@tanstack/ai-gemini 0.9.1 → 0.10.0

package/dist/esm/image/image-provider-options.d.ts

@@ -147,8 +147,10 @@ export declare function sizeToAspectRatio(size: string | undefined): GeminiAspec
  */
 export declare function validateImageSize(model: string, size: string | undefined): void;
 /**
- * Validates the number of images requested
- * Imagen models support 1-8 images per request (varies by model)
+ * Validates the number of images requested against the model's known cap.
+ * Uses a per-model table where available and falls back to the shared
+ * default otherwise — no more "some support up to 8" comments that don't
+ * match the error message.
  */
 export declare function validateNumberOfImages(model: string, numberOfImages: number | undefined): void;
 /**

package/dist/esm/image/image-provider-options.js

@@ -28,9 +28,16 @@ function validateImageSize(model, size) {
     );
   }
 }
+const IMAGEN_MAX_IMAGES_BY_MODEL = {
+  "imagen-3.0-generate-002": 4,
+  "imagen-4.0-generate-001": 4,
+  "imagen-4.0-ultra-generate-001": 4,
+  "imagen-4.0-fast-generate-001": 4
+};
+const DEFAULT_IMAGEN_MAX_IMAGES = 4;
 function validateNumberOfImages(model, numberOfImages) {
   if (numberOfImages === void 0) return;
-  const maxImages = 4;
+  const maxImages = IMAGEN_MAX_IMAGES_BY_MODEL[model] ?? DEFAULT_IMAGEN_MAX_IMAGES;
   if (numberOfImages < 1 || numberOfImages > maxImages) {
     throw new Error(
       `Invalid numberOfImages "${numberOfImages}" for model "${model}". Must be between 1 and ${maxImages}.`

package/dist/esm/image/image-provider-options.js.map

@@ -1 +1 @@
-{"version":3,"file":"image-provider-options.js","sources":["../../../src/image/image-provider-options.ts"],"sourcesContent":["import type { GeminiImageModels } from '../model-meta'\nimport type {\n  ImagePromptLanguage,\n  PersonGeneration,\n  SafetyFilterLevel,\n} from '@google/genai'\n\n// Re-export SDK types so users can use them directly\nexport type { ImagePromptLanguage, PersonGeneration, SafetyFilterLevel }\n\n/**\n * Gemini Imagen aspect ratio options\n * Controls the aspect ratio of generated images\n */\nexport type GeminiAspectRatio =\n  | '1:1'\n  | '3:4'\n  | '4:3'\n  | '9:16'\n  | '16:9'\n  | '9:21'\n  | '21:9'\n\n/**\n * Provider options for Gemini image generation\n * These options match the @google/genai GenerateImagesConfig interface\n * and can be spread directly into the API request.\n */\nexport interface GeminiImageProviderOptions {\n  /**\n   * The aspect ratio of generated images\n   * @default '1:1'\n   */\n  aspectRatio?: GeminiAspectRatio\n\n  /**\n   * Controls whether people can appear in generated images\n   * Use PersonGeneration enum values: DONT_ALLOW, ALLOW_ADULT, ALLOW_ALL\n   * @default 'ALLOW_ADULT'\n   */\n  personGeneration?: PersonGeneration\n\n  /**\n   * Safety filter level for content filtering\n   * Use SafetyFilterLevel enum values\n   */\n  safetyFilterLevel?: SafetyFilterLevel\n\n  /**\n   * Optional seed for reproducible image generation\n   * When the same seed is used with the same prompt and settings,\n   * you should get similar (though not identical) results\n   */\n  seed?: number\n\n  /**\n   * Whether to add a SynthID watermark to generated images\n   * SynthID helps identify AI-generated content\n   * @default true\n   */\n  addWatermark?: boolean\n\n  /**\n   * Language of the prompt\n   * Use ImagePromptLanguage enum values\n   */\n  language?: ImagePromptLanguage\n\n  /**\n   * Negative prompt - what to avoid in the generated image\n   * Not all models support negative prompts\n   */\n  negativePrompt?: string\n\n  /**\n   * Output MIME type for the generated image\n   * @default 'image/png'\n   */\n  outputMimeType?: 'image/png' | 'image/jpeg' | 'image/webp'\n\n  /**\n   * Compression quality for JPEG outputs (0-100)\n   * Higher values mean better quality but larger file sizes\n   * @default 75\n   */\n  outputCompressionQuality?: number\n\n  /**\n   * Controls how much the model adheres to the text prompt\n   * Large values increase output and prompt alignment,\n   * but may compromise image quality\n   */\n  guidanceScale?: number\n\n  /**\n   * Whether to use the prompt rewriting logic\n   */\n  enhancePrompt?: boolean\n\n  /**\n   * Whether to report the safety scores of each generated image\n   * and the positive prompt in the response\n   */\n  includeSafetyAttributes?: boolean\n\n  /**\n   * Whether to include the Responsible AI filter reason\n   * if the image is filtered out of the response\n   */\n  includeRaiReason?: boolean\n\n  /**\n   * Cloud Storage URI used to store the generated images\n   */\n  outputGcsUri?: string\n\n  /**\n   * User specified labels to track billing usage\n   */\n  labels?: Record<string, string>\n}\n\n/**\n * Model-specific provider options mapping\n * Currently all Imagen models use the same options structure\n */\nexport type GeminiImageModelProviderOptionsByName = {\n  [K in GeminiImageModels]: GeminiImageProviderOptions\n}\n\n/**\n * Supported size strings for Gemini Imagen models\n * These map to aspect ratios internally\n */\nexport type GeminiImageSize =\n  | '1024x1024'\n  | '512x512'\n  | '1024x768'\n  | '1536x1024'\n  | '1792x1024'\n  | '1920x1080'\n  | '768x1024'\n  | '1024x1536'\n  | '1024x1792'\n  | '1080x1920'\n\n/**\n * Aspect ratios supported by Gemini native image models (via generateContent API).\n * Matches the SDK's ImageConfig.aspectRatio values.\n */\nexport type GeminiNativeImageAspectRatio =\n  | '1:1'\n  | '2:3'\n  | '3:2'\n  | '3:4'\n  | '4:3'\n  | '9:16'\n  | '16:9'\n  | '21:9'\n\n/**\n * Resolution tiers for Gemini native image models.\n * Matches the SDK's ImageConfig.imageSize values.\n */\nexport type GeminiNativeImageResolution = '1K' | '2K' | '4K'\n\n/**\n * Template literal size type for Gemini native image models: \"16:9_4K\", \"1:1_2K\", etc.\n */\nexport type GeminiNativeImageSize =\n  `${GeminiNativeImageAspectRatio}_${GeminiNativeImageResolution}`\n\n/**\n * Gemini native image models that use the generateContent API path.\n * These models support template literal sizes (aspectRatio_resolution).\n */\nexport type GeminiNativeImageModels =\n  | 'gemini-3.1-flash-image-preview'\n  | 'gemini-3-pro-image-preview'\n  | 'gemini-2.5-flash-image'\n  | 'gemini-2.0-flash-preview-image-generation'\n\n/**\n * Model-specific size options mapping.\n * Gemini native image models use template literal sizes, Imagen models use pixel sizes.\n */\nexport type GeminiImageModelSizeByName = {\n  [K in GeminiNativeImageModels]: GeminiNativeImageSize\n} & {\n  [K in Exclude<GeminiImageModels, GeminiNativeImageModels>]: GeminiImageSize\n}\n\n/**\n * Valid sizes for Gemini Imagen models\n * Gemini uses aspect ratios, but we map common WIDTHxHEIGHT formats to aspect ratios\n * These are approximate mappings based on common image dimensions\n */\nexport const GEMINI_SIZE_TO_ASPECT_RATIO: Record<string, GeminiAspectRatio> = {\n  // Square\n  '1024x1024': '1:1',\n  '512x512': '1:1',\n  // Landscape\n  '1024x768': '4:3',\n  '1536x1024': '4:3',\n  '1792x1024': '16:9',\n  '1920x1080': '16:9',\n  // Portrait\n  '768x1024': '3:4',\n  '1024x1536': '3:4', // Inverted\n  '1024x1792': '9:16',\n  '1080x1920': '9:16',\n}\n\n/**\n * Maps a WIDTHxHEIGHT size string to a Gemini aspect ratio\n * Returns undefined if the size cannot be mapped\n */\nexport function sizeToAspectRatio(\n  size: string | undefined,\n): GeminiAspectRatio | undefined {\n  if (!size) return undefined\n  return GEMINI_SIZE_TO_ASPECT_RATIO[size]\n}\n\n/**\n * Validates that the provided size can be mapped to an aspect ratio\n * Throws an error if the size is invalid\n */\nexport function validateImageSize(\n  model: string,\n  size: string | undefined,\n): void {\n  if (!size) return\n\n  const aspectRatio = sizeToAspectRatio(size)\n  if (!aspectRatio) {\n    const validSizes = Object.keys(GEMINI_SIZE_TO_ASPECT_RATIO)\n    throw new Error(\n      `Invalid size \"${size}\" for model \"${model}\". ` +\n        `Gemini Imagen uses aspect ratios. Valid sizes that map to aspect ratios: ${validSizes.join(', ')}. ` +\n        `Alternatively, use providerOptions.aspectRatio directly with values: 1:1, 3:4, 4:3, 9:16, 16:9, 9:21, 21:9`,\n    )\n  }\n}\n\n/**\n * Validates the number of images requested\n * Imagen models support 1-8 images per request (varies by model)\n */\nexport function validateNumberOfImages(\n  model: string,\n  numberOfImages: number | undefined,\n): void {\n  if (numberOfImages === undefined) return\n\n  // Most Imagen models support 1-4 images, some support up to 8\n  const maxImages = 4\n  if (numberOfImages < 1 || numberOfImages > maxImages) {\n    throw new Error(\n      `Invalid numberOfImages \"${numberOfImages}\" for model \"${model}\". ` +\n        `Must be between 1 and ${maxImages}.`,\n    )\n  }\n}\n\n/**\n * Validates the prompt is not empty\n */\nexport function validatePrompt(options: {\n  prompt: string\n  model: string\n}): void {\n  const { prompt, model } = options\n  if (!prompt || prompt.trim().length === 0) {\n    throw new Error(`Prompt cannot be empty for model \"${model}\".`)\n  }\n}\n\n/**\n * Parses a Gemini native image size string into its components.\n * Format: \"aspectRatio_resolution\" e.g. \"16:9_4K\" → { aspectRatio: \"16:9\", resolution: \"4K\" }\n */\nexport function parseNativeImageSize(\n  size: string,\n): { aspectRatio: string; resolution: string } | undefined {\n  const match = size.match(/^(\\d+:\\d+)_(.+)$/)\n  if (!match) return undefined\n  return { aspectRatio: match[1]!, resolution: match[2]! }\n}\n"],"names":[],"mappings":"AAqMO,MAAM,8BAAiE;AAAA;AAAA,EAE5E,aAAa;AAAA,EACb,WAAW;AAAA;AAAA,EAEX,YAAY;AAAA,EACZ,aAAa;AAAA,EACb,aAAa;AAAA,EACb,aAAa;AAAA;AAAA,EAEb,YAAY;AAAA,EACZ,aAAa;AAAA;AAAA,EACb,aAAa;AAAA,EACb,aAAa;AACf;AAMO,SAAS,kBACd,MAC+B;AAC/B,MAAI,CAAC,KAAM,QAAO;AAClB,SAAO,4BAA4B,IAAI;AACzC;AAMO,SAAS,kBACd,OACA,MACM;AACN,MAAI,CAAC,KAAM;AAEX,QAAM,cAAc,kBAAkB,IAAI;AAC1C,MAAI,CAAC,aAAa;AAChB,UAAM,aAAa,OAAO,KAAK,2BAA2B;AAC1D,UAAM,IAAI;AAAA,MACR,iBAAiB,IAAI,gBAAgB,KAAK,+EACoC,WAAW,KAAK,IAAI,CAAC;AAAA,IAAA;AAAA,EAGvG;AACF;AAMO,SAAS,uBACd,OACA,gBACM;AACN,MAAI,mBAAmB,OAAW;AAGlC,QAAM,YAAY;AAClB,MAAI,iBAAiB,KAAK,iBAAiB,WAAW;AACpD,UAAM,IAAI;AAAA,MACR,2BAA2B,cAAc,gBAAgB,KAAK,4BACnC,SAAS;AAAA,IAAA;AAAA,EAExC;AACF;AAKO,SAAS,eAAe,SAGtB;AACP,QAAM,EAAE,QAAQ,MAAA,IAAU;AAC1B,MAAI,CAAC,UAAU,OAAO,KAAA,EAAO,WAAW,GAAG;AACzC,UAAM,IAAI,MAAM,qCAAqC,KAAK,IAAI;AAAA,EAChE;AACF;AAMO,SAAS,qBACd,MACyD;AACzD,QAAM,QAAQ,KAAK,MAAM,kBAAkB;AAC3C,MAAI,CAAC,MAAO,QAAO;AACnB,SAAO,EAAE,aAAa,MAAM,CAAC,GAAI,YAAY,MAAM,CAAC,EAAA;AACtD;"}
+{"version":3,"file":"image-provider-options.js","sources":["../../../src/image/image-provider-options.ts"],"sourcesContent":["import type { GeminiImageModels } from '../model-meta'\nimport type {\n  ImagePromptLanguage,\n  PersonGeneration,\n  SafetyFilterLevel,\n} from '@google/genai'\n\n// Re-export SDK types so users can use them directly\nexport type { ImagePromptLanguage, PersonGeneration, SafetyFilterLevel }\n\n/**\n * Gemini Imagen aspect ratio options\n * Controls the aspect ratio of generated images\n */\nexport type GeminiAspectRatio =\n  | '1:1'\n  | '3:4'\n  | '4:3'\n  | '9:16'\n  | '16:9'\n  | '9:21'\n  | '21:9'\n\n/**\n * Provider options for Gemini image generation\n * These options match the @google/genai GenerateImagesConfig interface\n * and can be spread directly into the API request.\n */\nexport interface GeminiImageProviderOptions {\n  /**\n   * The aspect ratio of generated images\n   * @default '1:1'\n   */\n  aspectRatio?: GeminiAspectRatio\n\n  /**\n   * Controls whether people can appear in generated images\n   * Use PersonGeneration enum values: DONT_ALLOW, ALLOW_ADULT, ALLOW_ALL\n   * @default 'ALLOW_ADULT'\n   */\n  personGeneration?: PersonGeneration\n\n  /**\n   * Safety filter level for content filtering\n   * Use SafetyFilterLevel enum values\n   */\n  safetyFilterLevel?: SafetyFilterLevel\n\n  /**\n   * Optional seed for reproducible image generation\n   * When the same seed is used with the same prompt and settings,\n   * you should get similar (though not identical) results\n   */\n  seed?: number\n\n  /**\n   * Whether to add a SynthID watermark to generated images\n   * SynthID helps identify AI-generated content\n   * @default true\n   */\n  addWatermark?: boolean\n\n  /**\n   * Language of the prompt\n   * Use ImagePromptLanguage enum values\n   */\n  language?: ImagePromptLanguage\n\n  /**\n   * Negative prompt - what to avoid in the generated image\n   * Not all models support negative prompts\n   */\n  negativePrompt?: string\n\n  /**\n   * Output MIME type for the generated image\n   * @default 'image/png'\n   */\n  outputMimeType?: 'image/png' | 'image/jpeg' | 'image/webp'\n\n  /**\n   * Compression quality for JPEG outputs (0-100)\n   * Higher values mean better quality but larger file sizes\n   * @default 75\n   */\n  outputCompressionQuality?: number\n\n  /**\n   * Controls how much the model adheres to the text prompt\n   * Large values increase output and prompt alignment,\n   * but may compromise image quality\n   */\n  guidanceScale?: number\n\n  /**\n   * Whether to use the prompt rewriting logic\n   */\n  enhancePrompt?: boolean\n\n  /**\n   * Whether to report the safety scores of each generated image\n   * and the positive prompt in the response\n   */\n  includeSafetyAttributes?: boolean\n\n  /**\n   * Whether to include the Responsible AI filter reason\n   * if the image is filtered out of the response\n   */\n  includeRaiReason?: boolean\n\n  /**\n   * Cloud Storage URI used to store the generated images\n   */\n  outputGcsUri?: string\n\n  /**\n   * User specified labels to track billing usage\n   */\n  labels?: Record<string, string>\n}\n\n/**\n * Model-specific provider options mapping\n * Currently all Imagen models use the same options structure\n */\nexport type GeminiImageModelProviderOptionsByName = {\n  [K in GeminiImageModels]: GeminiImageProviderOptions\n}\n\n/**\n * Supported size strings for Gemini Imagen models\n * These map to aspect ratios internally\n */\nexport type GeminiImageSize =\n  | '1024x1024'\n  | '512x512'\n  | '1024x768'\n  | '1536x1024'\n  | '1792x1024'\n  | '1920x1080'\n  | '768x1024'\n  | '1024x1536'\n  | '1024x1792'\n  | '1080x1920'\n\n/**\n * Aspect ratios supported by Gemini native image models (via generateContent API).\n * Matches the SDK's ImageConfig.aspectRatio values.\n */\nexport type GeminiNativeImageAspectRatio =\n  | '1:1'\n  | '2:3'\n  | '3:2'\n  | '3:4'\n  | '4:3'\n  | '9:16'\n  | '16:9'\n  | '21:9'\n\n/**\n * Resolution tiers for Gemini native image models.\n * Matches the SDK's ImageConfig.imageSize values.\n */\nexport type GeminiNativeImageResolution = '1K' | '2K' | '4K'\n\n/**\n * Template literal size type for Gemini native image models: \"16:9_4K\", \"1:1_2K\", etc.\n */\nexport type GeminiNativeImageSize =\n  `${GeminiNativeImageAspectRatio}_${GeminiNativeImageResolution}`\n\n/**\n * Gemini native image models that use the generateContent API path.\n * These models support template literal sizes (aspectRatio_resolution).\n */\nexport type GeminiNativeImageModels =\n  | 'gemini-3.1-flash-image-preview'\n  | 'gemini-3-pro-image-preview'\n  | 'gemini-2.5-flash-image'\n  | 'gemini-2.0-flash-preview-image-generation'\n\n/**\n * Model-specific size options mapping.\n * Gemini native image models use template literal sizes, Imagen models use pixel sizes.\n */\nexport type GeminiImageModelSizeByName = {\n  [K in GeminiNativeImageModels]: GeminiNativeImageSize\n} & {\n  [K in Exclude<GeminiImageModels, GeminiNativeImageModels>]: GeminiImageSize\n}\n\n/**\n * Valid sizes for Gemini Imagen models\n * Gemini uses aspect ratios, but we map common WIDTHxHEIGHT formats to aspect ratios\n * These are approximate mappings based on common image dimensions\n */\nexport const GEMINI_SIZE_TO_ASPECT_RATIO: Record<string, GeminiAspectRatio> = {\n  // Square\n  '1024x1024': '1:1',\n  '512x512': '1:1',\n  // Landscape\n  '1024x768': '4:3',\n  '1536x1024': '4:3',\n  '1792x1024': '16:9',\n  '1920x1080': '16:9',\n  // Portrait\n  '768x1024': '3:4',\n  '1024x1536': '3:4', // Inverted\n  '1024x1792': '9:16',\n  '1080x1920': '9:16',\n}\n\n/**\n * Maps a WIDTHxHEIGHT size string to a Gemini aspect ratio\n * Returns undefined if the size cannot be mapped\n */\nexport function sizeToAspectRatio(\n  size: string | undefined,\n): GeminiAspectRatio | undefined {\n  if (!size) return undefined\n  return GEMINI_SIZE_TO_ASPECT_RATIO[size]\n}\n\n/**\n * Validates that the provided size can be mapped to an aspect ratio\n * Throws an error if the size is invalid\n */\nexport function validateImageSize(\n  model: string,\n  size: string | undefined,\n): void {\n  if (!size) return\n\n  const aspectRatio = sizeToAspectRatio(size)\n  if (!aspectRatio) {\n    const validSizes = Object.keys(GEMINI_SIZE_TO_ASPECT_RATIO)\n    throw new Error(\n      `Invalid size \"${size}\" for model \"${model}\". ` +\n        `Gemini Imagen uses aspect ratios. Valid sizes that map to aspect ratios: ${validSizes.join(', ')}. ` +\n        `Alternatively, use providerOptions.aspectRatio directly with values: 1:1, 3:4, 4:3, 9:16, 16:9, 9:21, 21:9`,\n    )\n  }\n}\n\n/**\n * Per-model caps on images per request.\n * Imagen 3 and the Imagen 4 family all support up to 4 images per request\n * via the Gemini API (the rumored 8-image tier is Vertex-only and isn't\n * reachable through @google/genai today). Unknown models fall through to\n * the shared cap defined below.\n *\n * @see https://ai.google.dev/gemini-api/docs/imagen\n */\nconst IMAGEN_MAX_IMAGES_BY_MODEL: Record<string, number> = {\n  'imagen-3.0-generate-002': 4,\n  'imagen-4.0-generate-001': 4,\n  'imagen-4.0-ultra-generate-001': 4,\n  'imagen-4.0-fast-generate-001': 4,\n}\n\nconst DEFAULT_IMAGEN_MAX_IMAGES = 4\n\n/**\n * Validates the number of images requested against the model's known cap.\n * Uses a per-model table where available and falls back to the shared\n * default otherwise — no more \"some support up to 8\" comments that don't\n * match the error message.\n */\nexport function validateNumberOfImages(\n  model: string,\n  numberOfImages: number | undefined,\n): void {\n  if (numberOfImages === undefined) return\n\n  const maxImages =\n    IMAGEN_MAX_IMAGES_BY_MODEL[model] ?? DEFAULT_IMAGEN_MAX_IMAGES\n  if (numberOfImages < 1 || numberOfImages > maxImages) {\n    throw new Error(\n      `Invalid numberOfImages \"${numberOfImages}\" for model \"${model}\". ` +\n        `Must be between 1 and ${maxImages}.`,\n    )\n  }\n}\n\n/**\n * Validates the prompt is not empty\n */\nexport function validatePrompt(options: {\n  prompt: string\n  model: string\n}): void {\n  const { prompt, model } = options\n  if (!prompt || prompt.trim().length === 0) {\n    throw new Error(`Prompt cannot be empty for model \"${model}\".`)\n  }\n}\n\n/**\n * Parses a Gemini native image size string into its components.\n * Format: \"aspectRatio_resolution\" e.g. \"16:9_4K\" → { aspectRatio: \"16:9\", resolution: \"4K\" }\n */\nexport function parseNativeImageSize(\n  size: string,\n): { aspectRatio: string; resolution: string } | undefined {\n  const match = size.match(/^(\\d+:\\d+)_(.+)$/)\n  if (!match) return undefined\n  return { aspectRatio: match[1]!, resolution: match[2]! }\n}\n"],"names":[],"mappings":"AAqMO,MAAM,8BAAiE;AAAA;AAAA,EAE5E,aAAa;AAAA,EACb,WAAW;AAAA;AAAA,EAEX,YAAY;AAAA,EACZ,aAAa;AAAA,EACb,aAAa;AAAA,EACb,aAAa;AAAA;AAAA,EAEb,YAAY;AAAA,EACZ,aAAa;AAAA;AAAA,EACb,aAAa;AAAA,EACb,aAAa;AACf;AAMO,SAAS,kBACd,MAC+B;AAC/B,MAAI,CAAC,KAAM,QAAO;AAClB,SAAO,4BAA4B,IAAI;AACzC;AAMO,SAAS,kBACd,OACA,MACM;AACN,MAAI,CAAC,KAAM;AAEX,QAAM,cAAc,kBAAkB,IAAI;AAC1C,MAAI,CAAC,aAAa;AAChB,UAAM,aAAa,OAAO,KAAK,2BAA2B;AAC1D,UAAM,IAAI;AAAA,MACR,iBAAiB,IAAI,gBAAgB,KAAK,+EACoC,WAAW,KAAK,IAAI,CAAC;AAAA,IAAA;AAAA,EAGvG;AACF;AAWA,MAAM,6BAAqD;AAAA,EACzD,2BAA2B;AAAA,EAC3B,2BAA2B;AAAA,EAC3B,iCAAiC;AAAA,EACjC,gCAAgC;AAClC;AAEA,MAAM,4BAA4B;AAQ3B,SAAS,uBACd,OACA,gBACM;AACN,MAAI,mBAAmB,OAAW;AAElC,QAAM,YACJ,2BAA2B,KAAK,KAAK;AACvC,MAAI,iBAAiB,KAAK,iBAAiB,WAAW;AACpD,UAAM,IAAI;AAAA,MACR,2BAA2B,cAAc,gBAAgB,KAAK,4BACnC,SAAS;AAAA,IAAA;AAAA,EAExC;AACF;AAKO,SAAS,eAAe,SAGtB;AACP,QAAM,EAAE,QAAQ,MAAA,IAAU;AAC1B,MAAI,CAAC,UAAU,OAAO,KAAA,EAAO,WAAW,GAAG;AACzC,UAAM,IAAI,MAAM,qCAAqC,KAAK,IAAI;AAAA,EAChE;AACF;AAMO,SAAS,qBACd,MACyD;AACzD,QAAM,QAAQ,KAAK,MAAM,kBAAkB;AAC3C,MAAI,CAAC,MAAO,QAAO;AACnB,SAAO,EAAE,aAAa,MAAM,CAAC,GAAI,YAAY,MAAM,CAAC,EAAA;AACtD;"}

package/dist/esm/index.d.ts

@@ -6,11 +6,16 @@ export type { GeminiImageProviderOptions, GeminiImageModelProviderOptionsByName,
  * @experimental Gemini TTS is an experimental feature and may change.
  */
 export { GeminiTTSAdapter, createGeminiSpeech, geminiSpeech, type GeminiTTSConfig, type GeminiTTSProviderOptions, } from './adapters/tts.js';
+/**
+ * @experimental Gemini Lyria music generation is an experimental feature and may change.
+ */
+export { GeminiAudioAdapter, createGeminiAudio, geminiAudio, type GeminiAudioConfig, type GeminiAudioModel, type GeminiAudioProviderOptions, } from './adapters/audio.js';
 export { GEMINI_MODELS } from './model-meta.js';
 export { GEMINI_MODELS as GeminiTextModels } from './model-meta.js';
 export { GEMINI_IMAGE_MODELS as GeminiImageModels } from './model-meta.js';
 export { GEMINI_TTS_MODELS as GeminiTTSModels } from './model-meta.js';
 export { GEMINI_TTS_VOICES as GeminiTTSVoices } from './model-meta.js';
+export { GEMINI_AUDIO_MODELS as GeminiAudioModels } from './model-meta.js';
 export type { GeminiModels as GeminiTextModel } from './model-meta.js';
 export type { GeminiImageModels as GeminiImageModel } from './model-meta.js';
 export type { GeminiTTSVoice } from './model-meta.js';

package/dist/esm/index.js

@@ -2,9 +2,12 @@ import { GeminiTextAdapter, createGeminiChat, geminiText } from "./adapters/text
 import { GeminiSummarizeAdapter, GeminiSummarizeModels, createGeminiSummarize, geminiSummarize } from "./adapters/summarize.js";
 import { GeminiImageAdapter, createGeminiImage, geminiImage } from "./adapters/image.js";
 import { GeminiTTSAdapter, createGeminiSpeech, geminiSpeech } from "./adapters/tts.js";
-import { GEMINI_MODELS, GEMINI_IMAGE_MODELS, GEMINI_TTS_MODELS, GEMINI_TTS_VOICES, GEMINI_MODELS as GEMINI_MODELS2 } from "./model-meta.js";
+import { GeminiAudioAdapter, createGeminiAudio, geminiAudio } from "./adapters/audio.js";
+import { GEMINI_MODELS, GEMINI_AUDIO_MODELS, GEMINI_IMAGE_MODELS, GEMINI_TTS_MODELS, GEMINI_TTS_VOICES, GEMINI_MODELS as GEMINI_MODELS2 } from "./model-meta.js";
 export {
   GEMINI_MODELS,
+  GeminiAudioAdapter,
+  GEMINI_AUDIO_MODELS as GeminiAudioModels,
   GeminiImageAdapter,
   GEMINI_IMAGE_MODELS as GeminiImageModels,
   GeminiSummarizeAdapter,
@@ -14,10 +17,12 @@ export {
   GEMINI_TTS_VOICES as GeminiTTSVoices,
   GeminiTextAdapter,
   GEMINI_MODELS2 as GeminiTextModels,
+  createGeminiAudio,
   createGeminiChat,
   createGeminiImage,
   createGeminiSpeech,
   createGeminiSummarize,
+  geminiAudio,
   geminiImage,
   geminiSpeech,
   geminiSummarize,

package/dist/esm/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sources":[],"sourcesContent":[],"names":[],"mappings":";;;;;"}
+{"version":3,"file":"index.js","sources":[],"sourcesContent":[],"names":[],"mappings":";;;;;;"}

package/dist/esm/model-meta.d.ts

@@ -341,7 +341,12 @@ export declare const GEMINI_IMAGE_MODELS: readonly ["gemini-3.1-flash-image-prev
  * Text-to-speech models
  * @experimental Gemini TTS is an experimental feature and may change.
  */
-export declare const GEMINI_TTS_MODELS: readonly ["gemini-2.5-flash-preview-tts", "gemini-2.5-pro-preview-tts"];
+export declare const GEMINI_TTS_MODELS: readonly ["gemini-3.1-flash-tts-preview", "gemini-2.5-flash-preview-tts", "gemini-2.5-pro-preview-tts"];
+/**
+ * Audio generation models (Lyria music generation).
+ * @experimental Lyria music generation is an experimental feature and may change.
+ */
+export declare const GEMINI_AUDIO_MODELS: readonly ["lyria-3-pro-preview", "lyria-3-clip-preview"];
 /**
  * Available voice names for Gemini TTS
  * @see https://ai.google.dev/gemini-api/docs/speech-generation

package/dist/esm/model-meta.js

@@ -34,6 +34,15 @@ const GEMINI_2_5_FLASH_IMAGE = {
 const GEMINI_2_5_FLASH_TTS = {
   name: "gemini-2.5-flash-preview-tts"
 };
+const GEMINI_3_1_FLASH_TTS = {
+  name: "gemini-3.1-flash-tts-preview"
+};
+const LYRIA_3_PRO = {
+  name: "lyria-3-pro-preview"
+};
+const LYRIA_3_CLIP = {
+  name: "lyria-3-clip-preview"
+};
 const GEMINI_2_5_FLASH_LITE = {
   name: "gemini-2.5-flash-lite"
 };
@@ -85,9 +94,14 @@ const GEMINI_IMAGE_MODELS = [
   IMAGEN_4_GENERATE_ULTRA.name
 ];
 const GEMINI_TTS_MODELS = [
+  GEMINI_3_1_FLASH_TTS.name,
   GEMINI_2_5_FLASH_TTS.name,
   GEMINI_2_5_PRO_TTS.name
 ];
+const GEMINI_AUDIO_MODELS = [
+  LYRIA_3_PRO.name,
+  LYRIA_3_CLIP.name
+];
 const GEMINI_TTS_VOICES = [
   "Zephyr",
   "Puck",
@@ -121,6 +135,7 @@ const GEMINI_TTS_VOICES = [
   "Sulafat"
 ];
 export {
+  GEMINI_AUDIO_MODELS,
   GEMINI_IMAGE_MODELS,
   GEMINI_MODELS,
   GEMINI_TTS_MODELS,

package/dist/esm/model-meta.js.map

@@ -1 +1 @@
-{"version":3,"file":"model-meta.js","sources":["../../src/model-meta.ts"],"sourcesContent":["import type {\n  GeminiCachedContentOptions,\n  GeminiCommonConfigOptions,\n  GeminiSafetyOptions,\n  GeminiStructuredOutputOptions,\n  GeminiThinkingAdvancedOptions,\n  GeminiThinkingOptions,\n  GeminiToolConfigOptions,\n} from './text/text-provider-options'\n\ninterface ModelMeta<TProviderOptions = unknown> {\n  name: string\n  supports: {\n    input: Array<'text' | 'image' | 'audio' | 'video' | 'document'>\n    output: Array<'text' | 'image' | 'audio' | 'video'>\n    capabilities?: Array<\n      | 'audio_generation'\n      | 'batch_api'\n      | 'caching'\n      | 'function_calling'\n      | 'live_api'\n      | 'structured_output'\n      | 'thinking'\n    >\n    tools?: Array<\n      | 'code_execution'\n      | 'file_search'\n      | 'google_search'\n      | 'google_search_retrieval'\n      | 'google_maps'\n      | 'url_context'\n      | 'computer_use'\n    >\n  }\n  max_input_tokens?: number\n  max_output_tokens?: number\n  knowledge_cutoff?: string\n  pricing?: {\n    input: {\n      normal: number\n      cached?: number\n    }\n    output: {\n      normal: number\n    }\n  }\n  /**\n   * Type-level description of which provider options this model supports.\n   */\n  providerOptions?: TProviderOptions\n}\n\nconst GEMINI_3_1_PRO = {\n  name: 'gemini-3.1-pro-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 2.5,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions &\n    GeminiThinkingAdvancedOptions\n>\n\nconst GEMINI_3_PRO = {\n  name: 'gemini-3-pro-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 2.5,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions &\n    GeminiThinkingAdvancedOptions\n>\n\nconst GEMINI_3_FLASH = {\n  name: 'gemini-3-flash-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.5,\n    },\n    output: {\n      normal: 3,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions &\n    GeminiThinkingAdvancedOptions\n>\n\nconst GEMINI_3_PRO_IMAGE = {\n  name: 'gemini-3-pro-image-preview',\n  max_input_tokens: 65_536,\n  max_output_tokens: 32_768,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'structured_output', 'thinking'],\n    tools: ['google_search'],\n  },\n  pricing: {\n    input: {\n      normal: 2,\n    },\n    output: {\n      normal: 0.134,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions &\n    GeminiThinkingAdvancedOptions\n>\n\nconst GEMINI_3_1_FLASH_IMAGE = {\n  name: 'gemini-3.1-flash-image-preview',\n  max_input_tokens: 65_536,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'structured_output', 'thinking'],\n    tools: ['google_search'],\n  },\n  pricing: {\n    input: {\n      normal: 0.25,\n    },\n    output: {\n      normal: 1.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_3_1_FLASH_LITE = {\n  name: 'gemini-3.1-flash-lite-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.25,\n    },\n    output: {\n      normal: 1.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_PRO = {\n  name: 'gemini-2.5-pro',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: [\n      'code_execution',\n      'file_search',\n      'google_maps',\n      'google_search',\n      'url_context',\n    ],\n  },\n  pricing: {\n    input: {\n      normal: 2.5,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_PRO_TTS = {\n  name: 'gemini-2.5-pro-preview-tts',\n  max_input_tokens: 8_192,\n  max_output_tokens: 16_384,\n  knowledge_cutoff: '2025-05-01',\n  supports: {\n    input: ['text'],\n    output: ['audio'],\n    capabilities: ['audio_generation'],\n    tools: ['file_search'],\n  },\n  pricing: {\n    input: {\n      normal: 2.5,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst GEMINI_2_5_FLASH = {\n  name: 'gemini-2.5-flash',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: [\n      'code_execution',\n      'file_search',\n      'google_maps',\n      'google_search',\n      'url_context',\n    ],\n  },\n  pricing: {\n    input: {\n      normal: 1,\n    },\n    output: {\n      normal: 2.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_FLASH_PREVIEW = {\n  name: 'gemini-2.5-flash-preview-09-2025',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 1,\n    },\n    output: {\n      normal: 2.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_FLASH_IMAGE = {\n  name: 'gemini-2.5-flash-image',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-06-01',\n  supports: {\n    input: ['text', 'image'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'caching', 'structured_output'],\n    tools: ['file_search'],\n  },\n  pricing: {\n    input: {\n      normal: 0.3,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n/**\nconst GEMINI_2_5_FLASH_LIVE = {\n  name: 'gemini-2.5-flash-native-audio-preview-09-2025',\n  max_input_tokens: 141_072,\n  max_output_tokens: 8_192,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'audio', 'video'],\n    output: ['text', 'audio'],\n    capabilities: [\n      'audio_generation',\n      'file_search',\n      'function_calling',\n      'live_api',\n      'search_grounding',\n      'thinking',\n    ],\n  },\n  pricing: {\n    // todo find this info\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions &\n  GeminiThinkingOptions\n>\n*/\nconst GEMINI_2_5_FLASH_TTS = {\n  name: 'gemini-2.5-flash-preview-tts',\n  max_input_tokens: 8_192,\n  max_output_tokens: 16_384,\n  knowledge_cutoff: '2025-05-01',\n  supports: {\n    input: ['text'],\n    output: ['audio'],\n    capabilities: ['audio_generation', 'batch_api'],\n    tools: ['file_search'],\n  },\n  pricing: {\n    input: {\n      normal: 1,\n    },\n    output: {\n      normal: 2.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst GEMINI_2_5_FLASH_LITE = {\n  name: 'gemini-2.5-flash-lite',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'google_maps', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.1,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_FLASH_LITE_PREVIEW = {\n  name: 'gemini-2.5-flash-lite-preview-09-2025',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.1,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_FLASH = {\n  name: 'gemini-2.0-flash',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 8_192,\n  knowledge_cutoff: '2024-08-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'live_api',\n      'structured_output',\n    ],\n    tools: ['code_execution', 'google_maps', 'google_search'],\n  },\n  pricing: {\n    input: {\n      normal: 0.1,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions\n>\n\nconst GEMINI_2_FLASH_IMAGE = {\n  name: 'gemini-2.0-flash-preview-image-generation',\n  max_input_tokens: 32_768,\n  max_output_tokens: 8_192,\n  knowledge_cutoff: '2024-08-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video'],\n    output: ['text'],\n    capabilities: ['batch_api', 'caching', 'structured_output'],\n    tools: [],\n  },\n  pricing: {\n    input: {\n      normal: 0.1,\n    },\n    output: {\n      normal: 0.039,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n/* \nconst GEMINI_2_FLASH_LIVE = {\n  name: 'gemini-2.0-flash-live-001',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 8_192,\n  knowledge_cutoff: '2024-08-01',\n  supports: {\n    input: ['text', 'audio', 'video'],\n    output: ['text', 'audio'],\n    capabilities: [\n      'audio_generation',\n      'code_execution',\n      'function_calling',\n      'live_api',\n      'search_grounding',\n      'structured_output',\n      'url_context',\n    ],\n  },\n  pricing: {\n    // todo find this info\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions\n> */\n\nconst GEMINI_2_FLASH_LITE = {\n  name: 'gemini-2.0-flash-lite',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 8_192,\n  knowledge_cutoff: '2024-08-01',\n  supports: {\n    input: ['text', 'audio', 'video', 'image'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n    ],\n    tools: [],\n  },\n  pricing: {\n    input: {\n      normal: 0.075,\n    },\n    output: {\n      normal: 0.3,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions\n>\n\nconst IMAGEN_4_GENERATE = {\n  name: 'imagen-4.0-generate-001',\n  max_input_tokens: 480,\n  max_output_tokens: 4,\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst IMAGEN_4_GENERATE_ULTRA = {\n  name: 'imagen-4.0-ultra-generate-001',\n  max_input_tokens: 480,\n  max_output_tokens: 4,\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.6,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst IMAGEN_4_GENERATE_FAST = {\n  name: 'imagen-4.0-fast-generate-001',\n  max_input_tokens: 480,\n  max_output_tokens: 4,\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.2,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst IMAGEN_3 = {\n  name: 'imagen-3.0-generate-002',\n  max_output_tokens: 4,\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.03,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n/** \nconst VEO_3_1_PREVIEW = {\n  name: 'veo-3.1-generate-preview',\n  max_input_tokens: 1024,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions\n>\n\nconst VEO_3_1_FAST_PREVIEW = {\n  name: 'veo-3.1-fast-generate-preview',\n  max_input_tokens: 1024,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions\n>\n\nconst VEO_3 = {\n  name: 'veo-3.0-generate-001',\n  max_input_tokens: 1024,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions\n>\n\nconst VEO_3_FAST = {\n  name: 'veo-3.0-fast-generate-001',\n  max_input_tokens: 1024,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions\n>\n\nconst VEO_2 = {\n  name: 'veo-2.0-generate-001',\n  max_output_tokens: 2,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.35,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions\n> */\n\n/* const GEMINI_MODEL_META = {\n  [GEMINI_3_PRO.name]: GEMINI_3_PRO,\n  [GEMINI_2_5_PRO.name]: GEMINI_2_5_PRO,\n  [GEMINI_2_5_PRO_TTS.name]: GEMINI_2_5_PRO_TTS,\n  [GEMINI_2_5_FLASH.name]: GEMINI_2_5_FLASH,\n  [GEMINI_2_5_FLASH_PREVIEW.name]: GEMINI_2_5_FLASH_PREVIEW,\n  [GEMINI_2_5_FLASH_IMAGE.name]: GEMINI_2_5_FLASH_IMAGE,\n  [GEMINI_2_5_FLASH_LIVE.name]: GEMINI_2_5_FLASH_LIVE,\n  [GEMINI_2_5_FLASH_TTS.name]: GEMINI_2_5_FLASH_TTS,\n  [GEMINI_2_5_FLASH_LITE.name]: GEMINI_2_5_FLASH_LITE,\n  [GEMINI_2_5_FLASH_LITE_PREVIEW.name]: GEMINI_2_5_FLASH_LITE_PREVIEW,\n  [GEMINI_2_FLASH.name]: GEMINI_2_FLASH,\n  [GEMINI_2_FLASH_IMAGE.name]: GEMINI_2_FLASH_IMAGE,\n  [GEMINI_2_FLASH_LIVE.name]: GEMINI_2_FLASH_LIVE,\n  [GEMINI_2_FLASH_LITE.name]: GEMINI_2_FLASH_LITE,\n  [IMAGEN_4_GENERATE.name]: IMAGEN_4_GENERATE,\n  [IMAGEN_4_GENERATE_ULTRA.name]: IMAGEN_4_GENERATE_ULTRA,\n  [IMAGEN_4_GENERATE_FAST.name]: IMAGEN_4_GENERATE_FAST,\n  [IMAGEN_3.name]: IMAGEN_3,\n  [VEO_3_1_PREVIEW.name]: VEO_3_1_PREVIEW,\n  [VEO_3_1_FAST_PREVIEW.name]: VEO_3_1_FAST_PREVIEW,\n  [VEO_3.name]: VEO_3,\n  [VEO_3_FAST.name]: VEO_3_FAST,\n  [VEO_2.name]: VEO_2,\n} as const */\n\nexport const GEMINI_MODELS = [\n  GEMINI_3_1_PRO.name,\n  GEMINI_3_PRO.name,\n  GEMINI_3_FLASH.name,\n  GEMINI_3_1_FLASH_LITE.name,\n  GEMINI_2_5_PRO.name,\n  GEMINI_2_5_FLASH.name,\n  GEMINI_2_5_FLASH_PREVIEW.name,\n  GEMINI_2_5_FLASH_LITE.name,\n  GEMINI_2_5_FLASH_LITE_PREVIEW.name,\n  GEMINI_2_FLASH.name,\n  GEMINI_2_FLASH_LITE.name,\n] as const\n\nexport type GeminiModels = (typeof GEMINI_MODELS)[number]\n\nexport type GeminiImageModels = (typeof GEMINI_IMAGE_MODELS)[number]\n\nexport const GEMINI_IMAGE_MODELS = [\n  GEMINI_3_1_FLASH_IMAGE.name,\n  GEMINI_3_PRO_IMAGE.name,\n  GEMINI_2_5_FLASH_IMAGE.name,\n  GEMINI_2_FLASH_IMAGE.name,\n  IMAGEN_3.name,\n  IMAGEN_4_GENERATE.name,\n  IMAGEN_4_GENERATE_FAST.name,\n  IMAGEN_4_GENERATE_ULTRA.name,\n] as const\n\n/**\n * Text-to-speech models\n * @experimental Gemini TTS is an experimental feature and may change.\n */\nexport const GEMINI_TTS_MODELS = [\n  GEMINI_2_5_FLASH_TTS.name,\n  GEMINI_2_5_PRO_TTS.name,\n] as const\n\n/**\n * Available voice names for Gemini TTS\n * @see https://ai.google.dev/gemini-api/docs/speech-generation\n */\nexport const GEMINI_TTS_VOICES = [\n  'Zephyr',\n  'Puck',\n  'Charon',\n  'Kore',\n  'Fenrir',\n  'Leda',\n  'Orus',\n  'Aoede',\n  'Callirrhoe',\n  'Autonoe',\n  'Enceladus',\n  'Iapetus',\n  'Umbriel',\n  'Algieba',\n  'Despina',\n  'Erinome',\n  'Algenib',\n  'Rasalgethi',\n  'Laomedeia',\n  'Achernar',\n  'Alnilam',\n  'Schedar',\n  'Gacrux',\n  'Pulcherrima',\n  'Achird',\n  'Zubenelgenubi',\n  'Vindemiatrix',\n  'Sadachbia',\n  'Sadaltager',\n  'Sulafat',\n] as const\n\nexport type GeminiTTSVoice = (typeof GEMINI_TTS_VOICES)[number]\n\n/*   const GEMINI_AUDIO_MODELS = [\n  GEMINI_2_5_PRO_TTS.name,\n  GEMINI_2_5_FLASH_TTS.name,\n  GEMINI_2_5_FLASH_LIVE.name,\n  GEMINI_2_FLASH_LIVE.name,\n] as const\n\n  const GEMINI_VIDEO_MODELS = [\n  VEO_3_1_PREVIEW.name,\n  VEO_3_1_FAST_PREVIEW.name,\n  VEO_3.name,\n  VEO_3_FAST.name,\n  VEO_2.name,\n] as const */\n\n// Manual type map for per-model provider options\nexport type GeminiChatModelProviderOptionsByName = {\n  // Models with thinking and structured output support\n  [GEMINI_3_1_PRO.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions &\n    GeminiThinkingAdvancedOptions\n  [GEMINI_3_PRO.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions &\n    GeminiThinkingAdvancedOptions\n  [GEMINI_3_FLASH.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions &\n    GeminiThinkingAdvancedOptions\n  [GEMINI_3_1_FLASH_LITE.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_PRO.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_FLASH.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_FLASH_PREVIEW.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_FLASH_LITE.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_FLASH_LITE_PREVIEW.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  // Models with structured output but no thinking support\n  [GEMINI_2_FLASH.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions\n  [GEMINI_2_FLASH_LITE.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions\n}\n\n/**\n * Type-only map from chat model name to its supported tool capabilities.\n * Based on the 'supports.tools' arrays defined for each model.\n */\nexport type GeminiChatModelToolCapabilitiesByName = {\n  [GEMINI_3_1_PRO.name]: typeof GEMINI_3_1_PRO.supports.tools\n  [GEMINI_3_PRO.name]: typeof GEMINI_3_PRO.supports.tools\n  [GEMINI_3_FLASH.name]: typeof GEMINI_3_FLASH.supports.tools\n  [GEMINI_3_1_FLASH_LITE.name]: typeof GEMINI_3_1_FLASH_LITE.supports.tools\n  [GEMINI_2_5_PRO.name]: typeof GEMINI_2_5_PRO.supports.tools\n  [GEMINI_2_5_FLASH.name]: typeof GEMINI_2_5_FLASH.supports.tools\n  [GEMINI_2_5_FLASH_PREVIEW.name]: typeof GEMINI_2_5_FLASH_PREVIEW.supports.tools\n  [GEMINI_2_5_FLASH_LITE.name]: typeof GEMINI_2_5_FLASH_LITE.supports.tools\n  [GEMINI_2_5_FLASH_LITE_PREVIEW.name]: typeof GEMINI_2_5_FLASH_LITE_PREVIEW.supports.tools\n  [GEMINI_2_FLASH.name]: typeof GEMINI_2_FLASH.supports.tools\n  [GEMINI_2_FLASH_LITE.name]: typeof GEMINI_2_FLASH_LITE.supports.tools\n}\n\n/**\n * Type-only map from chat model name to its supported input modalities.\n * Based on the 'supports.input' arrays defined for each model.\n * Note: 'document' in the model meta is mapped to 'document' modality.\n * Used by the core AI types to constrain ContentPart types based on the selected model.\n * Note: These must be inlined as readonly arrays (not typeof) because the model\n * constants are not exported and typeof references don't work in .d.ts files\n * when consumed by external packages.\n *\n * @see https://ai.google.dev/gemini-api/docs/vision\n * @see https://ai.google.dev/gemini-api/docs/audio\n * @see https://ai.google.dev/gemini-api/docs/document-processing\n */\nexport type GeminiModelInputModalitiesByName = {\n  // Models with full multimodal support (text, image, audio, video, document)\n  [GEMINI_3_1_PRO.name]: typeof GEMINI_3_1_PRO.supports.input\n  [GEMINI_3_PRO.name]: typeof GEMINI_3_PRO.supports.input\n  [GEMINI_3_FLASH.name]: typeof GEMINI_3_FLASH.supports.input\n  [GEMINI_3_1_FLASH_LITE.name]: typeof GEMINI_3_1_FLASH_LITE.supports.input\n  [GEMINI_2_5_PRO.name]: typeof GEMINI_2_5_PRO.supports.input\n  [GEMINI_2_5_FLASH_LITE.name]: typeof GEMINI_2_5_FLASH_LITE.supports.input\n  [GEMINI_2_5_FLASH_LITE_PREVIEW.name]: typeof GEMINI_2_5_FLASH_LITE_PREVIEW.supports.input\n\n  // Models with text, image, audio, video (no document)\n  [GEMINI_2_5_FLASH.name]: typeof GEMINI_2_5_FLASH.supports.input\n  [GEMINI_2_5_FLASH_PREVIEW.name]: typeof GEMINI_2_5_FLASH_PREVIEW.supports.input\n  [GEMINI_2_FLASH.name]: typeof GEMINI_2_FLASH.supports.input\n  [GEMINI_2_FLASH_LITE.name]: typeof GEMINI_2_FLASH_LITE.supports.input\n}\n"],"names":[],"mappings":"AAoDA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AAwBR;AAUA,MAAM,eAAe;AAAA,EACnB,MAAM;AAwBR;AAUA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AAwBR;AAUA,MAAM,qBAAqB;AAAA,EACzB,MAAM;AAkBR;AAUA,MAAM,yBAAyB;AAAA,EAC7B,MAAM;AAkBR;AASA,MAAM,wBAAwB;AAAA,EAC5B,MAAM;AAwBR;AASA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AA8BR;AASA,MAAM,qBAAqB;AAAA,EACzB,MAAM;AAkBR;AAOA,MAAM,mBAAmB;AAAA,EACvB,MAAM;AA8BR;AASA,MAAM,2BAA2B;AAAA,EAC/B,MAAM;AAwBR;AASA,MAAM,yBAAyB;AAAA,EAC7B,MAAM;AAkBR;AAyCA,MAAM,uBAAuB;AAAA,EAC3B,MAAM;AAkBR;AAOA,MAAM,wBAAwB;AAAA,EAC5B,MAAM;AAwBR;AASA,MAAM,gCAAgC;AAAA,EACpC,MAAM;AAwBR;AASA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AAwBR;AAQA,MAAM,uBAAuB;AAAA,EAC3B,MAAM;AAkBR;AAyCA,MAAM,sBAAsB;AAAA,EAC1B,MAAM;AAuBR;AAQA,MAAM,oBAAoB;AAAA,EACxB,MAAM;AAeR;AAOA,MAAM,0BAA0B;AAAA,EAC9B,MAAM;AAeR;AAOA,MAAM,yBAAyB;AAAA,EAC7B,MAAM;AAeR;AAOA,MAAM,WAAW;AAAA,EACf,MAAM;AAcR;AAmJO,MAAM,gBAAgB;AAAA,EAC3B,eAAe;AAAA,EACf,aAAa;AAAA,EACb,eAAe;AAAA,EACf,sBAAsB;AAAA,EACtB,eAAe;AAAA,EACf,iBAAiB;AAAA,EACjB,yBAAyB;AAAA,EACzB,sBAAsB;AAAA,EACtB,8BAA8B;AAAA,EAC9B,eAAe;AAAA,EACf,oBAAoB;AACtB;AAMO,MAAM,sBAAsB;AAAA,EACjC,uBAAuB;AAAA,EACvB,mBAAmB;AAAA,EACnB,uBAAuB;AAAA,EACvB,qBAAqB;AAAA,EACrB,SAAS;AAAA,EACT,kBAAkB;AAAA,EAClB,uBAAuB;AAAA,EACvB,wBAAwB;AAC1B;AAMO,MAAM,oBAAoB;AAAA,EAC/B,qBAAqB;AAAA,EACrB,mBAAmB;AACrB;AAMO,MAAM,oBAAoB;AAAA,EAC/B;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;"}
+{"version":3,"file":"model-meta.js","sources":["../../src/model-meta.ts"],"sourcesContent":["import type {\n  GeminiCachedContentOptions,\n  GeminiCommonConfigOptions,\n  GeminiSafetyOptions,\n  GeminiStructuredOutputOptions,\n  GeminiThinkingAdvancedOptions,\n  GeminiThinkingOptions,\n  GeminiToolConfigOptions,\n} from './text/text-provider-options'\n\ninterface ModelMeta<TProviderOptions = unknown> {\n  name: string\n  supports: {\n    input: Array<'text' | 'image' | 'audio' | 'video' | 'document'>\n    output: Array<'text' | 'image' | 'audio' | 'video'>\n    capabilities?: Array<\n      | 'audio_generation'\n      | 'batch_api'\n      | 'caching'\n      | 'function_calling'\n      | 'live_api'\n      | 'structured_output'\n      | 'thinking'\n    >\n    tools?: Array<\n      | 'code_execution'\n      | 'file_search'\n      | 'google_search'\n      | 'google_search_retrieval'\n      | 'google_maps'\n      | 'url_context'\n      | 'computer_use'\n    >\n  }\n  max_input_tokens?: number\n  max_output_tokens?: number\n  knowledge_cutoff?: string\n  pricing?: {\n    input: {\n      normal: number\n      cached?: number\n    }\n    output: {\n      normal: number\n    }\n  }\n  /**\n   * Type-level description of which provider options this model supports.\n   */\n  providerOptions?: TProviderOptions\n}\n\nconst GEMINI_3_1_PRO = {\n  name: 'gemini-3.1-pro-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 2.5,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions &\n    GeminiThinkingAdvancedOptions\n>\n\nconst GEMINI_3_PRO = {\n  name: 'gemini-3-pro-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 2.5,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions &\n    GeminiThinkingAdvancedOptions\n>\n\nconst GEMINI_3_FLASH = {\n  name: 'gemini-3-flash-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.5,\n    },\n    output: {\n      normal: 3,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions &\n    GeminiThinkingAdvancedOptions\n>\n\nconst GEMINI_3_PRO_IMAGE = {\n  name: 'gemini-3-pro-image-preview',\n  max_input_tokens: 65_536,\n  max_output_tokens: 32_768,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'structured_output', 'thinking'],\n    tools: ['google_search'],\n  },\n  pricing: {\n    input: {\n      normal: 2,\n    },\n    output: {\n      normal: 0.134,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions &\n    GeminiThinkingAdvancedOptions\n>\n\nconst GEMINI_3_1_FLASH_IMAGE = {\n  name: 'gemini-3.1-flash-image-preview',\n  max_input_tokens: 65_536,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'structured_output', 'thinking'],\n    tools: ['google_search'],\n  },\n  pricing: {\n    input: {\n      normal: 0.25,\n    },\n    output: {\n      normal: 1.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_3_1_FLASH_LITE = {\n  name: 'gemini-3.1-flash-lite-preview',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.25,\n    },\n    output: {\n      normal: 1.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_PRO = {\n  name: 'gemini-2.5-pro',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: [\n      'code_execution',\n      'file_search',\n      'google_maps',\n      'google_search',\n      'url_context',\n    ],\n  },\n  pricing: {\n    input: {\n      normal: 2.5,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_PRO_TTS = {\n  name: 'gemini-2.5-pro-preview-tts',\n  max_input_tokens: 8_192,\n  max_output_tokens: 16_384,\n  knowledge_cutoff: '2025-05-01',\n  supports: {\n    input: ['text'],\n    output: ['audio'],\n    capabilities: ['audio_generation'],\n  },\n  pricing: {\n    input: {\n      normal: 2.5,\n    },\n    output: {\n      normal: 15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst GEMINI_2_5_FLASH = {\n  name: 'gemini-2.5-flash',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: [\n      'code_execution',\n      'file_search',\n      'google_maps',\n      'google_search',\n      'url_context',\n    ],\n  },\n  pricing: {\n    input: {\n      normal: 1,\n    },\n    output: {\n      normal: 2.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_FLASH_PREVIEW = {\n  name: 'gemini-2.5-flash-preview-09-2025',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'file_search', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 1,\n    },\n    output: {\n      normal: 2.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_FLASH_IMAGE = {\n  name: 'gemini-2.5-flash-image',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-06-01',\n  supports: {\n    input: ['text', 'image'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'caching', 'structured_output'],\n    tools: ['file_search'],\n  },\n  pricing: {\n    input: {\n      normal: 0.3,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n/**\nconst GEMINI_2_5_FLASH_LIVE = {\n  name: 'gemini-2.5-flash-native-audio-preview-09-2025',\n  max_input_tokens: 141_072,\n  max_output_tokens: 8_192,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'audio', 'video'],\n    output: ['text', 'audio'],\n    capabilities: [\n      'audio_generation',\n      'file_search',\n      'function_calling',\n      'live_api',\n      'search_grounding',\n      'thinking',\n    ],\n  },\n  pricing: {\n    // todo find this info\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions &\n  GeminiThinkingOptions\n>\n*/\nconst GEMINI_2_5_FLASH_TTS = {\n  name: 'gemini-2.5-flash-preview-tts',\n  max_input_tokens: 8_192,\n  max_output_tokens: 16_384,\n  knowledge_cutoff: '2025-05-01',\n  supports: {\n    input: ['text'],\n    output: ['audio'],\n    capabilities: ['audio_generation', 'batch_api'],\n  },\n  pricing: {\n    input: {\n      normal: 1,\n    },\n    output: {\n      normal: 2.5,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\n/**\n * Gemini 3.1 Flash TTS Preview - latest expressive TTS model with\n * 200+ audio tags, 70+ languages, and multi-speaker dialogue support.\n * @see https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-tts-preview\n */\nconst GEMINI_3_1_FLASH_TTS = {\n  name: 'gemini-3.1-flash-tts-preview',\n  max_input_tokens: 32_768,\n  max_output_tokens: 16_384,\n  knowledge_cutoff: '2025-05-01',\n  supports: {\n    input: ['text'],\n    output: ['audio'],\n    capabilities: ['audio_generation', 'batch_api'],\n  },\n  pricing: {\n    input: {\n      normal: 0.5,\n    },\n    output: {\n      normal: 10,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\n/**\n * Lyria 3 Pro Preview — Google's flagship music generation model.\n * Generates full-length songs with multiple verses, choruses, and bridges.\n * Outputs MP3 or WAV at 48 kHz stereo.\n * @see https://ai.google.dev/gemini-api/docs/models/lyria-3-pro-preview\n */\nconst LYRIA_3_PRO = {\n  name: 'lyria-3-pro-preview',\n  max_input_tokens: 131_072,\n  supports: {\n    input: ['text', 'image'],\n    output: ['audio'],\n    capabilities: ['audio_generation'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0,\n    },\n  },\n} as const satisfies ModelMeta\n\n/**\n * Lyria 3 Clip Preview — 30-second music clips in MP3 format.\n * @see https://ai.google.dev/gemini-api/docs/music-generation\n */\nconst LYRIA_3_CLIP = {\n  name: 'lyria-3-clip-preview',\n  max_input_tokens: 131_072,\n  supports: {\n    input: ['text', 'image'],\n    output: ['audio'],\n    capabilities: ['audio_generation'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0,\n    },\n  },\n} as const satisfies ModelMeta\n\nconst GEMINI_2_5_FLASH_LITE = {\n  name: 'gemini-2.5-flash-lite',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'google_maps', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.1,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_5_FLASH_LITE_PREVIEW = {\n  name: 'gemini-2.5-flash-lite-preview-09-2025',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 65_536,\n  knowledge_cutoff: '2025-01-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video', 'document'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n      'thinking',\n    ],\n    tools: ['code_execution', 'google_search', 'url_context'],\n  },\n  pricing: {\n    input: {\n      normal: 0.1,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n>\n\nconst GEMINI_2_FLASH = {\n  name: 'gemini-2.0-flash',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 8_192,\n  knowledge_cutoff: '2024-08-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'live_api',\n      'structured_output',\n    ],\n    tools: ['code_execution', 'google_maps', 'google_search'],\n  },\n  pricing: {\n    input: {\n      normal: 0.1,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions\n>\n\nconst GEMINI_2_FLASH_IMAGE = {\n  name: 'gemini-2.0-flash-preview-image-generation',\n  max_input_tokens: 32_768,\n  max_output_tokens: 8_192,\n  knowledge_cutoff: '2024-08-01',\n  supports: {\n    input: ['text', 'image', 'audio', 'video'],\n    output: ['text', 'image'],\n    capabilities: ['batch_api', 'caching', 'structured_output'],\n    tools: [],\n  },\n  pricing: {\n    input: {\n      normal: 0.1,\n    },\n    output: {\n      normal: 0.039,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n/* \nconst GEMINI_2_FLASH_LIVE = {\n  name: 'gemini-2.0-flash-live-001',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 8_192,\n  knowledge_cutoff: '2024-08-01',\n  supports: {\n    input: ['text', 'audio', 'video'],\n    output: ['text', 'audio'],\n    capabilities: [\n      'audio_generation',\n      'code_execution',\n      'function_calling',\n      'live_api',\n      'search_grounding',\n      'structured_output',\n      'url_context',\n    ],\n  },\n  pricing: {\n    // todo find this info\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions\n> */\n\nconst GEMINI_2_FLASH_LITE = {\n  name: 'gemini-2.0-flash-lite',\n  max_input_tokens: 1_048_576,\n  max_output_tokens: 8_192,\n  knowledge_cutoff: '2024-08-01',\n  supports: {\n    input: ['text', 'audio', 'video', 'image'],\n    output: ['text'],\n    capabilities: [\n      'batch_api',\n      'caching',\n      'function_calling',\n      'structured_output',\n    ],\n    tools: [],\n  },\n  pricing: {\n    input: {\n      normal: 0.075,\n    },\n    output: {\n      normal: 0.3,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions\n>\n\nconst IMAGEN_4_GENERATE = {\n  name: 'imagen-4.0-generate-001',\n  max_input_tokens: 480,\n  max_output_tokens: 4,\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst IMAGEN_4_GENERATE_ULTRA = {\n  name: 'imagen-4.0-ultra-generate-001',\n  max_input_tokens: 480,\n  max_output_tokens: 4,\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.6,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst IMAGEN_4_GENERATE_FAST = {\n  name: 'imagen-4.0-fast-generate-001',\n  max_input_tokens: 480,\n  max_output_tokens: 4,\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.2,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n\nconst IMAGEN_3 = {\n  name: 'imagen-3.0-generate-002',\n  max_output_tokens: 4,\n  supports: {\n    input: ['text'],\n    output: ['image'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.03,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions\n>\n/** \nconst VEO_3_1_PREVIEW = {\n  name: 'veo-3.1-generate-preview',\n  max_input_tokens: 1024,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions\n>\n\nconst VEO_3_1_FAST_PREVIEW = {\n  name: 'veo-3.1-fast-generate-preview',\n  max_input_tokens: 1024,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions\n>\n\nconst VEO_3 = {\n  name: 'veo-3.0-generate-001',\n  max_input_tokens: 1024,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.4,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions\n>\n\nconst VEO_3_FAST = {\n  name: 'veo-3.0-fast-generate-001',\n  max_input_tokens: 1024,\n  max_output_tokens: 1,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video', 'audio'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.15,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions\n>\n\nconst VEO_2 = {\n  name: 'veo-2.0-generate-001',\n  max_output_tokens: 2,\n  supports: {\n    input: ['text', 'image'],\n    output: ['video'],\n  },\n  pricing: {\n    input: {\n      normal: 0,\n    },\n    output: {\n      normal: 0.35,\n    },\n  },\n} as const satisfies ModelMeta<\n  GeminiToolConfigOptions &\n  GeminiSafetyOptions &\n  GeminiGenerationConfigOptions &\n  GeminiCachedContentOptions\n> */\n\n/* const GEMINI_MODEL_META = {\n  [GEMINI_3_PRO.name]: GEMINI_3_PRO,\n  [GEMINI_2_5_PRO.name]: GEMINI_2_5_PRO,\n  [GEMINI_2_5_PRO_TTS.name]: GEMINI_2_5_PRO_TTS,\n  [GEMINI_2_5_FLASH.name]: GEMINI_2_5_FLASH,\n  [GEMINI_2_5_FLASH_PREVIEW.name]: GEMINI_2_5_FLASH_PREVIEW,\n  [GEMINI_2_5_FLASH_IMAGE.name]: GEMINI_2_5_FLASH_IMAGE,\n  [GEMINI_2_5_FLASH_LIVE.name]: GEMINI_2_5_FLASH_LIVE,\n  [GEMINI_2_5_FLASH_TTS.name]: GEMINI_2_5_FLASH_TTS,\n  [GEMINI_2_5_FLASH_LITE.name]: GEMINI_2_5_FLASH_LITE,\n  [GEMINI_2_5_FLASH_LITE_PREVIEW.name]: GEMINI_2_5_FLASH_LITE_PREVIEW,\n  [GEMINI_2_FLASH.name]: GEMINI_2_FLASH,\n  [GEMINI_2_FLASH_IMAGE.name]: GEMINI_2_FLASH_IMAGE,\n  [GEMINI_2_FLASH_LIVE.name]: GEMINI_2_FLASH_LIVE,\n  [GEMINI_2_FLASH_LITE.name]: GEMINI_2_FLASH_LITE,\n  [IMAGEN_4_GENERATE.name]: IMAGEN_4_GENERATE,\n  [IMAGEN_4_GENERATE_ULTRA.name]: IMAGEN_4_GENERATE_ULTRA,\n  [IMAGEN_4_GENERATE_FAST.name]: IMAGEN_4_GENERATE_FAST,\n  [IMAGEN_3.name]: IMAGEN_3,\n  [VEO_3_1_PREVIEW.name]: VEO_3_1_PREVIEW,\n  [VEO_3_1_FAST_PREVIEW.name]: VEO_3_1_FAST_PREVIEW,\n  [VEO_3.name]: VEO_3,\n  [VEO_3_FAST.name]: VEO_3_FAST,\n  [VEO_2.name]: VEO_2,\n} as const */\n\nexport const GEMINI_MODELS = [\n  GEMINI_3_1_PRO.name,\n  GEMINI_3_PRO.name,\n  GEMINI_3_FLASH.name,\n  GEMINI_3_1_FLASH_LITE.name,\n  GEMINI_2_5_PRO.name,\n  GEMINI_2_5_FLASH.name,\n  GEMINI_2_5_FLASH_PREVIEW.name,\n  GEMINI_2_5_FLASH_LITE.name,\n  GEMINI_2_5_FLASH_LITE_PREVIEW.name,\n  GEMINI_2_FLASH.name,\n  GEMINI_2_FLASH_LITE.name,\n] as const\n\nexport type GeminiModels = (typeof GEMINI_MODELS)[number]\n\nexport type GeminiImageModels = (typeof GEMINI_IMAGE_MODELS)[number]\n\nexport const GEMINI_IMAGE_MODELS = [\n  GEMINI_3_1_FLASH_IMAGE.name,\n  GEMINI_3_PRO_IMAGE.name,\n  GEMINI_2_5_FLASH_IMAGE.name,\n  GEMINI_2_FLASH_IMAGE.name,\n  IMAGEN_3.name,\n  IMAGEN_4_GENERATE.name,\n  IMAGEN_4_GENERATE_FAST.name,\n  IMAGEN_4_GENERATE_ULTRA.name,\n] as const\n\n/**\n * Text-to-speech models\n * @experimental Gemini TTS is an experimental feature and may change.\n */\nexport const GEMINI_TTS_MODELS = [\n  GEMINI_3_1_FLASH_TTS.name,\n  GEMINI_2_5_FLASH_TTS.name,\n  GEMINI_2_5_PRO_TTS.name,\n] as const\n\n/**\n * Audio generation models (Lyria music generation).\n * @experimental Lyria music generation is an experimental feature and may change.\n */\nexport const GEMINI_AUDIO_MODELS = [\n  LYRIA_3_PRO.name,\n  LYRIA_3_CLIP.name,\n] as const\n\n/**\n * Available voice names for Gemini TTS\n * @see https://ai.google.dev/gemini-api/docs/speech-generation\n */\nexport const GEMINI_TTS_VOICES = [\n  'Zephyr',\n  'Puck',\n  'Charon',\n  'Kore',\n  'Fenrir',\n  'Leda',\n  'Orus',\n  'Aoede',\n  'Callirrhoe',\n  'Autonoe',\n  'Enceladus',\n  'Iapetus',\n  'Umbriel',\n  'Algieba',\n  'Despina',\n  'Erinome',\n  'Algenib',\n  'Rasalgethi',\n  'Laomedeia',\n  'Achernar',\n  'Alnilam',\n  'Schedar',\n  'Gacrux',\n  'Pulcherrima',\n  'Achird',\n  'Zubenelgenubi',\n  'Vindemiatrix',\n  'Sadachbia',\n  'Sadaltager',\n  'Sulafat',\n] as const\n\nexport type GeminiTTSVoice = (typeof GEMINI_TTS_VOICES)[number]\n\n/*   const GEMINI_AUDIO_MODELS = [\n  GEMINI_2_5_PRO_TTS.name,\n  GEMINI_2_5_FLASH_TTS.name,\n  GEMINI_2_5_FLASH_LIVE.name,\n  GEMINI_2_FLASH_LIVE.name,\n] as const\n\n  const GEMINI_VIDEO_MODELS = [\n  VEO_3_1_PREVIEW.name,\n  VEO_3_1_FAST_PREVIEW.name,\n  VEO_3.name,\n  VEO_3_FAST.name,\n  VEO_2.name,\n] as const */\n\n// Manual type map for per-model provider options\nexport type GeminiChatModelProviderOptionsByName = {\n  // Models with thinking and structured output support\n  [GEMINI_3_1_PRO.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions &\n    GeminiThinkingAdvancedOptions\n  [GEMINI_3_PRO.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions &\n    GeminiThinkingAdvancedOptions\n  [GEMINI_3_FLASH.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions &\n    GeminiThinkingAdvancedOptions\n  [GEMINI_3_1_FLASH_LITE.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_PRO.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_FLASH.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_FLASH_PREVIEW.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_FLASH_LITE.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  [GEMINI_2_5_FLASH_LITE_PREVIEW.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions &\n    GeminiThinkingOptions\n  // Models with structured output but no thinking support\n  [GEMINI_2_FLASH.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions\n  [GEMINI_2_FLASH_LITE.name]: GeminiToolConfigOptions &\n    GeminiSafetyOptions &\n    GeminiCommonConfigOptions &\n    GeminiCachedContentOptions &\n    GeminiStructuredOutputOptions\n}\n\n/**\n * Type-only map from chat model name to its supported tool capabilities.\n * Based on the 'supports.tools' arrays defined for each model.\n */\nexport type GeminiChatModelToolCapabilitiesByName = {\n  [GEMINI_3_1_PRO.name]: typeof GEMINI_3_1_PRO.supports.tools\n  [GEMINI_3_PRO.name]: typeof GEMINI_3_PRO.supports.tools\n  [GEMINI_3_FLASH.name]: typeof GEMINI_3_FLASH.supports.tools\n  [GEMINI_3_1_FLASH_LITE.name]: typeof GEMINI_3_1_FLASH_LITE.supports.tools\n  [GEMINI_2_5_PRO.name]: typeof GEMINI_2_5_PRO.supports.tools\n  [GEMINI_2_5_FLASH.name]: typeof GEMINI_2_5_FLASH.supports.tools\n  [GEMINI_2_5_FLASH_PREVIEW.name]: typeof GEMINI_2_5_FLASH_PREVIEW.supports.tools\n  [GEMINI_2_5_FLASH_LITE.name]: typeof GEMINI_2_5_FLASH_LITE.supports.tools\n  [GEMINI_2_5_FLASH_LITE_PREVIEW.name]: typeof GEMINI_2_5_FLASH_LITE_PREVIEW.supports.tools\n  [GEMINI_2_FLASH.name]: typeof GEMINI_2_FLASH.supports.tools\n  [GEMINI_2_FLASH_LITE.name]: typeof GEMINI_2_FLASH_LITE.supports.tools\n}\n\n/**\n * Type-only map from chat model name to its supported input modalities.\n * Based on the 'supports.input' arrays defined for each model.\n * Note: 'document' in the model meta is mapped to 'document' modality.\n * Used by the core AI types to constrain ContentPart types based on the selected model.\n * Note: These must be inlined as readonly arrays (not typeof) because the model\n * constants are not exported and typeof references don't work in .d.ts files\n * when consumed by external packages.\n *\n * @see https://ai.google.dev/gemini-api/docs/vision\n * @see https://ai.google.dev/gemini-api/docs/audio\n * @see https://ai.google.dev/gemini-api/docs/document-processing\n */\nexport type GeminiModelInputModalitiesByName = {\n  // Models with full multimodal support (text, image, audio, video, document)\n  [GEMINI_3_1_PRO.name]: typeof GEMINI_3_1_PRO.supports.input\n  [GEMINI_3_PRO.name]: typeof GEMINI_3_PRO.supports.input\n  [GEMINI_3_FLASH.name]: typeof GEMINI_3_FLASH.supports.input\n  [GEMINI_3_1_FLASH_LITE.name]: typeof GEMINI_3_1_FLASH_LITE.supports.input\n  [GEMINI_2_5_PRO.name]: typeof GEMINI_2_5_PRO.supports.input\n  [GEMINI_2_5_FLASH_LITE.name]: typeof GEMINI_2_5_FLASH_LITE.supports.input\n  [GEMINI_2_5_FLASH_LITE_PREVIEW.name]: typeof GEMINI_2_5_FLASH_LITE_PREVIEW.supports.input\n\n  // Models with text, image, audio, video (no document)\n  [GEMINI_2_5_FLASH.name]: typeof GEMINI_2_5_FLASH.supports.input\n  [GEMINI_2_5_FLASH_PREVIEW.name]: typeof GEMINI_2_5_FLASH_PREVIEW.supports.input\n  [GEMINI_2_FLASH.name]: typeof GEMINI_2_FLASH.supports.input\n  [GEMINI_2_FLASH_LITE.name]: typeof GEMINI_2_FLASH_LITE.supports.input\n}\n"],"names":[],"mappings":"AAoDA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AAwBR;AAUA,MAAM,eAAe;AAAA,EACnB,MAAM;AAwBR;AAUA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AAwBR;AAUA,MAAM,qBAAqB;AAAA,EACzB,MAAM;AAkBR;AAUA,MAAM,yBAAyB;AAAA,EAC7B,MAAM;AAkBR;AASA,MAAM,wBAAwB;AAAA,EAC5B,MAAM;AAwBR;AASA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AA8BR;AASA,MAAM,qBAAqB;AAAA,EACzB,MAAM;AAiBR;AAOA,MAAM,mBAAmB;AAAA,EACvB,MAAM;AA8BR;AASA,MAAM,2BAA2B;AAAA,EAC/B,MAAM;AAwBR;AASA,MAAM,yBAAyB;AAAA,EAC7B,MAAM;AAkBR;AAyCA,MAAM,uBAAuB;AAAA,EAC3B,MAAM;AAiBR;AAYA,MAAM,uBAAuB;AAAA,EAC3B,MAAM;AAiBR;AAaA,MAAM,cAAc;AAAA,EAClB,MAAM;AAeR;AAMA,MAAM,eAAe;AAAA,EACnB,MAAM;AAeR;AAEA,MAAM,wBAAwB;AAAA,EAC5B,MAAM;AAwBR;AASA,MAAM,gCAAgC;AAAA,EACpC,MAAM;AAwBR;AASA,MAAM,iBAAiB;AAAA,EACrB,MAAM;AAwBR;AAQA,MAAM,uBAAuB;AAAA,EAC3B,MAAM;AAkBR;AAyCA,MAAM,sBAAsB;AAAA,EAC1B,MAAM;AAuBR;AAQA,MAAM,oBAAoB;AAAA,EACxB,MAAM;AAeR;AAOA,MAAM,0BAA0B;AAAA,EAC9B,MAAM;AAeR;AAOA,MAAM,yBAAyB;AAAA,EAC7B,MAAM;AAeR;AAOA,MAAM,WAAW;AAAA,EACf,MAAM;AAcR;AAmJO,MAAM,gBAAgB;AAAA,EAC3B,eAAe;AAAA,EACf,aAAa;AAAA,EACb,eAAe;AAAA,EACf,sBAAsB;AAAA,EACtB,eAAe;AAAA,EACf,iBAAiB;AAAA,EACjB,yBAAyB;AAAA,EACzB,sBAAsB;AAAA,EACtB,8BAA8B;AAAA,EAC9B,eAAe;AAAA,EACf,oBAAoB;AACtB;AAMO,MAAM,sBAAsB;AAAA,EACjC,uBAAuB;AAAA,EACvB,mBAAmB;AAAA,EACnB,uBAAuB;AAAA,EACvB,qBAAqB;AAAA,EACrB,SAAS;AAAA,EACT,kBAAkB;AAAA,EAClB,uBAAuB;AAAA,EACvB,wBAAwB;AAC1B;AAMO,MAAM,oBAAoB;AAAA,EAC/B,qBAAqB;AAAA,EACrB,qBAAqB;AAAA,EACrB,mBAAmB;AACrB;AAMO,MAAM,sBAAsB;AAAA,EACjC,YAAY;AAAA,EACZ,aAAa;AACf;AAMO,MAAM,oBAAoB;AAAA,EAC/B;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tanstack/ai-gemini",
-  "version": "0.9.1",
+  "version": "0.10.0",
   "description": "Google Gemini adapter for TanStack AI",
   "author": "",
   "license": "MIT",
@@ -37,13 +37,13 @@
     "@google/genai": "^1.43.0"
   },
   "peerDependencies": {
-    "@tanstack/ai": "^0.13.0"
+    "@tanstack/ai": "^0.14.0"
   },
   "devDependencies": {
     "@vitest/coverage-v8": "4.0.14",
     "vite": "^7.2.7",
     "zod": "^4.2.0",
-    "@tanstack/ai": "0.13.0"
+    "@tanstack/ai": "0.14.0"
   },
   "scripts": {
     "build": "vite build",

package/src/adapters/audio.ts

@@ -0,0 +1,185 @@
+import { BaseAudioAdapter } from '@tanstack/ai/adapters'
+import {
+  createGeminiClient,
+  generateId,
+  getGeminiApiKeyFromEnv,
+} from '../utils'
+import type { GEMINI_AUDIO_MODELS } from '../model-meta'
+import type {
+  AudioGenerationOptions,
+  AudioGenerationResult,
+} from '@tanstack/ai'
+import type { GoogleGenAI } from '@google/genai'
+import type { GeminiClientConfig } from '../utils'
+
+/**
+ * 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 class GeminiAudioAdapter<
+  TModel extends GeminiAudioModel,
+> extends BaseAudioAdapter<TModel, GeminiAudioProviderOptions> {
+  readonly name = 'gemini' as const
+
+  private client: GoogleGenAI
+
+  constructor(config: GeminiAudioConfig, model: TModel) {
+    super(model, config)
+    this.client = createGeminiClient(config)
+  }
+
+  async generateAudio(
+    options: AudioGenerationOptions<GeminiAudioProviderOptions>,
+  ): Promise<AudioGenerationResult> {
+    const { model, prompt, modelOptions, logger } = options
+
+    logger.request(`activity=generateAudio provider=gemini model=${model}`, {
+      provider: 'gemini',
+      model,
+    })
+
+    try {
+      // FIXME (SDK audit): Lyria 3 music generation may not belong on
+      // generateContent at all — @google/genai exposes a `LiveMusicSession`
+      // (`ai.live.music.connect`) with a `musicGenerationConfig` object.
+      // `seed` is valid on GenerateContentConfig, and Lyria always returns
+      // MP3 today, so we don't forward `responseMimeType` either.
+      // The runtime test `emits only GenerateContentConfig-valid fields`
+      // asserts the config shape so a later SDK audit can catch regressions.
+      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: any) =>
+        part.inlineData?.mimeType?.startsWith('audio/'),
+      )
+
+      if (!audioPart?.inlineData?.data) {
+        throw new Error('No audio data in Gemini Lyria response')
+      }
+
+      // audioPart was selected because mimeType.startsWith('audio/') was
+      // truthy, so the mime type is guaranteed to be a string here. Trust the
+      // value Gemini returned rather than inventing a non-standard
+      // `audio/mp3` fallback (IANA is `audio/mpeg`).
+      const contentType = audioPart.inlineData.mimeType
+
+      return {
+        id: generateId(this.name),
+        model,
+        audio: {
+          b64Json: audioPart.inlineData.data,
+          contentType,
+        },
+      }
+    } catch (error) {
+      logger.errors('gemini.generateAudio fatal', {
+        error,
+        source: 'gemini.generateAudio',
+      })
+      throw error
+    }
+  }
+}
+
+/**
+ * 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 function createGeminiAudio<TModel extends GeminiAudioModel>(
+  model: TModel,
+  apiKey: string,
+  config?: Omit<GeminiAudioConfig, 'apiKey'>,
+): GeminiAudioAdapter<TModel> {
+  // Put apiKey LAST so caller-supplied config can't silently override the
+  // explicit argument.
+  return new GeminiAudioAdapter({ ...config, apiKey }, model)
+}
+
+/**
+ * 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 function geminiAudio<TModel extends GeminiAudioModel>(
+  model: TModel,
+  config?: Omit<GeminiAudioConfig, 'apiKey'>,
+): GeminiAudioAdapter<TModel> {
+  const apiKey = getGeminiApiKeyFromEnv()
+  return createGeminiAudio(model, apiKey, config)
+}