diff --git a/src/lib/openrouter.ts b/src/lib/openrouter.ts
index 38e6616..6e42403 100644
--- a/src/lib/openrouter.ts
+++ b/src/lib/openrouter.ts
@@ -7,6 +7,8 @@ interface OpenRouterModel {
     modality: string
     tokenizer: string
     instruct_type?: string
+    input_modalities?: string[]
+    output_modalities?: string[]
   }
   pricing: {
     prompt: string
@@ -63,12 +65,15 @@ export class OpenRouterService {
 
   // 将 OpenRouter 模型转换为我们数据库的格式
   transformModelForDB(model: OpenRouterModel) {
+    const allOutputTypes = this.determineAllOutputTypes(model)
+    const primaryOutputType = this.determinePrimaryOutputType(allOutputTypes)
+    
     return {
       modelId: model.id,
       name: model.name,
       provider: this.extractProvider(model.id),
       serviceProvider: 'openrouter',
-      outputType: 'text',
+      outputType: primaryOutputType, // 主要输出类型
       description: model.description || null,
       maxTokens: model.context_length || null,
       inputCostPer1k: parseFloat(model.pricing.prompt) * 1000 || null,
@@ -77,6 +82,9 @@ export class OpenRouterService {
         modality: model.architecture.modality,
         tokenizer: model.architecture.tokenizer,
         instruct_type: model.architecture.instruct_type,
+        input_modalities: model.architecture.input_modalities,
+        output_modalities: model.architecture.output_modalities,
+        all_output_types: allOutputTypes, // 记录所有支持的输出类型
         is_moderated: model.top_provider.is_moderated,
         max_completion_tokens: model.top_provider.max_completion_tokens,
       },
@@ -88,6 +96,83 @@ export class OpenRouterService {
     }
   }
 
+  // 根据模型的架构信息确定所有支持的输出类型
+  private determineAllOutputTypes(model: OpenRouterModel): string[] {
+    const supportedTypes: Set<string> = new Set()
+    
+    // 优先使用 output_modalities 字段
+    if (model.architecture.output_modalities && Array.isArray(model.architecture.output_modalities)) {
+      model.architecture.output_modalities.forEach(modality => {
+        supportedTypes.add(modality)
+      })
+    }
+    
+    // 如果没有 output_modalities，则通过 modality 字段推断
+    if (model.architecture.modality) {
+      const modality = model.architecture.modality.toLowerCase()
+      
+      // 检查 modality 字段中的输出类型指示
+      if (modality.includes('->image') || (modality.includes('image') && modality.includes('->'))) {
+        supportedTypes.add('image')
+      }
+      
+      if (modality.includes('->video') || (modality.includes('video') && modality.includes('->'))) {
+        supportedTypes.add('video')
+      }
+      
+      if (modality.includes('->audio') || (modality.includes('audio') && modality.includes('->'))) {
+        supportedTypes.add('audio')
+      }
+      
+      if (modality.includes('->text') || (modality.includes('text') && modality.includes('->'))) {
+        supportedTypes.add('text')
+      }
+    }
+    
+    // 通过模型名称进行启发式判断
+    const modelName = model.name?.toLowerCase() || ''
+    const modelId = model.id?.toLowerCase() || ''
+    
+    // 图像生成模型关键词
+    const imageKeywords = ['dall-e', 'dalle', 'midjourney', 'stable-diffusion', 'flux', 'imagen', 'ideogram']
+    if (imageKeywords.some(keyword => modelName.includes(keyword) || modelId.includes(keyword))) {
+      supportedTypes.add('image')
+    }
+    
+    // 视频生成模型关键词
+    const videoKeywords = ['runway', 'gen-2', 'pika', 'video', 'sora']
+    if (videoKeywords.some(keyword => modelName.includes(keyword) || modelId.includes(keyword))) {
+      supportedTypes.add('video')
+    }
+    
+    // 音频生成模型关键词
+    const audioKeywords = ['whisper', 'musicgen', 'bark', 'audio', 'speech', 'tts']
+    if (audioKeywords.some(keyword => modelName.includes(keyword) || modelId.includes(keyword))) {
+      supportedTypes.add('audio')
+    }
+    
+    // 如果没有找到任何输出类型，默认为 text
+    if (supportedTypes.size === 0) {
+      supportedTypes.add('text')
+    }
+    
+    return Array.from(supportedTypes)
+  }
+
+  // 确定主要输出类型（用于数据库的 outputType 字段）
+  private determinePrimaryOutputType(supportedTypes: string[]): string {
+    // 按优先级返回主要类型
+    const priority = ['image', 'video', 'audio', 'text']
+    
+    for (const type of priority) {
+      if (supportedTypes.includes(type)) {
+        return type
+      }
+    }
+    
+    return 'text' // 默认返回 text
+  }
+
   private extractProvider(modelId: string): string {
     // 从模型 ID 中提取提供商名称，如 "openai/gpt-4" -> "OpenAI"
     const providerMap: Record<string, string> = {