@tanstack/ai-gemini 0.17.2 → 0.18.1

package/src/experimental/text-interactions/events.ts

@@ -15,55 +15,60 @@ export interface GeminiInteractionIdEvent extends Omit<
 }
 
 /**
- * `CUSTOM` event carrying a raw `google_search_call` content delta from the
+ * `CUSTOM` event carrying a raw `google_search_call` Step from the
  * Interactions API. Payload shape is owned by `@google/genai`.
+ *
+ * SDK 2.x: server-side tool activity is now first-class `Step`s (emitted
+ * via `step.start`) rather than content deltas, so each variant below
+ * carries the full Step (with its `id` / `call_id`) instead of a delta
+ * fragment.
  */
 export interface GeminiGoogleSearchCallEvent extends Omit<
   CustomEvent,
   'name' | 'value'
 > {
   name: 'gemini.googleSearchCall'
-  value: Interactions.ContentDelta.GoogleSearchCallDelta
+  value: Interactions.GoogleSearchCallStep
 }
 
 /**
- * `CUSTOM` event carrying a raw `google_search_result` content delta from
- * the Interactions API.
+ * `CUSTOM` event carrying a raw `google_search_result` Step from the
+ * Interactions API.
  */
 export interface GeminiGoogleSearchResultEvent extends Omit<
   CustomEvent,
   'name' | 'value'
 > {
   name: 'gemini.googleSearchResult'
-  value: Interactions.ContentDelta.GoogleSearchResultDelta
+  value: Interactions.GoogleSearchResultStep
 }
 
 /**
- * `CUSTOM` event carrying a raw `code_execution_call` content delta from
- * the Interactions API.
+ * `CUSTOM` event carrying a raw `code_execution_call` Step from the
+ * Interactions API.
  */
 export interface GeminiCodeExecutionCallEvent extends Omit<
   CustomEvent,
   'name' | 'value'
 > {
   name: 'gemini.codeExecutionCall'
-  value: Interactions.ContentDelta.CodeExecutionCallDelta
+  value: Interactions.CodeExecutionCallStep
 }
 
 /**
- * `CUSTOM` event carrying a raw `code_execution_result` content delta from
- * the Interactions API.
+ * `CUSTOM` event carrying a raw `code_execution_result` Step from the
+ * Interactions API.
  */
 export interface GeminiCodeExecutionResultEvent extends Omit<
   CustomEvent,
   'name' | 'value'
 > {
   name: 'gemini.codeExecutionResult'
-  value: Interactions.ContentDelta.CodeExecutionResultDelta
+  value: Interactions.CodeExecutionResultStep
 }
 
 /**
- * `CUSTOM` event carrying a raw `url_context_call` content delta from the
+ * `CUSTOM` event carrying a raw `url_context_call` Step from the
  * Interactions API.
  */
 export interface GeminiUrlContextCallEvent extends Omit<
@@ -71,11 +76,11 @@ export interface GeminiUrlContextCallEvent extends Omit<
   'name' | 'value'
 > {
   name: 'gemini.urlContextCall'
-  value: Interactions.ContentDelta.URLContextCallDelta
+  value: Interactions.URLContextCallStep
 }
 
 /**
- * `CUSTOM` event carrying a raw `url_context_result` content delta from the
+ * `CUSTOM` event carrying a raw `url_context_result` Step from the
  * Interactions API.
  */
 export interface GeminiUrlContextResultEvent extends Omit<
@@ -83,11 +88,11 @@ export interface GeminiUrlContextResultEvent extends Omit<
   'name' | 'value'
 > {
   name: 'gemini.urlContextResult'
-  value: Interactions.ContentDelta.URLContextResultDelta
+  value: Interactions.URLContextResultStep
 }
 
 /**
- * `CUSTOM` event carrying a raw `file_search_call` content delta from the
+ * `CUSTOM` event carrying a raw `file_search_call` Step from the
  * Interactions API.
  */
 export interface GeminiFileSearchCallEvent extends Omit<
@@ -95,11 +100,11 @@ export interface GeminiFileSearchCallEvent extends Omit<
   'name' | 'value'
 > {
   name: 'gemini.fileSearchCall'
-  value: Interactions.ContentDelta.FileSearchCallDelta
+  value: Interactions.FileSearchCallStep
 }
 
 /**
- * `CUSTOM` event carrying a raw `file_search_result` content delta from the
+ * `CUSTOM` event carrying a raw `file_search_result` Step from the
  * Interactions API.
  */
 export interface GeminiFileSearchResultEvent extends Omit<
@@ -107,7 +112,7 @@ export interface GeminiFileSearchResultEvent extends Omit<
   'name' | 'value'
 > {
   name: 'gemini.fileSearchResult'
-  value: Interactions.ContentDelta.FileSearchResultDelta
+  value: Interactions.FileSearchResultStep
 }
 
 /**
@@ -127,12 +132,13 @@ export interface GeminiFileSearchResultEvent extends Omit<
  * }
  * ```
  *
- * The four tool-delta variant pairs (`google_search`, `code_execution`,
+ * The four tool variant pairs (`google_search`, `code_execution`,
  * `url_context`, `file_search`) forward the raw
- * `Interactions.ContentDelta.*Delta` payload from `@google/genai` so
- * consumers stay in sync with SDK shape changes automatically.
+ * `Interactions.*CallStep` / `Interactions.*ResultStep` payload from
+ * `@google/genai`. SDK 2.x emits these as full Steps via `step.start`
+ * (carrying the call's `id` / `call_id`) rather than as content deltas.
  * `computer_use` is accepted as a request tool but the API does not
- * stream per-delta CUSTOM events for it.
+ * surface a dedicated CUSTOM event for it.
  */
 export type GeminiInteractionsCustomEvent =
   | GeminiInteractionIdEvent

package/src/experimental/text-interactions/provider-options.ts

@@ -21,6 +21,5 @@ export type ExternalTextInteractionsProviderOptions = Pick<
   | 'system_instruction'
   | 'response_modalities'
   | 'response_format'
-  | 'response_mime_type'
   | 'generation_config'
 >

package/dist/esm/adapters/audio.d.ts

@@ -1,92 +0,0 @@
-import { BaseAudioAdapter } from '@tanstack/ai/adapters';
-import { GEMINI_AUDIO_MODELS } from '../model-meta.js';
-import { AudioGenerationOptions, AudioGenerationResult } from '@tanstack/ai';
-import { GeminiClientConfig } from '../utils.js';
-/**
- * Provider options for Gemini Lyria music generation.
- *
- * Notes on the Lyria 3 surface area:
- * - `lyria-3-clip-preview` always returns MP3 (30-second clips). It does
- *   not accept `responseMimeType`, and duration is fixed at 30 seconds —
- *   the generic `duration` option on `AudioActivityOptions` is ignored.
- * - `lyria-3-pro-preview` returns MP3 by default. Duration is controlled
- *   via the natural-language prompt, not a separate SDK field, so the
- *   generic `duration` option is similarly ignored.
- * - `negativePrompt` is NOT accepted by `GenerateContentConfig` and has
- *   therefore been removed from this surface to avoid giving callers a
- *   silently-dropped knob.
- *
- * @see https://ai.google.dev/gemini-api/docs/music-generation
- */
-export interface GeminiAudioProviderOptions {
-    /**
-     * Seed for deterministic generation.
-     */
-    seed?: number;
-}
-export interface GeminiAudioConfig extends GeminiClientConfig {
-}
-/** Model type for Gemini Lyria audio generation */
-export type GeminiAudioModel = (typeof GEMINI_AUDIO_MODELS)[number];
-/**
- * Gemini Lyria Music Generation Adapter.
- *
- * Tree-shakeable adapter for Google Lyria music generation via the Gemini API.
- *
- * Models:
- * - `lyria-3-pro-preview` — flagship model, full-length songs with verses,
- *   choruses, and bridges. Outputs MP3 or WAV at 48 kHz stereo.
- * - `lyria-3-clip-preview` — 30-second clips in MP3.
- *
- * @see https://ai.google.dev/gemini-api/docs/music-generation
- *
- * @example
- * ```typescript
- * const adapter = geminiAudio('lyria-3-pro-preview')
- * const result = await generateAudio({
- *   adapter,
- *   prompt: 'An upbeat jazz track with saxophone and drums',
- * })
- * ```
- */
-export declare class GeminiAudioAdapter<TModel extends GeminiAudioModel> extends BaseAudioAdapter<TModel, GeminiAudioProviderOptions> {
-    readonly name: "gemini";
-    private readonly client;
-    constructor(config: GeminiAudioConfig, model: TModel);
-    generateAudio(options: AudioGenerationOptions<GeminiAudioProviderOptions>): Promise<AudioGenerationResult>;
-}
-/**
- * Creates a Gemini Lyria audio adapter with an explicit API key.
- *
- * @param model - The Lyria model name (e.g., 'lyria-3-pro-preview')
- * @param apiKey - Your Google API key
- * @param config - Optional additional configuration
- *
- * @example
- * ```typescript
- * const adapter = createGeminiAudio('lyria-3-pro-preview', 'your-api-key')
- * const result = await generateAudio({
- *   adapter,
- *   prompt: 'Ambient electronic music with soft pads',
- * })
- * ```
- */
-export declare function createGeminiAudio<TModel extends GeminiAudioModel>(model: TModel, apiKey: string, config?: Omit<GeminiAudioConfig, 'apiKey'>): GeminiAudioAdapter<TModel>;
-/**
- * Creates a Gemini Lyria audio adapter with automatic API key detection.
- *
- * Looks for `GOOGLE_API_KEY` or `GEMINI_API_KEY` in the environment.
- *
- * @param model - The Lyria model name (e.g., 'lyria-3-pro-preview')
- * @param config - Optional configuration (excluding apiKey)
- *
- * @example
- * ```typescript
- * const adapter = geminiAudio('lyria-3-pro-preview')
- * const result = await generateAudio({
- *   adapter,
- *   prompt: 'An orchestral piece with strings and brass',
- * })
- * ```
- */
-export declare function geminiAudio<TModel extends GeminiAudioModel>(model: TModel, config?: Omit<GeminiAudioConfig, 'apiKey'>): GeminiAudioAdapter<TModel>;

package/dist/esm/adapters/audio.js

@@ -1,66 +0,0 @@
-import { BaseAudioAdapter } from "@tanstack/ai/adapters";
-import { createGeminiClient, generateId, getGeminiApiKeyFromEnv } from "../utils/client.js";
-import { buildGeminiUsage } from "../usage.js";
-class GeminiAudioAdapter extends BaseAudioAdapter {
-  name = "gemini";
-  client;
-  constructor(config, model) {
-    super(model, config);
-    this.client = createGeminiClient(config);
-  }
-  async generateAudio(options) {
-    const { model, prompt, modelOptions, logger } = options;
-    logger.request(`activity=generateAudio provider=gemini model=${model}`, {
-      provider: "gemini",
-      model
-    });
-    try {
-      const response = await this.client.models.generateContent({
-        model,
-        contents: [{ role: "user", parts: [{ text: prompt }] }],
-        config: {
-          responseModalities: ["AUDIO", "TEXT"],
-          ...modelOptions?.seed != null ? { seed: modelOptions.seed } : {}
-        }
-      });
-      const parts = response.candidates?.[0]?.content?.parts ?? [];
-      const audioPart = parts.find(
-        (part) => part.inlineData?.mimeType?.startsWith("audio/")
-      );
-      if (!audioPart?.inlineData?.data) {
-        throw new Error("No audio data in Gemini Lyria response");
-      }
-      const contentType = audioPart.inlineData.mimeType;
-      return {
-        id: generateId(this.name),
-        model,
-        audio: {
-          b64Json: audioPart.inlineData.data,
-          ...contentType !== void 0 && { contentType }
-        },
-        // Surface token usage (with per-modality breakdown) when Gemini reports
-        // it. Spread conditionally for exactOptionalPropertyTypes.
-        ...response.usageMetadata ? { usage: buildGeminiUsage(response.usageMetadata) } : {}
-      };
-    } catch (error) {
-      logger.errors("gemini.generateAudio fatal", {
-        error,
-        source: "gemini.generateAudio"
-      });
-      throw error;
-    }
-  }
-}
-function createGeminiAudio(model, apiKey, config) {
-  return new GeminiAudioAdapter({ ...config, apiKey }, model);
-}
-function geminiAudio(model, config) {
-  const apiKey = getGeminiApiKeyFromEnv();
-  return createGeminiAudio(model, apiKey, config);
-}
-export {
-  GeminiAudioAdapter,
-  createGeminiAudio,
-  geminiAudio
-};
-//# sourceMappingURL=audio.js.map

package/dist/esm/adapters/audio.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"audio.js","sources":["../../../src/adapters/audio.ts"],"sourcesContent":["import { BaseAudioAdapter } from '@tanstack/ai/adapters'\nimport {\n  createGeminiClient,\n  generateId,\n  getGeminiApiKeyFromEnv,\n} from '../utils'\nimport { buildGeminiUsage } from '../usage'\nimport type { GEMINI_AUDIO_MODELS } from '../model-meta'\nimport type {\n  AudioGenerationOptions,\n  AudioGenerationResult,\n} from '@tanstack/ai'\nimport type { GoogleGenAI } from '@google/genai'\nimport type { GeminiClientConfig } from '../utils'\n\n/**\n * Provider options for Gemini Lyria music generation.\n *\n * Notes on the Lyria 3 surface area:\n * - `lyria-3-clip-preview` always returns MP3 (30-second clips). It does\n *   not accept `responseMimeType`, and duration is fixed at 30 seconds —\n *   the generic `duration` option on `AudioActivityOptions` is ignored.\n * - `lyria-3-pro-preview` returns MP3 by default. Duration is controlled\n *   via the natural-language prompt, not a separate SDK field, so the\n *   generic `duration` option is similarly ignored.\n * - `negativePrompt` is NOT accepted by `GenerateContentConfig` and has\n *   therefore been removed from this surface to avoid giving callers a\n *   silently-dropped knob.\n *\n * @see https://ai.google.dev/gemini-api/docs/music-generation\n */\nexport interface GeminiAudioProviderOptions {\n  /**\n   * Seed for deterministic generation.\n   */\n  seed?: number\n}\n\nexport interface GeminiAudioConfig extends GeminiClientConfig {}\n\n/** Model type for Gemini Lyria audio generation */\nexport type GeminiAudioModel = (typeof GEMINI_AUDIO_MODELS)[number]\n\n/**\n * Gemini Lyria Music Generation Adapter.\n *\n * Tree-shakeable adapter for Google Lyria music generation via the Gemini API.\n *\n * Models:\n * - `lyria-3-pro-preview` — flagship model, full-length songs with verses,\n *   choruses, and bridges. Outputs MP3 or WAV at 48 kHz stereo.\n * - `lyria-3-clip-preview` — 30-second clips in MP3.\n *\n * @see https://ai.google.dev/gemini-api/docs/music-generation\n *\n * @example\n * ```typescript\n * const adapter = geminiAudio('lyria-3-pro-preview')\n * const result = await generateAudio({\n *   adapter,\n *   prompt: 'An upbeat jazz track with saxophone and drums',\n * })\n * ```\n */\nexport class GeminiAudioAdapter<\n  TModel extends GeminiAudioModel,\n> extends BaseAudioAdapter<TModel, GeminiAudioProviderOptions> {\n  readonly name = 'gemini' as const\n\n  private readonly client: GoogleGenAI\n\n  constructor(config: GeminiAudioConfig, model: TModel) {\n    super(model, config)\n    this.client = createGeminiClient(config)\n  }\n\n  async generateAudio(\n    options: AudioGenerationOptions<GeminiAudioProviderOptions>,\n  ): Promise<AudioGenerationResult> {\n    const { model, prompt, modelOptions, logger } = options\n\n    logger.request(`activity=generateAudio provider=gemini model=${model}`, {\n      provider: 'gemini',\n      model,\n    })\n\n    try {\n      // FIXME (SDK audit): Lyria 3 music generation may not belong on\n      // generateContent at all — @google/genai exposes a `LiveMusicSession`\n      // (`ai.live.music.connect`) with a `musicGenerationConfig` object.\n      // `seed` is valid on GenerateContentConfig, and Lyria always returns\n      // MP3 today, so we don't forward `responseMimeType` either.\n      // The runtime test `emits only GenerateContentConfig-valid fields`\n      // asserts the config shape so a later SDK audit can catch regressions.\n      const response = await this.client.models.generateContent({\n        model,\n        contents: [{ role: 'user', parts: [{ text: prompt }] }],\n        config: {\n          responseModalities: ['AUDIO', 'TEXT'],\n          ...(modelOptions?.seed != null ? { seed: modelOptions.seed } : {}),\n        },\n      })\n\n      const parts = response.candidates?.[0]?.content?.parts ?? []\n      const audioPart = parts.find((part: any) =>\n        part.inlineData?.mimeType?.startsWith('audio/'),\n      )\n\n      if (!audioPart?.inlineData?.data) {\n        throw new Error('No audio data in Gemini Lyria response')\n      }\n\n      // audioPart was selected because mimeType.startsWith('audio/') was\n      // truthy, so the mime type is guaranteed to be a string here. Trust the\n      // value Gemini returned rather than inventing a non-standard\n      // `audio/mp3` fallback (IANA is `audio/mpeg`).\n      const contentType = audioPart.inlineData.mimeType\n\n      return {\n        id: generateId(this.name),\n        model,\n        audio: {\n          b64Json: audioPart.inlineData.data,\n          ...(contentType !== undefined && { contentType }),\n        },\n        // Surface token usage (with per-modality breakdown) when Gemini reports\n        // it. Spread conditionally for exactOptionalPropertyTypes.\n        ...(response.usageMetadata\n          ? { usage: buildGeminiUsage(response.usageMetadata) }\n          : {}),\n      }\n    } catch (error) {\n      logger.errors('gemini.generateAudio fatal', {\n        error,\n        source: 'gemini.generateAudio',\n      })\n      throw error\n    }\n  }\n}\n\n/**\n * Creates a Gemini Lyria audio adapter with an explicit API key.\n *\n * @param model - The Lyria model name (e.g., 'lyria-3-pro-preview')\n * @param apiKey - Your Google API key\n * @param config - Optional additional configuration\n *\n * @example\n * ```typescript\n * const adapter = createGeminiAudio('lyria-3-pro-preview', 'your-api-key')\n * const result = await generateAudio({\n *   adapter,\n *   prompt: 'Ambient electronic music with soft pads',\n * })\n * ```\n */\nexport function createGeminiAudio<TModel extends GeminiAudioModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<GeminiAudioConfig, 'apiKey'>,\n): GeminiAudioAdapter<TModel> {\n  // Put apiKey LAST so caller-supplied config can't silently override the\n  // explicit argument.\n  return new GeminiAudioAdapter({ ...config, apiKey }, model)\n}\n\n/**\n * Creates a Gemini Lyria audio adapter with automatic API key detection.\n *\n * Looks for `GOOGLE_API_KEY` or `GEMINI_API_KEY` in the environment.\n *\n * @param model - The Lyria model name (e.g., 'lyria-3-pro-preview')\n * @param config - Optional configuration (excluding apiKey)\n *\n * @example\n * ```typescript\n * const adapter = geminiAudio('lyria-3-pro-preview')\n * const result = await generateAudio({\n *   adapter,\n *   prompt: 'An orchestral piece with strings and brass',\n * })\n * ```\n */\nexport function geminiAudio<TModel extends GeminiAudioModel>(\n  model: TModel,\n  config?: Omit<GeminiAudioConfig, 'apiKey'>,\n): GeminiAudioAdapter<TModel> {\n  const apiKey = getGeminiApiKeyFromEnv()\n  return createGeminiAudio(model, apiKey, config)\n}\n"],"names":[],"mappings":";;;AAgEO,MAAM,2BAEH,iBAAqD;AAAA,EACpD,OAAO;AAAA,EAEC;AAAA,EAEjB,YAAY,QAA2B,OAAe;AACpD,UAAM,OAAO,MAAM;AACnB,SAAK,SAAS,mBAAmB,MAAM;AAAA,EACzC;AAAA,EAEA,MAAM,cACJ,SACgC;AAChC,UAAM,EAAE,OAAO,QAAQ,cAAc,WAAW;AAEhD,WAAO,QAAQ,gDAAgD,KAAK,IAAI;AAAA,MACtE,UAAU;AAAA,MACV;AAAA,IAAA,CACD;AAED,QAAI;AAQF,YAAM,WAAW,MAAM,KAAK,OAAO,OAAO,gBAAgB;AAAA,QACxD;AAAA,QACA,UAAU,CAAC,EAAE,MAAM,QAAQ,OAAO,CAAC,EAAE,MAAM,OAAA,CAAQ,GAAG;AAAA,QACtD,QAAQ;AAAA,UACN,oBAAoB,CAAC,SAAS,MAAM;AAAA,UACpC,GAAI,cAAc,QAAQ,OAAO,EAAE,MAAM,aAAa,SAAS,CAAA;AAAA,QAAC;AAAA,MAClE,CACD;AAED,YAAM,QAAQ,SAAS,aAAa,CAAC,GAAG,SAAS,SAAS,CAAA;AAC1D,YAAM,YAAY,MAAM;AAAA,QAAK,CAAC,SAC5B,KAAK,YAAY,UAAU,WAAW,QAAQ;AAAA,MAAA;AAGhD,UAAI,CAAC,WAAW,YAAY,MAAM;AAChC,cAAM,IAAI,MAAM,wCAAwC;AAAA,MAC1D;AAMA,YAAM,cAAc,UAAU,WAAW;AAEzC,aAAO;AAAA,QACL,IAAI,WAAW,KAAK,IAAI;AAAA,QACxB;AAAA,QACA,OAAO;AAAA,UACL,SAAS,UAAU,WAAW;AAAA,UAC9B,GAAI,gBAAgB,UAAa,EAAE,YAAA;AAAA,QAAY;AAAA;AAAA;AAAA,QAIjD,GAAI,SAAS,gBACT,EAAE,OAAO,iBAAiB,SAAS,aAAa,MAChD,CAAA;AAAA,MAAC;AAAA,IAET,SAAS,OAAO;AACd,aAAO,OAAO,8BAA8B;AAAA,QAC1C;AAAA,QACA,QAAQ;AAAA,MAAA,CACT;AACD,YAAM;AAAA,IACR;AAAA,EACF;AACF;AAkBO,SAAS,kBACd,OACA,QACA,QAC4B;AAG5B,SAAO,IAAI,mBAAmB,EAAE,GAAG,QAAQ,OAAA,GAAU,KAAK;AAC5D;AAmBO,SAAS,YACd,OACA,QAC4B;AAC5B,QAAM,SAAS,uBAAA;AACf,SAAO,kBAAkB,OAAO,QAAQ,MAAM;AAChD;"}

package/dist/esm/adapters/image.d.ts

@@ -1,101 +0,0 @@
-import { BaseImageAdapter } from '@tanstack/ai/adapters';
-import { GEMINI_IMAGE_MODELS } from '../model-meta.js';
-import { GeminiImageModelInputModalitiesByName, GeminiImageModelProviderOptionsByName, GeminiImageModelSizeByName, GeminiImageProviderOptions } from '../image/image-provider-options.js';
-import { ImageGenerationOptions, ImageGenerationResult } from '@tanstack/ai';
-import { GeminiClientConfig } from '../utils.js';
-/**
- * Configuration for Gemini image adapter
- */
-export interface GeminiImageConfig extends GeminiClientConfig {
-}
-/** Model type for Gemini Image */
-export type GeminiImageModel = (typeof GEMINI_IMAGE_MODELS)[number];
-/**
- * Gemini Image Generation Adapter
- *
- * Tree-shakeable adapter for Gemini image generation functionality.
- * Supports Imagen 3/4 models (via generateImages API) and Gemini native
- * image models like Nano Banana 2 (via generateContent API).
- *
- * Features:
- * - Aspect ratio-based image sizing
- * - Person generation controls
- * - Safety filtering
- * - Watermark options
- * - Extended resolution tiers (Nano Banana 2)
- */
-export declare class GeminiImageAdapter<TModel extends GeminiImageModel> extends BaseImageAdapter<TModel, GeminiImageProviderOptions, GeminiImageModelProviderOptionsByName, GeminiImageModelSizeByName, GeminiImageModelInputModalitiesByName> {
-    readonly kind: "image";
-    readonly name: "gemini";
-    '~types': {
-        providerOptions: GeminiImageProviderOptions;
-        modelProviderOptionsByName: GeminiImageModelProviderOptionsByName;
-        modelSizeByName: GeminiImageModelSizeByName;
-        modelInputModalitiesByName: GeminiImageModelInputModalitiesByName;
-    };
-    private readonly client;
-    constructor(config: GeminiImageConfig, model: TModel);
-    generateImages(options: ImageGenerationOptions<GeminiImageProviderOptions>): Promise<ImageGenerationResult>;
-    private isGeminiImageModel;
-    private generateWithGeminiApi;
-    /**
-     * Build the multimodal `contents` payload. Text-only prompts pass through
-     * as a plain string (the SDK accepts it directly); prompts with image
-     * parts become a single user `Content` whose `parts` mirror the prompt's
-     * interleaved order — position is meaningful to Gemini ("not like this
-     * *(image)*, more like this *(image)*").
-     *
-     * The generateContent API has no numberOfImages parameter, so when more
-     * than one image is requested a trailing instruction is appended.
-     */
-    private buildContents;
-    private imagePartToGeminiPart;
-    private transformGeminiResponse;
-    private buildImagenConfig;
-    private transformImagenResponse;
-}
-/**
- * Creates a Gemini image adapter with explicit API key.
- * Type resolution happens here at the call site.
- *
- * @param model - The model name (e.g., 'imagen-3.0-generate-002')
- * @param apiKey - Your Google API key
- * @param config - Optional additional configuration
- * @returns Configured Gemini image adapter instance with resolved types
- *
- * @example
- * ```typescript
- * const adapter = createGeminiImage('imagen-3.0-generate-002', "your-api-key");
- *
- * const result = await generateImage({
- *   adapter,
- *   prompt: 'A cute baby sea otter'
- * });
- * ```
- */
-export declare function createGeminiImage<TModel extends GeminiImageModel>(model: TModel, apiKey: string, config?: Omit<GeminiImageConfig, 'apiKey'>): GeminiImageAdapter<TModel>;
-/**
- * Creates a Gemini image adapter with automatic API key detection from environment variables.
- * Type resolution happens here at the call site.
- *
- * Looks for `GOOGLE_API_KEY` or `GEMINI_API_KEY` in:
- * - `process.env` (Node.js)
- * - `window.env` (Browser with injected env)
- *
- * @param model - The model name (e.g., 'imagen-4.0-generate-001')
- * @param config - Optional configuration (excluding apiKey which is auto-detected)
- * @returns Configured Gemini image adapter instance with resolved types
- * @throws Error if GOOGLE_API_KEY or GEMINI_API_KEY is not found in environment
- *
- * @example
- * ```typescript
- * // Automatically uses GOOGLE_API_KEY from environment
- * const adapter = geminiImage('imagen-4.0-generate-001');
- *
- * const result = await generateImage({
- *   adapter,
- *   prompt: 'A beautiful sunset over mountains'
- * });
- * ```
- */
-export declare function geminiImage<TModel extends GeminiImageModel>(model: TModel, config?: Omit<GeminiImageConfig, 'apiKey'>): GeminiImageAdapter<TModel>;

package/dist/esm/adapters/image.js

@@ -1,253 +0,0 @@
-import { resolveMediaPrompt } from "@tanstack/ai";
-import { BaseImageAdapter } from "@tanstack/ai/adapters";
-import { arrayBufferToBase64 } from "@tanstack/ai-utils";
-import { createGeminiClient, generateId, getGeminiApiKeyFromEnv } from "../utils/client.js";
-import { buildGeminiUsage } from "../usage.js";
-import { validatePrompt, validateImageSize, validateNumberOfImages, parseNativeImageSize, sizeToAspectRatio } from "../image/image-provider-options.js";
-class GeminiImageAdapter extends BaseImageAdapter {
-  kind = "image";
-  name = "gemini";
-  client;
-  constructor(config, model) {
-    super(model, config);
-    this.client = createGeminiClient(config);
-  }
-  async generateImages(options) {
-    const { model, logger } = options;
-    logger.request(
-      `activity=generateImage provider=gemini model=${this.model}`,
-      {
-        provider: "gemini",
-        model: this.model
-      }
-    );
-    try {
-      const resolved = resolveMediaPrompt(options.prompt);
-      if (resolved.images.length === 0) {
-        validatePrompt({ prompt: resolved.text, model });
-      }
-      if (resolved.videos.length > 0) {
-        throw new Error(
-          `${this.name}.generateImages does not support video prompt parts (model: ${model}).`
-        );
-      }
-      if (resolved.audios.length > 0) {
-        throw new Error(
-          `${this.name}.generateImages does not support audio prompt parts (model: ${model}).`
-        );
-      }
-      if (this.isGeminiImageModel(model)) {
-        return await this.generateWithGeminiApi(options, resolved);
-      }
-      if (resolved.images.length > 0) {
-        throw new Error(
-          `${this.name}: model "${model}" (Imagen) does not support image prompt parts. Use a Gemini-native image model (e.g. gemini-2.5-flash-image, "nano-banana") for image-conditioned generation.`
-        );
-      }
-      validateImageSize(model, options.size);
-      validateNumberOfImages(model, options.numberOfImages);
-      const config = this.buildImagenConfig(options);
-      const response = await this.client.models.generateImages({
-        model,
-        prompt: resolved.text,
-        config
-      });
-      return this.transformImagenResponse(model, response);
-    } catch (error) {
-      logger.errors("gemini.generateImage fatal", {
-        error,
-        source: "gemini.generateImage"
-      });
-      throw error;
-    }
-  }
-  isGeminiImageModel(model) {
-    return model.startsWith("gemini-");
-  }
-  async generateWithGeminiApi(options, resolved) {
-    const { model, size, numberOfImages, modelOptions } = options;
-    const parsedSize = size ? parseNativeImageSize(size) : void 0;
-    const nativeConfig = {};
-    if (modelOptions?.seed !== void 0) {
-      nativeConfig.seed = modelOptions.seed;
-    }
-    const config = {
-      ...nativeConfig,
-      // Include TEXT so the model can interleave descriptions between images.
-      // IMPORTANT: responseModalities is a protected default — set it AFTER
-      // nativeConfig so nothing can silently disable image output.
-      responseModalities: ["TEXT", "IMAGE"],
-      ...parsedSize && {
-        imageConfig: {
-          ...parsedSize.aspectRatio && {
-            aspectRatio: parsedSize.aspectRatio
-          },
-          ...parsedSize.resolution && {
-            imageSize: parsedSize.resolution
-          }
-        }
-      }
-    };
-    const contents = await this.buildContents(resolved, numberOfImages);
-    const response = await this.client.models.generateContent({
-      model,
-      contents,
-      config
-    });
-    return this.transformGeminiResponse(model, response);
-  }
-  /**
-   * Build the multimodal `contents` payload. Text-only prompts pass through
-   * as a plain string (the SDK accepts it directly); prompts with image
-   * parts become a single user `Content` whose `parts` mirror the prompt's
-   * interleaved order — position is meaningful to Gemini ("not like this
-   * *(image)*, more like this *(image)*").
-   *
-   * The generateContent API has no numberOfImages parameter, so when more
-   * than one image is requested a trailing instruction is appended.
-   */
-  async buildContents(resolved, numberOfImages) {
-    const countInstruction = numberOfImages && numberOfImages > 1 ? `Generate ${numberOfImages} distinct images.` : void 0;
-    if (resolved.images.length === 0) {
-      return countInstruction ? `${resolved.text} ${countInstruction}` : resolved.text;
-    }
-    const parts = await Promise.all(
-      resolved.parts.map((part) => {
-        if (part.type === "text") {
-          return Promise.resolve({ text: part.content });
-        }
-        if (part.type === "image") {
-          return this.imagePartToGeminiPart(part);
-        }
-        throw new Error(
-          `gemini: unsupported prompt part type "${part.type}" in image generation.`
-        );
-      })
-    );
-    if (countInstruction) {
-      parts.push({ text: countInstruction });
-    }
-    return [{ role: "user", parts }];
-  }
-  async imagePartToGeminiPart(part) {
-    if (part.source.type === "data") {
-      return {
-        inlineData: {
-          mimeType: part.source.mimeType || "image/png",
-          data: part.source.value
-        }
-      };
-    }
-    if (part.source.value.startsWith("gs://") || /^https?:\/\/generativelanguage\.googleapis\.com\//.test(
-      part.source.value
-    )) {
-      return {
-        fileData: {
-          fileUri: part.source.value,
-          ...part.source.mimeType && { mimeType: part.source.mimeType }
-        }
-      };
-    }
-    const response = await fetch(part.source.value);
-    if (!response.ok) {
-      throw new Error(
-        `Failed to fetch image input (${response.status} ${response.statusText}): ${part.source.value}`
-      );
-    }
-    const blob = await response.blob();
-    const buffer = await blob.arrayBuffer();
-    const base64 = arrayBufferToBase64(buffer);
-    return {
-      inlineData: {
-        mimeType: part.source.mimeType || blob.type || "image/png",
-        data: base64
-      }
-    };
-  }
-  transformGeminiResponse(model, response) {
-    const images = [];
-    const textParts = [];
-    const parts = response.candidates?.[0]?.content?.parts ?? [];
-    for (const part of parts) {
-      if (part.inlineData?.data && typeof part.inlineData.data === "string" && part.inlineData.data.length > 0) {
-        images.push({ b64Json: part.inlineData.data });
-      } else if (typeof part.text === "string" && part.text.length > 0) {
-        textParts.push(part.text);
-      }
-    }
-    if (images.length === 0) {
-      const reason = textParts.length > 0 ? `: ${textParts.join(" ").trim()}` : " (no inline image or text parts were returned).";
-      throw new Error(`Gemini ${model} returned no images${reason}`);
-    }
-    return {
-      id: generateId(this.name),
-      model,
-      images,
-      // Surface token usage (with per-modality breakdown) when the model
-      // reports it (e.g. Nano Banana via generateContent). Conditionally spread
-      // to satisfy exactOptionalPropertyTypes — only include usage when
-      // present. See #330.
-      ...response.usageMetadata ? { usage: buildGeminiUsage(response.usageMetadata) } : {}
-    };
-  }
-  buildImagenConfig(options) {
-    const { size, numberOfImages, modelOptions } = options;
-    const sizeAspectRatio = size ? sizeToAspectRatio(size) : void 0;
-    return {
-      numberOfImages: numberOfImages ?? 1,
-      // Map size to aspect ratio if provided (modelOptions.aspectRatio will override)
-      ...sizeAspectRatio !== void 0 && { aspectRatio: sizeAspectRatio },
-      ...modelOptions
-    };
-  }
-  transformImagenResponse(model, response) {
-    const entries = response.generatedImages ?? [];
-    const images = [];
-    const filterReasons = [];
-    for (const item of entries) {
-      const b64Json = item.image?.imageBytes;
-      if (b64Json) {
-        images.push({
-          b64Json,
-          ...item.enhancedPrompt !== void 0 && {
-            revisedPrompt: item.enhancedPrompt
-          }
-        });
-        continue;
-      }
-      const reason = item.raiFilteredReason;
-      if (reason) {
-        filterReasons.push(reason);
-      }
-    }
-    if (entries.length > 0 && images.length === 0) {
-      const joined = filterReasons.length > 0 ? filterReasons.join("; ") : "";
-      throw new Error(
-        `Imagen ${model} returned no images: all ${entries.length} generated image(s) were filtered by Responsible-AI${joined ? ` (${joined})` : ""}.`
-      );
-    }
-    if (filterReasons.length > 0 && typeof console !== "undefined") {
-      console.warn(
-        `[gemini-image] ${filterReasons.length} of ${entries.length} images from ${model} were filtered by Responsible-AI: ${filterReasons.join("; ")}`
-      );
-    }
-    return {
-      id: generateId(this.name),
-      model,
-      images
-    };
-  }
-}
-function createGeminiImage(model, apiKey, config) {
-  return new GeminiImageAdapter({ apiKey, ...config }, model);
-}
-function geminiImage(model, config) {
-  const apiKey = getGeminiApiKeyFromEnv();
-  return createGeminiImage(model, apiKey, config);
-}
-export {
-  GeminiImageAdapter,
-  createGeminiImage,
-  geminiImage
-};
-//# sourceMappingURL=image.js.map