Nano Banana

Instructions

Implementation Reference

• mcp-server/src/server.ts:121-172 (schema)

  Tool schema definition including input parameters for generate_image tool (prompt, outputCount, styles, variations, etc.) used in ListTools response.
  { name: "generate_image", description: "Generate single or multiple images from text prompts with style and variation options", inputSchema: { type: "object", properties: { prompt: { type: "string", description: "The text prompt describing the image to generate", }, outputCount: { type: "number", description: "Number of variations to generate (1-8, default: 1)", minimum: 1, maximum: 8, default: 1, }, styles: { type: "array", items: { type: "string" }, description: "Array of artistic styles: photorealistic, watercolor, oil-painting, sketch, pixel-art, anime, vintage, modern, abstract, minimalist", }, variations: { type: "array", items: { type: "string" }, description: "Array of variation types: lighting, angle, color-palette, composition, mood, season, time-of-day", }, format: { type: "string", enum: ["grid", "separate"], description: "Output format: separate files or single grid image", default: "separate", }, seed: { type: "number", description: "Seed for reproducible variations", }, preview: { type: "boolean", description: "Automatically open generated images in default viewer", default: false, }, }, required: ["prompt"], }, },
• mcp-server/src/server.ts:494-512 (handler)

  MCP CallToolRequest handler case for 'generate_image': parses arguments, builds ImageGenerationRequest, and delegates execution to ImageGenerator.generateTextToImage().
  case "generate_image": { const imageRequest: ImageGenerationRequest = { prompt: args?.prompt as string, outputCount: (args?.outputCount as number) || 1, mode: "generate", styles: args?.styles as string[], variations: args?.variations as string[], format: (args?.format as "grid" | "separate") || "separate", seed: args?.seed as number, preview: args?.preview as boolean, fileFormat: (args?.fileFormat as "png" | "jpeg") || "png", noPreview: (args?.noPreview as boolean) || (args?.["no-preview"] as boolean), }; response = await this.imageGenerator.generateTextToImage(imageRequest); break; }
• mcp-server/src/imageGenerator.ts:532-638 (handler)

  Core handler function implementing generate_image tool logic: builds batch prompts, calls OpenRouter API via postJson, parses base64 image data, saves files using FileHandler, handles multiple variations, previews, and returns generated files list.
  async generateTextToImage( request: ImageGenerationRequest ): Promise<ImageGenerationResponse> { try { const outputPath = FileHandler.ensureOutputDirectory(); const generatedFiles: string[] = []; const prompts = this.buildBatchPrompts(request); let firstError: string | null = null; const fileFormat = this.resolveFileFormat(request); logger.debug(`Generating ${prompts.length} image variation(s)`); for (let i = 0; i < prompts.length; i++) { const currentPrompt = prompts[i]; logger.debug( `Generating variation ${i + 1}/${prompts.length}:`, currentPrompt ); try { const payload: Record<string, unknown> = { model: this.modelName, input: [ { role: "user", content: [ { type: "input_text", text: currentPrompt, }, ], }, ], }; if (request.seed !== undefined) { payload.seed = request.seed; } const response = await this.postJson<OpenRouterImageResponse>( this.generationPath, payload ); const imageBase64 = this.parseImageFromResponse(response); if (imageBase64) { const filename = FileHandler.generateFilename( request.styles || request.variations ? currentPrompt : request.prompt, fileFormat, i ); const fullPath = await FileHandler.saveImageFromBase64( imageBase64, outputPath, filename ); generatedFiles.push(fullPath); logger.debug("Image saved to:", fullPath); } else { logger.warn("No valid image data found in OpenRouter response"); } } catch (error: unknown) { const errorMessage = this.handleApiError(error); if (!firstError) { firstError = errorMessage; } logger.warn(`Error generating variation ${i + 1}:`, errorMessage); if (errorMessage.toLowerCase().includes("authentication failed")) { return { success: false, message: "Image generation failed", error: errorMessage, }; } } } if (generatedFiles.length === 0) { return { success: false, message: "Failed to generate any images", error: firstError || "No image data returned from OpenRouter. Try adjusting your prompt.", }; } await this.handlePreview(generatedFiles, request); return { success: true, message: `Successfully generated ${generatedFiles.length} image variation(s)`, generatedFiles, }; } catch (error: unknown) { logger.error("Error in generateTextToImage:", error); return { success: false, message: "Failed to generate image", error: this.handleApiError(error), }; } }
• mcp-server/src/imageGenerator.ts:346-422 (helper)

  Helper utility to generate multiple prompt variations from styles, variations parameters, and outputCount for batch image generation.
  private buildBatchPrompts(request: ImageGenerationRequest): string[] { const prompts: string[] = []; const basePrompt = request.prompt; if (!request.styles && !request.variations && !request.outputCount) { return [basePrompt]; } if (request.styles && request.styles.length > 0) { for (const style of request.styles) { prompts.push(`${basePrompt}, ${style} style`); } } if (request.variations && request.variations.length > 0) { const basePrompts = prompts.length > 0 ? prompts : [basePrompt]; const variationPrompts: string[] = []; for (const baseP of basePrompts) { for (const variation of request.variations) { switch (variation) { case "lighting": variationPrompts.push(`${baseP}, dramatic lighting`); variationPrompts.push(`${baseP}, soft lighting`); break; case "angle": variationPrompts.push(`${baseP}, from above`); variationPrompts.push(`${baseP}, close-up view`); break; case "color-palette": variationPrompts.push(`${baseP}, warm color palette`); variationPrompts.push(`${baseP}, cool color palette`); break; case "composition": variationPrompts.push(`${baseP}, centered composition`); variationPrompts.push(`${baseP}, rule of thirds composition`); break; case "mood": variationPrompts.push(`${baseP}, cheerful mood`); variationPrompts.push(`${baseP}, dramatic mood`); break; case "season": variationPrompts.push(`${baseP}, in spring`); variationPrompts.push(`${baseP}, in winter`); break; case "time-of-day": variationPrompts.push(`${baseP}, at sunrise`); variationPrompts.push(`${baseP}, at sunset`); break; default: variationPrompts.push(`${baseP}, ${variation}`); break; } } } if (variationPrompts.length > 0) { prompts.splice(0, prompts.length, ...variationPrompts); } } if ( prompts.length === 0 && request.outputCount && request.outputCount > 1 ) { for (let i = 0; i < request.outputCount; i++) { prompts.push(basePrompt); } } if (request.outputCount && prompts.length > request.outputCount) { prompts.splice(request.outputCount); } return prompts.length > 0 ? prompts : [basePrompt]; }