions &
+    GeminiSafetyOptions &
+    GeminiCommonConfigOptions &
+    GeminiCachedContentOptions
+>
+
+/**
+ * @deprecated `gemini-omni-flash-preview` shuts down on 2026-09-30. Use the
+ * GA id `gemini-omni-1.1-flash` instead. Kept in the model union so existing
+ * code still compiles until shutdown.
+ * @see https://ai.google.dev/gemini-api/docs/models/gemini-omni-flash
  * @experimental Omni video generation is an experimental feature and may change.
  */
 const GEMINI_OMNI_FLASH_PREVIEW = {
@@ -1143,15 +1176,19 @@ export const GEMINI_VIDEO_MODELS = [
   VEO_3_1_PREVIEW.name,
   VEO_3_1_FAST_PREVIEW.name,
   VEO_3_1_LITE_PREVIEW.name,
+  GEMINI_OMNI_1_1_FLASH.name,
+  // Deprecated alias — shuts down 2026-09-30.
   GEMINI_OMNI_FLASH_PREVIEW.name,
 ] as const
 
 /**
  * Video models served by the Interactions API rather than Veo's
- * `:predictLongRunning` operations flow.
+ * `:predictLongRunning` operations flow. GA id first; the trailing
+ * `-preview` id is a shutdown alias kept so existing code compiles.
  * @experimental Omni video generation is an experimental feature and may change.
  */
 export const GEMINI_INTERACTIONS_VIDEO_MODELS = [
+  GEMINI_OMNI_1_1_FLASH.name,
   GEMINI_OMNI_FLASH_PREVIEW.name,
 ] as const