generate_image
Generate custom images using Baidu's iRAG API by inputting descriptive prompts and optional parameters like size, quantity, and reference images. Ideal for creating visuals tailored to specific needs or ideas.
Instructions
使用百度iRAG API生成图片
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| guidance | No | 指导密度值（仅flux.1-schnell模型支持），默认为3.5 | |
| n | No | 生成图片数量，默认为1 | |
| prompt | Yes | 图片生成提示词，描述要生成的图片内容 | |
| refer_image | No | 参考图片URL（可选） | |
| seed | No | 随机种子（仅flux.1-schnell模型支持） | |
| size | No | 图片尺寸，默认为1024x1024 | 1024x1024 |
| steps | No | 采样步数（仅flux.1-schnell模型支持） |
Implementation Reference
- src/server/mcp-server.ts:166-249 (handler)Main MCP tool handler for 'generate_image': validates params with Zod schema, maps to Irag API request, calls IragClient.generateImage, processes image URLs with image-saver utility, constructs CallToolResult with text descriptions of generated/saved images.
private async handleGenerateImage(params: GenerateImageParams): Promise<CallToolResult> { // 验证参数 const validatedParams = generateImageParamsSchema.parse(params); logger.info('开始处理图片生成请求', validatedParams); try { // 调用百度iRAG API const requestParams: IragGenerationRequest = { model: this.config.defaultModel, prompt: validatedParams.prompt, n: validatedParams.n || 1, size: validatedParams.size || '1024x1024', }; // 只有当参数存在时才添加可选参数 if (validatedParams.refer_image !== undefined) { requestParams.refer_image = validatedParams.refer_image; } if (validatedParams.steps !== undefined) { requestParams.steps = validatedParams.steps; } if (validatedParams.seed !== undefined) { requestParams.seed = validatedParams.seed; } if (validatedParams.guidance !== undefined) { requestParams.guidance = validatedParams.guidance; } const response = await this.iragClient.generateImage(requestParams); // 构建响应内容 const content: TextContent[] = [ { type: 'text', text: `成功生成${response.data.length}张图片`, } as TextContent, ]; // 处理图片 const imageUrls = response.data.map(item => item.url); const imageResults = await processImages( imageUrls, this.config.resourceMode, this.config.basePath ); // 添加图片处理结果 for (const [index, result] of imageResults.entries()) { if (result.success) { // 添加文本说明 let textMessage = `图片 ${index + 1} 生成成功`; if (result.localPath) { textMessage += `\n📁 已保存到: ${result.localPath}`; } if (this.config.resourceMode === 'url') { textMessage += `\n🔗 原始URL: ${result.url}`; } content.push({ type: 'text', text: textMessage, } as TextContent); } else { // 处理失败的情况 logger.error(`图片 ${index + 1} 处理失败`, result.error); content.push({ type: 'text', text: `图片 ${index + 1} 处理失败: ${result.error || '未知错误'}\n🔗 原始URL: ${result.url}`, } as TextContent); } } logger.info('图片生成成功', { requestId: response.id, imageCount: response.data.length, }); return { content }; } catch (error) { logger.error('图片生成失败', error); throw error; } } - src/server/mcp-server.ts:73-126 (registration)Tool registration in ListToolsRequestHandler: defines 'generate_image' tool with detailed inputSchema matching GenerateImageParams.
{ name: 'generate_image', description: '使用百度iRAG API生成图片', inputSchema: { type: 'object', properties: { prompt: { type: 'string', description: '图片生成提示词，描述要生成的图片内容', }, refer_image: { type: 'string', format: 'uri', description: '参考图片URL（可选）', }, n: { type: 'integer', minimum: 1, maximum: 4, description: '生成图片数量，默认为1', default: 1, }, size: { type: 'string', enum: [ '512x512', '768x768', '1024x768', '1024x1024' ], description: '图片尺寸，默认为1024x1024', default: '1024x1024', }, steps: { type: 'integer', minimum: 1, maximum: 50, description: '采样步数（仅flux.1-schnell模型支持）', }, seed: { type: 'integer', minimum: 0, maximum: 4294967295, description: '随机种子（仅flux.1-schnell模型支持）', }, guidance: { type: 'number', minimum: 0, maximum: 30, description: '指导密度值（仅flux.1-schnell模型支持），默认为3.5', default: 3.5, }, }, required: ['prompt'], }, }, - src/server/mcp-server.ts:18-28 (schema)Zod validation schema for GenerateImageParams used in the tool handler.
const generateImageParamsSchema = z.object({ prompt: z.string().min(1, '提示词不能为空'), refer_image: z.string().url().optional(), n: z.number().int().min(1).max(4).optional(), size: z.enum([ '512x512', '768x768', '1024x768', '1024x1024' ]).optional(), steps: z.number().int().min(1).max(50).optional(), seed: z.number().int().min(0).max(4294967295).optional(), guidance: z.number().min(0).max(30).optional(), }); - src/client/irag-client.ts:144-172 (helper)Core helper function in IragClient that performs the HTTP POST to Baidu Qianfan iRAG API endpoint '/v2/images/generations', includes param validation, retry logic with exponential backoff, and detailed error handling.
async generateImage(params: IragGenerationRequest): Promise<IragGenerationResponse> { const validatedParams = this.validateRequest(params); logger.info('开始生成图片', { params: validatedParams }); return this.withRetry(async () => { try { const response = await this.client.post<IragGenerationResponse>( '/v2/images/generations', validatedParams ); logger.info('图片生成成功', { id: response.data.id, imageCount: response.data.data.length, }); return response.data; } catch (error) { if (error instanceof AxiosError && error.response) { const errorData = error.response.data as IragErrorResponse; const errorMessage = `API错误 (${error.response.status}): ${errorData.message || error.message}`; logger.error(errorMessage, errorData); throw new Error(errorMessage); } throw error; } }); } - src/types/index.ts:44-52 (schema)TypeScript interface defining the input parameters for the generate_image tool.
export interface GenerateImageParams { prompt: string; refer_image?: string | undefined; n?: number | undefined; size?: ImageSize | undefined; steps?: number | undefined; seed?: number | undefined; guidance?: number | undefined; }