@lobehub/chat 1.105.0 → 1.105.2

package/src/config/aiModels/siliconcloud.ts

@@ -1,4 +1,4 @@
-import { AIChatModelCard } from '@/types/aiModel';
+import { AIChatModelCard, AIImageModelCard } from '@/types/aiModel';
 
 // https://siliconflow.cn/zh-cn/models
 const siliconcloudChatModels: AIChatModelCard[] = [
@@ -830,6 +830,28 @@ const siliconcloudChatModels: AIChatModelCard[] = [
   },
 ];
 
-export const allModels = [...siliconcloudChatModels];
+const siliconcloudImageModels: AIImageModelCard[] = [
+  {
+    description:
+      'Kolors 是由快手 Kolors 团队开发的基于潜在扩散的大规模文本到图像生成模型。该模型通过数十亿文本-图像对的训练，在视觉质量、复杂语义准确性以及中英文字符渲染方面展现出显著优势。它不仅支持中英文输入，在理解和生成中文特定内容方面也表现出色',
+    displayName: 'Kolors',
+    enabled: true,
+    id: 'Kwai-Kolors/Kolors',
+    parameters: {
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      size: {
+        default: '1024x1024',
+        enum: ['1024x1024', '960x1280', '768x1024', '720x1440', '720x1280'],
+      },
+    },
+    releasedAt: '2024-07-06',
+    type: 'image',
+  },
+];
+
+export const allModels = [...siliconcloudChatModels, ...siliconcloudImageModels];
 
 export default allModels;

package/src/config/aiModels/stepfun.ts

@@ -1,4 +1,4 @@
-import { AIChatModelCard } from '@/types/aiModel';
+import { AIChatModelCard, AIImageModelCard } from '@/types/aiModel';
 
 // https://platform.stepfun.com/docs/pricing/details
 
@@ -275,6 +275,71 @@ const stepfunChatModels: AIChatModelCard[] = [
   },
 ];
 
-export const allModels = [...stepfunChatModels];
+const stepfunImageModels: AIImageModelCard[] = [
+  // https://platform.stepfun.com/docs/llm/image
+  {
+    description:
+      '阶跃星辰新一代生图模型,该模型专注于图像生成任务,能够根据用户提供的文本描述,生成高质量的图像。新模型生成图片质感更真实，中英文文字生成能力更强。',
+    displayName: 'Step 2X Large',
+    enabled: true,
+    id: 'step-2x-large',
+    parameters: {
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      size: {
+        default: '1024x1024',
+        enum: ['256x256', '512x512', '768x768', '1024x1024', '1280x800', '800x1280'],
+      },
+      steps: { default: 50, max: 100, min: 1 },
+    },
+    releasedAt: '2024-08-07',
+    type: 'image',
+  },
+  {
+    description:
+      '该模型拥有强大的图像生成能力，支持文本描述作为输入方式。具备原生的中文支持，能够更好的理解和处理中文文本描述，并且能够更准确地捕捉文本描述中的语义信息，并将其转化为图像特征，从而实现更精准的图像生成。模型能够根据输入生成高分辨率、高质量的图像，并具备一定的风格迁移能力。',
+    displayName: 'Step 1X Medium',
+    enabled: true,
+    id: 'step-1x-medium',
+    parameters: {
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      size: {
+        default: '1024x1024',
+        enum: ['256x256', '512x512', '768x768', '1024x1024', '1280x800', '800x1280'],
+      },
+      steps: { default: 50, max: 100, min: 1 },
+    },
+    releasedAt: '2025-07-15',
+    type: 'image',
+  },
+  {
+    description:
+      '该模型专注于图像编辑任务，能够根据用户提供的图片和文本描述，对图片进行修改和增强。支持多种输入格式，包括文本描述和示例图像。模型能够理解用户的意图，并生成符合要求的图像编辑结果。',
+    displayName: 'Step 1X Edit',
+    enabled: true,
+    id: 'step-1x-edit',
+    parameters: {
+      imageUrl: { default: null },
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      size: {
+        default: '1024x1024',
+        enum: ['512x512', '768x768', '1024x1024'],
+      },
+      steps: { default: 28, max: 100, min: 1 },
+    },
+    releasedAt: '2025-03-04',
+    type: 'image',
+  },
+];
+
+export const allModels = [...stepfunChatModels, ...stepfunImageModels];
 
 export default allModels;

package/src/config/aiModels/volcengine.ts

@@ -1,4 +1,4 @@
-import { AIChatModelCard } from '@/types/aiModel';
+import { AIChatModelCard, AIImageModelCard } from '@/types/aiModel';
 
 // modelInfo https://www.volcengine.com/docs/82379/1330310
 // pricing https://console.volcengine.com/ark/region:ark+cn-beijing/openManagement
@@ -509,6 +509,60 @@ const doubaoChatModels: AIChatModelCard[] = [
   },
 ];
 
-export const allModels = [...doubaoChatModels];
+const volcengineImageModels: AIImageModelCard[] = [
+  {
+    /*
+    // TODO: AIImageModelCard 不支持 config.deploymentName
+    config: {
+      deploymentName: 'doubao-seedream-3-0-t2i-250415',
+    },
+    */
+    description:
+      'Doubao图片生成模型由字节跳动 Seed 团队研发，支持文字与图片输入，提供高可控、高质量的图片生成体验。基于文本提示词生成图片。',
+    displayName: 'Doubao Seedream 3.0 t2i',
+    enabled: true,
+    id: 'doubao-seedream-3-0-t2i-250415',
+    parameters: {
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      size: {
+        default: '1024x1024',
+        enum: ['1024x1024', '864x1152', '1152x864', '1280x720', '720x1280', '832x1248', '1248x832', '1512x648'],
+      },
+    },
+    releasedAt: '2025-04-15',
+    type: 'image',
+  },
+  /*
+  // Note: Doubao 图生图模型与文生图模型公用一个 Endpoint，当前如果存在 imageUrl 会切换至 edit endpoint 下
+  {
+    config: {
+      deploymentName: 'doubao-seededit-3-0-i2i-250628',
+    },
+    description:
+      'Doubao图片生成模型由字节跳动 Seed 团队研发，支持文字与图片输入，提供高可控、高质量的图片生成体验。支持通过文本指令编辑图像，生成图像的边长在512～1536之间。',
+    displayName: 'Doubao SeedEdit 3.0 i2i',
+    enabled: true,
+    id: 'doubao-seededit-3-0-i2i-250628',
+    parameters: {
+      imageUrl: { default: null },
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      size: {
+        default: '1024x1024',
+        enum: ['1024x1024', '864x1152', '1152x864', '1280x720', '720x1280', '832x1248', '1248x832', '1512x648'],
+      },
+    },
+    releasedAt: '2025-06-28',
+    type: 'image',
+  },
+  */
+];
+
+export const allModels = [...doubaoChatModels, ...volcengineImageModels];
 
 export default allModels;

package/src/config/aiModels/wenxin.ts

@@ -1,4 +1,4 @@
-import { AIChatModelCard } from '@/types/aiModel';
+import { AIChatModelCard, AIImageModelCard } from '@/types/aiModel';
 
 const wenxinChatModels: AIChatModelCard[] = [
   {
@@ -564,6 +564,66 @@ const wenxinChatModels: AIChatModelCard[] = [
   },
 ];
 
-export const allModels = [...wenxinChatModels];
+const wenxinImageModels: AIImageModelCard[] = [
+  {
+    description:
+      '百度自研的iRAG（image based RAG），检索增强的文生图技术，将百度搜索的亿级图片资源跟强大的基础模型能力相结合，就可以生成各种超真实的图片，整体效果远远超过文生图原生系统，去掉了AI味儿，而且成本很低。iRAG具备无幻觉、超真实、立等可取等特点。',
+    displayName: 'ERNIE iRAG',
+    enabled: true,
+    id: 'irag-1.0',
+    parameters: {
+      prompt: {
+        default: '',
+      },
+      size: {
+        default: '1024x1024',
+        enum: ['768x768', '1024x1024', '1536x1536', '2048x2048', '1024x768', '2048x1536', '768x1024', '1536x2048', '1024x576', '2048x1152', '576x1024', '1152x2048'],
+      },
+    },
+    releasedAt: '2025-02-05',
+    type: 'image',
+  },
+  {
+    description:
+      '百度自研的ERNIE iRAG Edit图像编辑模型支持基于图片进行erase（消除对象）、repaint（重绘对象）、variation（生成变体）等操作。',
+    displayName: 'ERNIE iRAG Edit',
+    enabled: true,
+    id: 'ernie-irag-edit',
+    parameters: {
+      imageUrl: { default: null },
+      prompt: {
+        default: '',
+      },
+      size: {
+        default: '1024x1024',
+        enum: ['768x768', '1024x1024', '1536x1536', '2048x2048', '1024x768', '2048x1536', '768x1024', '1536x2048', '1024x576', '2048x1152', '576x1024', '1152x2048'],
+      },
+    },
+    releasedAt: '2025-04-17',
+    type: 'image',
+  },
+  {
+    description:
+      '具有120亿参数的修正流变换器，能够根据文本描述生成图像。',
+    displayName: 'FLUX.1-schnell',
+    enabled: true,
+    id: 'flux.1-schnell',
+    parameters: {
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      size: {
+        default: '1024x1024',
+        enum: ['768x768', '1024x1024', '1536x1536', '2048x2048', '1024x768', '2048x1536', '768x1024', '1536x2048', '1024x576', '2048x1152', '576x1024', '1152x2048'],
+      },
+      steps: { default: 25, max: 50, min: 1 },
+    },
+    releasedAt: '2025-03-27',
+    type: 'image',
+  },
+];
+
+export const allModels = [...wenxinChatModels, ...wenxinImageModels];
 
 export default allModels;

package/src/config/aiModels/xai.ts

@@ -1,4 +1,4 @@
-import { AIChatModelCard } from '@/types/aiModel';
+import { AIChatModelCard, AIImageModelCard } from '@/types/aiModel';
 
 // https://docs.x.ai/docs/models
 const xaiChatModels: AIChatModelCard[] = [
@@ -158,6 +158,23 @@ const xaiChatModels: AIChatModelCard[] = [
   },
 ];
 
-export const allModels = [...xaiChatModels];
+const xaiImageModels: AIImageModelCard[] = [
+  {
+    description:
+      '我们最新的图像生成模型可以根据文本提示生成生动逼真的图像。它在营销、社交媒体和娱乐等领域的图像生成方面表现出色。',
+    displayName: 'Grok 2 Image 1212',
+    enabled: true,
+    id: 'grok-2-image-1212',
+    parameters: {
+      prompt: {
+        default: '',
+      },
+    },
+    releasedAt: '2024-12-12',
+    type: 'image',
+  },
+];
+
+export const allModels = [...xaiChatModels, ...xaiImageModels];
 
 export default allModels;

package/src/const/auth.ts

@@ -9,11 +9,10 @@ export const LOBE_CHAT_OIDC_AUTH_HEADER = 'Oidc-Auth';
 
 export const OAUTH_AUTHORIZED = 'X-oauth-authorized';
 
-export const JWT_SECRET_KEY = 'LobeHub · LobeChat';
-export const NON_HTTP_PREFIX = 'http_nosafe';
+export const SECRET_XOR_KEY = 'LobeHub · LobeHub';
 
 /* eslint-disable typescript-sort-keys/interface */
-export interface JWTPayload {
+export interface ClientSecretPayload {
   /**
    * password
    */

package/src/libs/model-runtime/ModelRuntime.test.ts

@@ -4,7 +4,7 @@ import { LangfuseGenerationClient, LangfuseTraceClient } from 'langfuse-core';
 import { beforeEach, describe, expect, it, vi } from 'vitest';
 
 import * as langfuseCfg from '@/config/langfuse';
-import { JWTPayload } from '@/const/auth';
+import { ClientSecretPayload } from '@/const/auth';
 import { TraceNameMap } from '@/const/trace';
 import { AgentRuntime, ChatStreamPayload, LobeOpenAI, ModelProvider } from '@/libs/model-runtime';
 import { providerRuntimeMap } from '@/libs/model-runtime/runtimeMap';
@@ -51,7 +51,7 @@ const specialProviders = [
 const testRuntime = (providerId: string, payload?: any) => {
   describe(`${providerId} provider runtime`, () => {
     it('should initialize correctly', async () => {
-      const jwtPayload: JWTPayload = { apiKey: 'user-key', ...payload };
+      const jwtPayload: ClientSecretPayload = { apiKey: 'user-key', ...payload };
       const runtime = await AgentRuntime.initializeWithProvider(providerId, jwtPayload);
 
       // @ts-ignore
@@ -66,7 +66,7 @@ const testRuntime = (providerId: string, payload?: any) => {
 
 let mockModelRuntime: AgentRuntime;
 beforeEach(async () => {
-  const jwtPayload: JWTPayload = { apiKey: 'user-openai-key', baseURL: 'user-endpoint' };
+  const jwtPayload: ClientSecretPayload = { apiKey: 'user-openai-key', baseURL: 'user-endpoint' };
   mockModelRuntime = await AgentRuntime.initializeWithProvider(ModelProvider.OpenAI, jwtPayload);
 });
 

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: {

package/src/libs/trpc/async/context.ts

@@ -1,14 +1,14 @@
 import debug from 'debug';
 import { NextRequest } from 'next/server';
 
-import { JWTPayload, LOBE_CHAT_AUTH_HEADER } from '@/const/auth';
+import { ClientSecretPayload, LOBE_CHAT_AUTH_HEADER } from '@/const/auth';
 import { LobeChatDatabase } from '@/database/type';
 import { KeyVaultsGateKeeper } from '@/server/modules/KeyVaultsEncrypt';
 
 const log = debug('lobe-async:context');
 
 export interface AsyncAuthContext {
-  jwtPayload: JWTPayload;
+  jwtPayload: ClientSecretPayload;
   secret: string;
   serverDB?: LobeChatDatabase;
   userId?: string | null;
@@ -19,7 +19,7 @@ export interface AsyncAuthContext {
  * This is useful for testing when we don't want to mock Next.js' request/response
  */
 export const createAsyncContextInner = async (params?: {
-  jwtPayload?: JWTPayload;
+  jwtPayload?: ClientSecretPayload;
   secret?: string;
   userId?: string | null;
 }): Promise<AsyncAuthContext> => ({

package/src/libs/trpc/edge/context.ts

@@ -1,13 +1,18 @@
 import { User } from 'next-auth';
 import { NextRequest } from 'next/server';
 
-import { JWTPayload, LOBE_CHAT_AUTH_HEADER, enableClerk, enableNextAuth } from '@/const/auth';
+import {
+  ClientSecretPayload,
+  LOBE_CHAT_AUTH_HEADER,
+  enableClerk,
+  enableNextAuth,
+} from '@/const/auth';
 import { ClerkAuth, IClerkAuth } from '@/libs/clerk-auth';
 
 export interface AuthContext {
   authorizationHeader?: string | null;
   clerkAuth?: IClerkAuth;
-  jwtPayload?: JWTPayload | null;
+  jwtPayload?: ClientSecretPayload | null;
   nextAuth?: User;
   userId?: string | null;
 }

package/src/libs/trpc/edge/middleware/jwtPayload.test.ts

@@ -5,7 +5,7 @@ import { beforeEach, describe, expect, it, vi } from 'vitest';
 import { createCallerFactory } from '@/libs/trpc/edge';
 import { AuthContext, createContextInner } from '@/libs/trpc/edge/context';
 import { edgeTrpc as trpc } from '@/libs/trpc/edge/init';
-import * as utils from '@/utils/server/jwt';
+import * as utils from '@/utils/server/xor';
 
 import { jwtPayloadChecker } from './jwtPayload';
 
@@ -40,7 +40,7 @@ describe('passwordChecker middleware', () => {
   it('should call next with jwtPayload in context if access code is correct', async () => {
     const jwtPayload = { accessCode: '123' };
 
-    vi.spyOn(utils, 'getJWTPayload').mockResolvedValue(jwtPayload);
+    vi.spyOn(utils, 'getXorPayload').mockResolvedValue(jwtPayload);
 
     ctx = await createContextInner({ authorizationHeader: 'Bearer token' });
     router = createCaller(ctx);
@@ -52,7 +52,7 @@ describe('passwordChecker middleware', () => {
 
   it('should call next with jwtPayload in context if no access codes are set', async () => {
     const jwtPayload = {};
-    vi.spyOn(utils, 'getJWTPayload').mockResolvedValue(jwtPayload);
+    vi.spyOn(utils, 'getXorPayload').mockResolvedValue(jwtPayload);
 
     ctx = await createContextInner({ authorizationHeader: 'Bearer token' });
     router = createCaller(ctx);
@@ -63,7 +63,7 @@ describe('passwordChecker middleware', () => {
   });
   it('should call next with jwtPayload in context if  access codes is undefined', async () => {
     const jwtPayload = {};
-    vi.spyOn(utils, 'getJWTPayload').mockResolvedValue(jwtPayload);
+    vi.spyOn(utils, 'getXorPayload').mockResolvedValue(jwtPayload);
 
     ctx = await createContextInner({ authorizationHeader: 'Bearer token' });
     router = createCaller(ctx);

package/src/libs/trpc/edge/middleware/jwtPayload.ts

@@ -1,6 +1,6 @@
 import { TRPCError } from '@trpc/server';
 
-import { getJWTPayload } from '@/utils/server/jwt';
+import { getXorPayload } from '@/utils/server/xor';
 
 import { edgeTrpc } from '../init';
 
@@ -9,7 +9,7 @@ export const jwtPayloadChecker = edgeTrpc.middleware(async (opts) => {
 
   if (!ctx.authorizationHeader) throw new TRPCError({ code: 'UNAUTHORIZED' });
 
-  const jwtPayload = await getJWTPayload(ctx.authorizationHeader);
+  const jwtPayload = getXorPayload(ctx.authorizationHeader);
 
   return opts.next({ ctx: { jwtPayload } });
 });

package/src/libs/trpc/lambda/context.ts

@@ -4,7 +4,7 @@ import { User } from 'next-auth';
 import { NextRequest } from 'next/server';
 
 import {
-  JWTPayload,
+  ClientSecretPayload,
   LOBE_CHAT_AUTH_HEADER,
   LOBE_CHAT_OIDC_AUTH_HEADER,
   enableClerk,
@@ -29,7 +29,7 @@ export interface OIDCAuth {
 export interface AuthContext {
   authorizationHeader?: string | null;
   clerkAuth?: IClerkAuth;
-  jwtPayload?: JWTPayload | null;
+  jwtPayload?: ClientSecretPayload | null;
   marketAccessToken?: string;
   nextAuth?: User;
   // Add OIDC authentication information

package/src/libs/trpc/lambda/middleware/keyVaults.ts

@@ -1,6 +1,6 @@
 import { TRPCError } from '@trpc/server';
 
-import { getJWTPayload } from '@/utils/server/jwt';
+import { getXorPayload } from '@/utils/server/xor';
 
 import { trpc } from '../init';
 
@@ -10,7 +10,7 @@ export const keyVaults = trpc.middleware(async (opts) => {
   if (!ctx.authorizationHeader) throw new TRPCError({ code: 'UNAUTHORIZED' });
 
   try {
-    const jwtPayload = await getJWTPayload(ctx.authorizationHeader);
+    const jwtPayload = getXorPayload(ctx.authorizationHeader);
 
     return opts.next({ ctx: { jwtPayload } });
   } catch (e) {