@core-ai/google-genai 0.13.1 → 0.15.0

package/README.md

@@ -26,3 +26,23 @@ const result = await generate({
 
 console.log(result.content);
 ```
+
+## Image generation
+
+`imageModel()` supports Gemini native image models and dedicated Imagen models:
+
+```ts
+import { generateImage } from '@core-ai/core-ai';
+
+const result = await generateImage({
+    model: google.imageModel('gemini-2.5-flash-image'),
+    prompt: 'A watercolor robot in a mountain cabin at sunrise',
+    size: '1024x1024',
+});
+
+console.log(result.images[0]?.base64);
+```
+
+Gemini model IDs use native multimodal generation and return base64 image
+data. Imagen model IDs such as `imagen-4.0-generate-001` use the dedicated
+Imagen API and also support generating multiple images with `n`.

package/dist/index.d.ts

@@ -2,17 +2,25 @@ import { GoogleGenAI } from '@google/genai';
 import { ChatModel, EmbeddingModel, ImageModel, ModelCapabilities } from '@core-ai/core-ai';
 import { z } from 'zod';
 
-type GoogleGenAIProviderOptions = {
+type GoogleGenAIClient = {
+    models: GoogleGenAI['models'];
+};
+type GoogleGenAIProviderBaseOptions = {
     apiKey?: string;
     apiVersion?: string;
     baseUrl?: string;
-    client?: GoogleGenAI;
+    client?: GoogleGenAIClient;
+};
+type GoogleGenAIProviderFactoryOptions = {
+    providerId?: string;
 };
+type GoogleGenAIProviderOptions = GoogleGenAIProviderBaseOptions;
 type GoogleGenAIProvider = {
     chatModel(modelId: string): ChatModel;
     embeddingModel(modelId: string): EmbeddingModel;
     imageModel(modelId: string): ImageModel;
 };
+declare function createGoogleGenAIProvider(options?: GoogleGenAIProviderBaseOptions, factoryOptions?: GoogleGenAIProviderFactoryOptions): GoogleGenAIProvider;
 declare function createGoogleGenAI(options?: GoogleGenAIProviderOptions): GoogleGenAIProvider;
 
 type GoogleReasoningMetadata = {
@@ -90,4 +98,4 @@ declare const googleProviderOptionsSchema: z.ZodObject<{
 }, z.core.$strict>;
 type GoogleProviderOptions = GoogleGenerateProviderOptions;
 
-export { type GoogleEmbedProviderOptions, type GoogleGenAIProvider, type GoogleGenAIProviderOptions, type GoogleGenerateProviderOptions, type GoogleImageProviderOptions, type GoogleModelCapabilities, type GoogleProviderOptions as GoogleModelProviderOptions, type GoogleReasoningMetadata, createGoogleGenAI, getGoogleModelCapabilities, googleEmbedProviderOptionsSchema, googleGenerateProviderOptionsSchema, googleImageProviderOptionsSchema, googleProviderOptionsSchema };
+export { type GoogleEmbedProviderOptions, type GoogleGenAIClient, type GoogleGenAIProvider, type GoogleGenAIProviderBaseOptions, type GoogleGenAIProviderFactoryOptions, type GoogleGenAIProviderOptions, type GoogleGenerateProviderOptions, type GoogleImageProviderOptions, type GoogleModelCapabilities, type GoogleProviderOptions as GoogleModelProviderOptions, type GoogleReasoningMetadata, createGoogleGenAI, createGoogleGenAIProvider, getGoogleModelCapabilities, googleEmbedProviderOptionsSchema, googleGenerateProviderOptionsSchema, googleImageProviderOptionsSchema, googleProviderOptionsSchema };

package/dist/index.js

@@ -439,6 +439,7 @@ async function* transformStream(stream) {
   const bufferedToolCalls = /* @__PURE__ */ new Map();
   let finishReason = "unknown";
   let sawToolCalls = false;
+  let textOpen = false;
   let reasoningOpen = false;
   let usage = {
     inputTokens: 0,
@@ -459,10 +460,25 @@ async function* transformStream(stream) {
       providerMetadata: { google: {} }
     };
   };
+  const startText = function* () {
+    if (textOpen) {
+      return;
+    }
+    textOpen = true;
+    yield { type: "text-start" };
+  };
+  const closeText = function* () {
+    if (!textOpen) {
+      return;
+    }
+    textOpen = false;
+    yield { type: "text-end" };
+  };
   for await (const chunk of stream) {
     usage = mapUsage(chunk, usage);
     const reasoningDeltas = extractReasoningDeltas(chunk);
     if (reasoningDeltas.length > 0) {
+      yield* closeText();
       if (!reasoningOpen) {
         reasoningOpen = true;
         yield {
@@ -481,6 +497,7 @@ async function* transformStream(stream) {
       if (reasoningEnd2) {
         yield reasoningEnd2;
       }
+      yield* startText();
       yield {
         type: "text-delta",
         text: chunk.text
@@ -488,6 +505,7 @@ async function* transformStream(stream) {
     }
     const functionCalls = chunk.functionCalls ?? [];
     if (functionCalls.length > 0) {
+      yield* closeText();
       const reasoningEnd2 = closeReasoning();
       if (reasoningEnd2) {
         yield reasoningEnd2;
@@ -538,6 +556,7 @@ async function* transformStream(stream) {
   if (reasoningEnd) {
     yield reasoningEnd;
   }
+  yield* closeText();
   for (const toolCall of bufferedToolCalls.values()) {
     yield {
       type: "tool-call-end",
@@ -665,33 +684,32 @@ function mapUsage(response, fallback) {
 // src/google-error.ts
 import { ApiError } from "@google/genai";
 import { ProviderError } from "@core-ai/core-ai";
-function wrapGoogleError(error) {
+function wrapGoogleError(error, provider = "google") {
   if (error instanceof ApiError) {
-    return new ProviderError(error.message, "google", error.status, error);
+    return new ProviderError(error.message, provider, error.status, error);
   }
   return new ProviderError(
     error instanceof Error ? error.message : String(error),
-    "google",
+    provider,
     void 0,
     error
   );
 }
 
 // src/chat-model.ts
-function createGoogleGenAIChatModel(client, modelId) {
-  const provider = "google";
+function createGoogleGenAIChatModel(client, modelId, provider = "google") {
   async function callGenerateContentApi(request) {
     try {
       return await client.models.generateContent(request);
     } catch (error) {
-      throw wrapGoogleError(error);
+      throw wrapGoogleError(error, provider);
     }
   }
   async function callGenerateContentStreamApi(request) {
     try {
       return await client.models.generateContentStream(request);
     } catch (error) {
-      throw wrapGoogleError(error);
+      throw wrapGoogleError(error, provider);
     }
   }
   async function generateChat(options) {
@@ -899,9 +917,9 @@ function formatZodIssues(issues) {
 }
 
 // src/embedding-model.ts
-function createGoogleGenAIEmbeddingModel(client, modelId) {
+function createGoogleGenAIEmbeddingModel(client, modelId, provider = "google") {
   return {
-    provider: "google",
+    provider,
     modelId,
     async embed(options) {
       try {
@@ -942,7 +960,7 @@ function createGoogleGenAIEmbeddingModel(client, modelId) {
           usage
         };
       } catch (error) {
-        throw wrapGoogleError(error);
+        throw wrapGoogleError(error, provider);
       }
     }
   };
@@ -957,46 +975,107 @@ function mapGoogleEmbedProviderOptionsToConfig(options) {
 }
 
 // src/image-model.ts
-function createGoogleGenAIImageModel(client, modelId) {
+function createGoogleGenAIImageModel(client, modelId, provider = "google") {
   return {
-    provider: "google",
+    provider,
     modelId,
     async generate(options) {
       try {
-        const baseRequest = {
-          model: modelId,
-          prompt: options.prompt,
-          config: {
-            ...options.n !== void 0 ? { numberOfImages: options.n } : {},
-            ...mapSizeToImageConfig(options.size)
-          }
-        };
         const googleOptions = parseGoogleImageProviderOptions(
           options.providerOptions
         );
-        const providerConfig = mapGoogleImageProviderOptionsToConfig(googleOptions);
-        const request = Object.keys(providerConfig).length > 0 ? {
-          ...baseRequest,
-          config: {
-            ...baseRequest.config,
-            ...providerConfig
-          }
-        } : baseRequest;
-        const response = await client.models.generateImages(request);
-        return {
-          images: (response.generatedImages ?? []).map((image) => ({
-            base64: image.image?.imageBytes ?? void 0,
-            url: image.image?.gcsUri ?? void 0,
-            revisedPrompt: image.enhancedPrompt ?? void 0
-          }))
-        };
+        return isGeminiImageModel(modelId) ? await generateGeminiImages(
+          client,
+          modelId,
+          options,
+          googleOptions
+        ) : await generateImagenImages(
+          client,
+          modelId,
+          options,
+          googleOptions
+        );
       } catch (error) {
-        throw wrapGoogleError(error);
+        throw wrapGoogleError(error, provider);
+      }
+    }
+  };
+}
+function isGeminiImageModel(modelId) {
+  return modelId.startsWith("gemini-");
+}
+async function generateGeminiImages(client, modelId, options, googleOptions) {
+  if (options.n !== void 0 && options.n > 1) {
+    throw new Error(
+      "Gemini image generation does not support n greater than 1."
+    );
+  }
+  const sizeConfig = mapSizeToImageConfig(options.size);
+  const aspectRatio = googleOptions?.aspectRatio ?? sizeConfig["aspectRatio"];
+  const imageSize = supportsGeminiImageSize(modelId) ? googleOptions?.imageSize ?? sizeConfig["imageSize"] : void 0;
+  const imageConfig = {
+    ...aspectRatio ? { aspectRatio } : {},
+    ...imageSize ? { imageSize } : {}
+  };
+  const request = {
+    model: modelId,
+    contents: [
+      {
+        role: "user",
+        parts: [{ text: options.prompt }]
+      }
+    ],
+    config: {
+      responseModalities: ["IMAGE"],
+      ...Object.keys(imageConfig).length > 0 ? { imageConfig } : {},
+      ...googleOptions?.seed !== void 0 ? { seed: googleOptions.seed } : {}
+    }
+  };
+  const response = await client.models.generateContent(request);
+  return mapGeminiImageResponse(response);
+}
+function supportsGeminiImageSize(modelId) {
+  return !modelId.startsWith("gemini-2.5-flash-image");
+}
+function mapGeminiImageResponse(response) {
+  const parts = response.candidates?.[0]?.content?.parts ?? [];
+  return {
+    images: parts.flatMap((part) => {
+      const inlineData = part.inlineData;
+      if (!inlineData?.data || !inlineData.mimeType?.startsWith("image/")) {
+        return [];
       }
+      return [{ base64: inlineData.data }];
+    })
+  };
+}
+async function generateImagenImages(client, modelId, options, googleOptions) {
+  const baseRequest = {
+    model: modelId,
+    prompt: options.prompt,
+    config: {
+      ...options.n !== void 0 ? { numberOfImages: options.n } : {},
+      ...mapSizeToImageConfig(options.size)
     }
   };
+  const providerConfig = mapImagenImageProviderOptionsToConfig(googleOptions);
+  const request = Object.keys(providerConfig).length > 0 ? {
+    ...baseRequest,
+    config: {
+      ...baseRequest.config,
+      ...providerConfig
+    }
+  } : baseRequest;
+  const response = await client.models.generateImages(request);
+  return {
+    images: (response.generatedImages ?? []).map((image) => ({
+      base64: image.image?.imageBytes ?? void 0,
+      url: image.image?.gcsUri ?? void 0,
+      revisedPrompt: image.enhancedPrompt ?? void 0
+    }))
+  };
 }
-function mapGoogleImageProviderOptionsToConfig(options) {
+function mapImagenImageProviderOptionsToConfig(options) {
   return {
     ...options?.outputGcsUri !== void 0 ? { outputGcsUri: options.outputGcsUri } : {},
     ...options?.negativePrompt !== void 0 ? { negativePrompt: options.negativePrompt } : {},
@@ -1052,7 +1131,8 @@ function greatestCommonDivisor(a, b) {
 }
 
 // src/provider.ts
-function createGoogleGenAI(options = {}) {
+var DEFAULT_PROVIDER_ID = "google";
+function createGoogleGenAIProvider(options = {}, factoryOptions = {}) {
   const client = options.client ?? new GoogleGenAI({
     apiKey: options.apiKey,
     ...options.apiVersion ? { apiVersion: options.apiVersion } : {},
@@ -1062,14 +1142,19 @@ function createGoogleGenAI(options = {}) {
       }
     } : {}
   });
+  const providerId = factoryOptions.providerId ?? DEFAULT_PROVIDER_ID;
   return {
-    chatModel: (modelId) => createGoogleGenAIChatModel(client, modelId),
-    embeddingModel: (modelId) => createGoogleGenAIEmbeddingModel(client, modelId),
-    imageModel: (modelId) => createGoogleGenAIImageModel(client, modelId)
+    chatModel: (modelId) => createGoogleGenAIChatModel(client, modelId, providerId),
+    embeddingModel: (modelId) => createGoogleGenAIEmbeddingModel(client, modelId, providerId),
+    imageModel: (modelId) => createGoogleGenAIImageModel(client, modelId, providerId)
   };
 }
+function createGoogleGenAI(options = {}) {
+  return createGoogleGenAIProvider(options);
+}
 export {
   createGoogleGenAI,
+  createGoogleGenAIProvider,
   getGoogleModelCapabilities,
   googleEmbedProviderOptionsSchema,
   googleGenerateProviderOptionsSchema,

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@core-ai/google-genai",
-    "version": "0.13.1",
+    "version": "0.15.0",
     "description": "Google GenAI provider package for @core-ai/core-ai",
     "license": "MIT",
     "author": "Omnifact (https://omnifact.ai)",
@@ -45,7 +45,7 @@
         "test:watch": "vitest"
     },
     "dependencies": {
-        "@core-ai/core-ai": "^0.13.1",
+        "@core-ai/core-ai": "^0.15.0",
         "@google/genai": "^1.42.0"
     },
     "peerDependencies": {