@lobehub/chat 1.105.0 → 1.105.1

package/CHANGELOG.md

@@ -2,6 +2,31 @@
 
 # Changelog
 
+### [Version 1.105.1](https://github.com/lobehub/lobe-chat/compare/v1.105.0...v1.105.1)
+
+<sup>Released on **2025-07-29**</sup>
+
+#### 💄 Styles
+
+- **misc**: Support more Text2Image from Qwen.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### Styles
+
+- **misc**: Support more Text2Image from Qwen, closes [#8574](https://github.com/lobehub/lobe-chat/issues/8574) ([b8c0e2d](https://github.com/lobehub/lobe-chat/commit/b8c0e2d))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
 ## [Version 1.105.0](https://github.com/lobehub/lobe-chat/compare/v1.104.5...v1.105.0)
 
 <sup>Released on **2025-07-28**</sup>

package/changelog/v1.json

@@ -1,4 +1,13 @@
 [
+  {
+    "children": {
+      "improvements": [
+        "Support more Text2Image from Qwen."
+      ]
+    },
+    "date": "2025-07-29",
+    "version": "1.105.1"
+  },
   {
     "children": {
       "features": [

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lobehub/chat",
-  "version": "1.105.0",
+  "version": "1.105.1",
   "description": "Lobe Chat - an open-source, high-performance chatbot framework that supports speech synthesis, multimodal, and extensible Function Call plugin system. Supports one-click free deployment of your private ChatGPT/LLM web application.",
   "keywords": [
     "framework",

package/src/config/aiModels/qwen.ts

@@ -958,9 +958,42 @@ const qwenChatModels: AIChatModelCard[] = [
 
 const qwenImageModels: AIImageModelCard[] = [
   {
-    description: '阿里云通义旗下的文生图模型',
-    displayName: 'Wanxiang T2I Turbo',
+    description: '万相2.2极速版，当前最新模型。在创意性、稳定性、写实质感上全面升级，生成速度快，性价比高。',
+    displayName: 'Wanxiang2.2 T2I Flash',
     enabled: true,
+    id: 'wan2.2-t2i-flash',
+    organization: 'Qwen',
+    parameters: {
+      height: { default: 1024, max: 1440, min: 512, step: 1 },
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      width: { default: 1024, max: 1440, min: 512, step: 1 },
+    },
+    releasedAt: '2025-07-28',
+    type: 'image',
+  },
+  {
+    description: '万相2.2专业版，当前最新模型。在创意性、稳定性、写实质感上全面升级，生成细节丰富。',
+    displayName: 'Wanxiang2.2 T2I Plus',
+    enabled: true,
+    id: 'wan2.2-t2i-plus',
+    organization: 'Qwen',
+    parameters: {
+      height: { default: 1024, max: 1440, min: 512, step: 1 },
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      width: { default: 1024, max: 1440, min: 512, step: 1 },
+    },
+    releasedAt: '2025-07-28',
+    type: 'image',
+  },
+  {
+    description: '全面升级版本。生成速度快、效果全面、综合性价比高。对应通义万相官网2.1极速模型。',
+    displayName: 'Wanxiang2.1 T2I Turbo',
     id: 'wanx2.1-t2i-turbo',
     organization: 'Qwen',
     parameters: {
@@ -974,6 +1007,178 @@ const qwenImageModels: AIImageModelCard[] = [
     releasedAt: '2025-01-08',
     type: 'image',
   },
+  {
+    description: '全面升级版本。生成图像细节更丰富，速度稍慢。对应通义万相官网2.1专业模型。',
+    displayName: 'Wanxiang2.1 T2I Plus',
+    id: 'wanx2.1-t2i-plus',
+    organization: 'Qwen',
+    parameters: {
+      height: { default: 1024, max: 1440, min: 512, step: 1 },
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      width: { default: 1024, max: 1440, min: 512, step: 1 },
+    },
+    releasedAt: '2025-01-08',
+    type: 'image',
+  },
+  {
+    description: '擅长质感人像，速度中等、成本较低。对应通义万相官网2.0极速模型。',
+    displayName: 'Wanxiang2.0 T2I Turbo',
+    id: 'wanx2.0-t2i-turbo',
+    organization: 'Qwen',
+    parameters: {
+      height: { default: 1024, max: 1440, min: 512, step: 1 },
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      width: { default: 1024, max: 1440, min: 512, step: 1 },
+    },
+    releasedAt: '2025-01-17',
+    type: 'image',
+  },
+  {
+    description: '基础文生图模型。对应通义万相官网1.0通用模型。',
+    displayName: 'Wanxiang v1',
+    id: 'wanx-v1',
+    organization: 'Qwen',
+    parameters: {
+      height: { default: 1024, max: 1440, min: 512, step: 1 },
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      width: { default: 1024, max: 1440, min: 512, step: 1 },
+    },
+    releasedAt: '2024-05-22',
+    type: 'image',
+  },
+  {
+    description: 'FLUX.1 [schnell] 作为目前开源最先进的少步模型，不仅超越了同类竞争者，甚至还优于诸如 Midjourney v6.0 和 DALL·E 3 (HD) 等强大的非精馏模型。该模型经过专门微调，以保留预训练阶段的全部输出多样性，相较于当前市场上的最先进模型，FLUX.1 [schnell] 显著提升了在视觉质量、指令遵从、尺寸/比例变化、字体处理及输出多样性等方面的可能，为用户带来更为丰富多样的创意图像生成体验。',
+    displayName: 'FLUX.1 [schnell]',
+    enabled: true,
+    id: 'flux-schnell',
+    organization: 'Qwen',
+    parameters: {
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      size: {
+        default: '1024x1024',
+        enum: ['512x1024', '768x512', '768x1024', '1024x576', '576x1024', '1024x1024'],
+      },
+      steps: { default: 4, max: 12, min: 1 },
+    },
+    releasedAt: '2024-08-07',
+    type: 'image',
+  },
+  {
+    description: 'FLUX.1 [dev] 是一款面向非商业应用的开源权重、精炼模型。FLUX.1 [dev] 在保持了与FLUX专业版相近的图像质量和指令遵循能力的同时，具备更高的运行效率。相较于同尺寸的标准模型，它在资源利用上更为高效。',
+    displayName: 'FLUX.1 [dev]',
+    enabled: true,
+    id: 'flux-dev',
+    organization: 'Qwen',
+    parameters: {
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      size: {
+        default: '1024x1024',
+        enum: ['512x1024', '768x512', '768x1024', '1024x576', '576x1024', '1024x1024'],
+      },
+      steps: { default: 50, max: 50, min: 1 },
+    },
+    releasedAt: '2024-08-07',
+    type: 'image',
+  },
+  {
+    description: 'FLUX.1-merged 模型结合了 "DEV" 在开发阶段探索的深度特性和 "Schnell" 所代表的高速执行优势。通过这一举措，FLUX.1-merged 不仅提升了模型的性能界限，还拓宽了其应用范围。',
+    displayName: 'FLUX.1-merged',
+    enabled: true,
+    id: 'flux-merged',
+    organization: 'Qwen',
+    parameters: {
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      size: {
+        default: '1024x1024',
+        enum: ['512x1024', '768x512', '768x1024', '1024x576', '576x1024', '1024x1024'],
+      },
+      steps: { default: 30, max: 30, min: 1 },
+    },
+    releasedAt: '2024-08-22',
+    type: 'image',
+  },
+  {
+    description: 'stable-diffusion-3.5-large 是一个具有8亿参数的多模态扩散变压器（MMDiT）文本到图像生成模型，具备卓越的图像质量和提示词匹配度，支持生成 100 万像素的高分辨率图像，且能够在普通消费级硬件上高效运行。',
+    displayName: 'StableDiffusion 3.5 Large',
+    id: 'stable-diffusion-3.5-large',
+    organization: 'Qwen',
+    parameters: {
+      height: { default: 1024, max: 1024, min: 512, step: 128 },
+      prompt: {
+        default: '',
+      },
+      steps: { default: 40, max: 500, min: 1 },
+      width: { default: 1024, max: 1024, min: 512, step: 128 },
+    },
+    releasedAt: '2024-10-25',
+    type: 'image',
+  },
+  {
+    description: 'stable-diffusion-3.5-large-turbo 是在 stable-diffusion-3.5-large 的基础上采用对抗性扩散蒸馏（ADD）技术的模型，具备更快的速度。',
+    displayName: 'StableDiffusion 3.5 Large Turbo',
+    id: 'stable-diffusion-3.5-large-turbo',
+    organization: 'Qwen',
+    parameters: {
+      height: { default: 1024, max: 1024, min: 512, step: 128 },
+      prompt: {
+        default: '',
+      },
+      steps: { default: 40, max: 500, min: 1 },
+      width: { default: 1024, max: 1024, min: 512, step: 128 },
+    },
+    releasedAt: '2024-10-25',
+    type: 'image',
+  },
+  {
+    description: 'stable-diffusion-xl 相比于 v1.5 做了重大的改进，并且与当前开源的文生图 SOTA 模型 midjourney 效果相当。具体改进之处包括： 更大的 unet backbone，是之前的 3 倍； 增加了 refinement 模块用于改善生成图片的质量；更高效的训练技巧等。',
+    displayName: 'StableDiffusion xl',
+    id: 'stable-diffusion-xl',
+    organization: 'Qwen',
+    parameters: {
+      height: { default: 1024, max: 1024, min: 512, step: 128 },
+      prompt: {
+        default: '',
+      },
+      steps: { default: 50, max: 500, min: 1 },
+      width: { default: 1024, max: 1024, min: 512, step: 128 },
+    },
+    releasedAt: '2024-04-09',
+    type: 'image',
+  },
+  {
+    description: 'stable-diffusion-v1.5 是以 stable-diffusion-v1.2 检查点的权重进行初始化，并在 "laion-aesthetics v2 5+" 上以 512x512 的分辨率进行了595k步的微调，减少了 10% 的文本条件化，以提高无分类器的引导采样。',
+    displayName: 'StableDiffusion v1.5',
+    id: 'stable-diffusion-v1.5',
+    organization: 'Qwen',
+    parameters: {
+      height: { default: 512, max: 1024, min: 512, step: 128 },
+      prompt: {
+        default: '',
+      },
+      steps: { default: 50, max: 500, min: 1 },
+      width: { default: 512, max: 1024, min: 512, step: 128 },
+    },
+    releasedAt: '2024-04-09',
+    type: 'image',
+  },
 ];
 
 export const allModels = [...qwenChatModels, ...qwenImageModels];

package/src/libs/model-runtime/qwen/createImage.ts

@@ -18,20 +18,38 @@ interface QwenImageTaskResponse {
   request_id: string;
 }
 
+const QwenText2ImageModels = [
+  'wan2.2-t2i',
+  'wanx2.1-t2i',
+  'wanx2.0-t2i',
+  'wanx-v1',
+  'flux',
+  'stable-diffusion'
+];
+
+const getModelType = (model: string): string => {
+  // 可以添加其他模型类型的判断
+  // if (QwenImage2ImageModels.some(prefix => model.startsWith(prefix))) {
+  //   return 'image2image';
+  // }
+
+  if (QwenText2ImageModels.some(prefix => model.startsWith(prefix))) {
+    return 'text2image';
+  }
+
+  throw new Error(`Unsupported model: ${model}`);
+}
+
 /**
  * Create an image generation task with Qwen API
  */
 async function createImageTask(payload: CreateImagePayload, apiKey: string): Promise<string> {
   const { model, params } = payload;
   // I can only say that the design of Alibaba Cloud's API is really bad; each model has a different endpoint path.
-  const modelEndpointMap: Record<string, string> = {
-    'wanx2.1-t2i-turbo':
-      'https://dashscope.aliyuncs.com/api/v1/services/aigc/text2image/image-synthesis',
-  };
-
-  const endpoint = modelEndpointMap[model];
+  const modelType = getModelType(model);
+  const endpoint = `https://dashscope.aliyuncs.com/api/v1/services/aigc/${modelType}/image-synthesis`
   if (!endpoint) {
-    throw new Error(`Unsupported model: ${model}`);
+    throw new Error(`No endpoint configured for model type: ${modelType}`);
   }
   log('Creating image task with model: %s, endpoint: %s', model, endpoint);
 
@@ -45,10 +63,12 @@ async function createImageTask(payload: CreateImagePayload, apiKey: string): Pro
       model,
       parameters: {
         n: 1,
-        ...(params.seed !== undefined ? { seed: params.seed } : {}),
+        ...(typeof params.seed === 'number' ? { seed: params.seed } : {}),
         ...(params.width && params.height
           ? { size: `${params.width}*${params.height}` }
-          : { size: '1024*1024' }),
+          : params.size 
+            ? { size: params.size.replaceAll('x', '*') }
+            : { size: '1024*1024' }),
       },
     }),
     headers: {