cerevox 3.0.0-beta.2 → 3.0.0-beta.20

package/dist/mcp/servers/prompts/rules/story-telling.md

@@ -25,13 +25,13 @@
     * 分组场景图片 `generate-image-serials` （一次性生成所有分镜图片）
     * 配音 `generate-scene-tts` （⚠️ 务必严格采用storyboard中场景的script或dialog作为配音输入文本）
     * 视频 `generate-video` 
-    * 背景音乐 `generate-music`
+    * 背景音乐 `generate-music-or-mv`
    **模式二：角色三视图生成**
     * 角色三视图 `generate-character-image` （为主要角色生成三视图）
     * 分镜图片 `generate-image` （依次生成各分镜图片，使用角色三视图作为参考）
     * 配音 `generate-scene-tts` （⚠️ 务必严格采用storyboard中场景的script或dialog作为配音输入文本）
     * 视频 `generate-video`
-    * 背景音乐 `generate-music`
+    * 背景音乐 `generate-music-or-mv`
 10. 技术规范 → 调用`get-schema（type: draft_content）`获取 draft_content 规范 → 根据规范创建 draft_content.json
 11. 执行渲染 → `compile-and-run` 输出成品并自动下载到本地
 12. 关闭项目 → `project-close`
@@ -55,7 +55,7 @@
 - **配音同步:** 确保 `generate-scene-tts` 的输入文本与 storyboard 中的 script 或 dialog 完全一致
 - **时长规范:** 视频时长必须为整秒数，配音、音效等可以精确到毫秒，如有对应配音，默认视频时长为 ceil(配音时长) 秒数
 - **内容一致性:** storyboard 文案 script 和配音内容必须严格一致，如生成配音时修改了文案，必须及时更新 storyboard
-- **音画协调:** 背景音乐 `generate-music` 必须与故事情感基调和节奏相匹配
+- **音画协调:** 背景音乐 `generate-music-or-mv` 必须与故事情感基调和节奏相匹配
 
 ## 专业技巧
 

package/dist/mcp/servers/prompts/skills/workflows/general-video.md

@@ -19,7 +19,7 @@ description: 制作通用视频时，可以根据用户需求，按照这个流
   1) 配音 `generate-scene-tts` 
   2) 图片 `generate-image` + 视频 `generate-video` 
     或 `generate-video-by-ref` 参考图生视频
-  3) 背景音乐 `generate-music`
+  3) 背景音乐 `generate-music-or-mv`
 6. 技术规范 → 调用`get-schema（type: draft_content）`获取 draft_content 规范 → 根据规范创建 draft_content.json
 7. 执行渲染 → `compile-and-run` 输出成品并自动下载到本地
 8. 关闭项目 → `project-close`
@@ -56,7 +56,7 @@ description: 制作通用视频时，可以根据用户需求，按照这个流
   * 方式二（storyboard 的当前 scene 设置 video_type: references）：
    1) `generate-scene-tts`生成配音(确定时长：接口返回数据中durationMs为配音时长)
    2) `generate-video-by-ref`参考图生视频
-5. `generate-music`生成背景音乐
+5. `generate-music-or-mv`生成背景音乐
 6. 创建`draft_content.json`：
    - ⚠️ 必须包含完整的VideoProject结构
    - 除非用户明确拒绝，否则`draft_content.json`中必须包含字幕：

package/dist/mcp/servers/prompts/skills/workflows/music-video.md

@@ -12,7 +12,7 @@ description: 创作专业音乐MV，基于 Zerocut 自主完成音乐MV成片的
 1. 确保项目已启动 → `project-open`
 2. 资料收集（可选）→ 使用搜索工具收集相关资料
 3. 音乐创作 → 根据主题构思音乐氛围 → 创作歌词 lyrics.txt
-4. 音乐生成 → 根据 lyrics.txt 调用 `generate-music` → 获得歌曲和 captions
+4. 音乐生成 → 根据 lyrics.txt 调用 `generate-music-or-mv` → 获得歌曲和 captions
 5. 分析歌曲 → 创建 timeline_analysis.json 得到 captions 的时间线
 6. 设计分镜场景 → `get-schema（type: storyboard）` 获取分镜规范 → 创建初始 storyboard.json
 7. 主要角色形象塑造 → `generate-character-image` → 生成主要角色形象参考图（三视图）

package/dist/mcp/servers/prompts/zerocut-core.md

@@ -70,11 +70,6 @@ projects/<id>/
   └─ draft_content.json  # 技术规范
 ```
 
-### materials 资源命名规范
-
-- 场景素材：`sc01_bg.png`、`sc01_motion.mp4`、`sc01_vo.mp3`
-- 通用素材：`main_bgm_60s.wav`
-
 ### output 输出规范
 
 - 画幅：提前确定横竖屏，竖屏720x1280，横屏1280x720，如无特殊要求，竖屏(720x1280)优先

package/dist/mcp/servers/zerocut.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"zerocut.d.ts","sourceRoot":"","sources":["../../../src/mcp/servers/zerocut.ts"],"names":[],"mappings":";AAm/KA,wBAAsB,GAAG,kBAKxB"}
+{"version":3,"file":"zerocut.d.ts","sourceRoot":"","sources":["../../../src/mcp/servers/zerocut.ts"],"names":[],"mappings":";AA0pLA,wBAAsB,GAAG,kBAKxB"}

package/dist/mcp/servers/zerocut.js

@@ -731,7 +731,7 @@ server.registerTool('generate-character-image', {
     description: 'Generate a turnaround image or portrait for any character.',
     inputSchema: {
         type: zod_1.z
-            .enum(['banana', 'banana-pro', 'seedream'])
+            .enum(['banana', 'banana-pro', 'seedream', 'seedream-pro'])
             .optional()
             .default('banana'),
         name: zod_1.z.string().describe('The name of the character.'),
@@ -761,7 +761,9 @@ server.registerTool('generate-character-image', {
             .boolean()
             .default(true)
             .describe('是否生成三视图。true: 生成4096x3072的三视图，false: 生成2304x4096的竖版人物正视图'),
-        saveToFileName: zod_1.z.string().describe('The filename to save.'),
+        saveToFileName: zod_1.z
+            .string()
+            .describe('The filename to save. 应该是png文件'),
     },
 }, async ({ type, name, gender, age, appearance, clothing, personality, detail_features, style, saveToFileName, referenceImage, referenceImagePrompt, isTurnaround, }) => {
     try {
@@ -937,7 +939,7 @@ server.registerTool('generate-line-sketch', {
         prompt: zod_1.z.string().describe('The prompt to generate line sketch.'),
         saveToFileName: zod_1.z
             .string()
-            .describe('The filename to save the generated line sketch.'),
+            .describe('The filename to save the generated line sketch. 应该是png文件'),
     },
 }, async ({ prompt, saveToFileName }) => {
     try {
@@ -1064,7 +1066,7 @@ server.registerTool('generate-image', {
     description: `生成图片`,
     inputSchema: {
         type: zod_1.z
-            .enum(['banana', 'banana-pro', 'seedream'])
+            .enum(['banana', 'banana-pro', 'seedream', 'seedream-pro'])
             .optional()
             .default('seedream'),
         prompt: zod_1.z
@@ -1124,17 +1126,14 @@ server.registerTool('generate-image', {
         ])
             .default('720x1280')
             .describe('The size of the image.'),
-        saveToFileName: zod_1.z.string().describe('The filename to save.'),
+        saveToFileName: zod_1.z
+            .string()
+            .describe('The filename to save. 应该是png文件'),
         watermark: zod_1.z
             .boolean()
             .optional()
             .default(false)
             .describe('Whether to add watermark to the image.'),
-        optimizePrompt: zod_1.z
-            .boolean()
-            .optional()
-            .default(false)
-            .describe('Whether to optimize the prompt.'),
         referenceImages: zod_1.z
             .array(zod_1.z.object({
             image: zod_1.z.string().describe('Local image file path'),
@@ -1163,7 +1162,7 @@ server.registerTool('generate-image', {
 \`\`\`
           `),
     },
-}, async ({ type = 'seedream', prompt, sceneIndex, storyBoardFile = 'storyboard.json', skipConsistencyCheck = false, size = '720x1280', saveToFileName, watermark, referenceImages, optimizePrompt, }) => {
+}, async ({ type = 'seedream', prompt, sceneIndex, storyBoardFile = 'storyboard.json', skipConsistencyCheck = false, size = '720x1280', saveToFileName, watermark, referenceImages, }) => {
     try {
         // 验证session状态
         const currentSession = await validateSession('generate-image');
@@ -1257,53 +1256,71 @@ server.registerTool('generate-image', {
         // 检查并替换英文单引号包裹的中文内容为中文双引号
         // 这样才能让 seedream 生成更好的中文文字
         let processedPrompt = prompt.replace(/'([^']*[\u4e00-\u9fff][^']*)'/g, '“$1”');
-        if (optimizePrompt) {
-            try {
-                const ai = currentSession.ai;
-                const promptOptimizer = await (0, promises_1.readFile)((0, node_path_1.resolve)(__dirname, './prompts/image-prompt-optimizer.md'), 'utf8');
-                const completion = await ai.getCompletions({
-                    model: 'Doubao-Seed-1.6-flash',
-                    messages: [
-                        {
-                            role: 'system',
-                            content: promptOptimizer,
+        try {
+            const ai = currentSession.ai;
+            const promptOptimizer = await (0, promises_1.readFile)((0, node_path_1.resolve)(__dirname, './prompts/image-prompt-optimizer.md'), 'utf8');
+            const schema = {
+                name: 'optimize_image_prompt',
+                schema: {
+                    type: 'object',
+                    properties: {
+                        prompt_optimized: {
+                            type: 'string',
+                            description: '优化后的提示词',
                         },
-                        {
-                            role: 'user',
-                            content: processedPrompt.trim(),
+                        metaphor_modifiers: {
+                            type: 'array',
+                            description: '从 prompt_optimized 中抽取的所有比喻修饰词（字符串数组）',
+                            items: {
+                                type: 'string',
+                                description: '比喻性修饰词，例如 “如羽毛般轻盈”、“像晨雾一样柔和”',
+                            },
                         },
-                    ],
-                });
-                let optimizedPrompt = completion.choices[0]?.message?.content.trim();
-                if (optimizedPrompt) {
-                    if (optimizedPrompt.startsWith('```json')) {
-                        // 提取 JSON 代码块中的内容
-                        const jsonMatch = optimizedPrompt.match(/```json\s*([\s\S]*?)\s*```/);
-                        if (jsonMatch && jsonMatch[1]) {
-                            optimizedPrompt = jsonMatch[1];
-                        }
-                    }
-                    if (optimizedPrompt.startsWith('{')) {
-                        try {
-                            const { prompt_optimized, metaphor_modifiers } = JSON.parse(optimizedPrompt);
-                            processedPrompt = `${prompt_optimized}`;
-                            if (metaphor_modifiers?.length) {
-                                processedPrompt += `\n\n注意：下面这些是形象比喻，并不是输出内容。\n${metaphor_modifiers}`;
-                            }
-                        }
-                        catch (ex) {
-                            processedPrompt = optimizedPrompt;
-                        }
-                    }
-                    else {
-                        processedPrompt = optimizedPrompt;
+                    },
+                    required: ['prompt_optimized', 'metaphor_modifiers'],
+                },
+            };
+            const completion = await ai.getCompletions({
+                model: 'Doubao-Seed-1.6',
+                messages: [
+                    {
+                        role: 'system',
+                        content: promptOptimizer,
+                    },
+                    {
+                        role: 'user',
+                        content: `## 用户指令
+
+${processedPrompt.trim()}
+
+## 参考图
+
+${referenceImages?.map((ref, index) => `图${index + 1}：${ref.image}`).join('\n') || '无'}`,
+                    },
+                ],
+                response_format: {
+                    type: 'json_schema',
+                    json_schema: schema,
+                },
+            });
+            const optimizedPrompt = completion.choices[0]?.message?.content.trim();
+            if (optimizedPrompt) {
+                try {
+                    const { prompt_optimized, metaphor_modifiers } = JSON.parse(optimizedPrompt);
+                    processedPrompt = `${prompt_optimized}`;
+                    if (metaphor_modifiers?.length) {
+                        processedPrompt += `\n\n注意：下面这些是形象比喻，并不是输出内容。\n${metaphor_modifiers}`;
                     }
                 }
-            }
-            catch (error) {
-                console.error('Failed to optimize prompt:', error);
+                catch (ex) {
+                    console.error('Failed to parse optimized prompt:', ex);
+                    processedPrompt = optimizedPrompt;
+                }
             }
         }
+        catch (error) {
+            console.error('Failed to optimize prompt:', error);
+        }
         console.log(`Generating image with prompt: ${processedPrompt.substring(0, 100)}...`);
         // 处理参考图片
         let imageBase64Array;
@@ -1432,12 +1449,14 @@ server.registerTool('edit-image', {
     inputSchema: {
         prompt: zod_1.z.string().describe('要编辑图片的中文提示词'),
         type: zod_1.z
-            .enum(['banana-pro', 'banana', 'seedream'])
+            .enum(['banana-pro', 'banana', 'seedream', 'seedream-pro'])
             .optional()
             .default('seedream')
             .describe('The type of image model to use.'),
         sourceImageFileName: zod_1.z.string().describe('The source image file name.'),
-        saveToFileName: zod_1.z.string().describe('The filename to save.'),
+        saveToFileName: zod_1.z
+            .string()
+            .describe('The filename to save. 应该是png文件'),
         size: zod_1.z
             .enum([
             '1024x1024',
@@ -1592,6 +1611,8 @@ server.registerTool('generate-video', {
             'hailuo-fast',
             'vidu',
             'vidu-pro',
+            'vidu-uc',
+            'vidu-uc-pro',
             'kling',
             'kling-pro',
             'pixv',
@@ -1601,7 +1622,9 @@ server.registerTool('generate-video', {
         ])
             .default('lite')
             .describe('除非用户明确提出使用其他模型，否则一律用lite模型；zero 系列模型适合创作8-23秒带故事情节的短片'),
-        saveToFileName: zod_1.z.string().describe('The filename to save.'),
+        saveToFileName: zod_1.z
+            .string()
+            .describe('The filename to save. 应该是mp4文件'),
         start_frame: zod_1.z
             .string()
             .optional()
@@ -2008,7 +2031,7 @@ server.registerTool('generate-video', {
                     console.warn('Failed to send progress update:', progressError);
                 }
             },
-            waitForFinish: true,
+            waitForFinish: type !== 'zero',
         });
         if (!res) {
             throw new Error('Failed to generate video: no response from AI service');
@@ -2056,7 +2079,7 @@ server.registerTool('generate-video', {
                         type: 'text',
                         text: JSON.stringify({
                             success: true,
-                            message: '该视频生成任务正在运行中，它是异步任务，且执行时间较长，你应立即调用工具 wait-for-task-finish 来等待任务结束，如该工具调用超时，你应立即再次重新调用直到任务结束。',
+                            message: '该视频生成任务正在运行中，它是异步任务，且执行时间较长，你应立即调用工具 wait-for-task-finish 来等待任务结束，如 wait-for-task-finish 工具调用超时，你应立即再次重新调用直到任务结束。',
                             taskUrl: res.taskUrl,
                         }),
                     },
@@ -2094,7 +2117,7 @@ server.registerTool('wait-for-task-finish', {
             .describe('The taskUrl of the video task to wait for.'),
         saveToFileName: zod_1.z
             .string()
-            .describe('The file name to save the video to.'),
+            .describe('The file name to save the video to. 应该是mp4文件'),
     },
 }, async ({ taskUrl, saveToFileName }, context) => {
     try {
@@ -2169,7 +2192,7 @@ server.registerTool('generate-sound-effect', {
             .describe('The duration of the sound which will be generated in seconds. Must be at least 0.5 and at most 30. If set to None we will guess the optimal duration using the prompt. Defaults to None.'),
         saveToFileName: zod_1.z
             .string()
-            .describe('The filename to save. The mime type is audio/mpeg (mp3).'),
+            .describe('The filename to save. 应该是mp3文件'),
     },
 }, async ({ prompt_in_english, loop, saveToFileName, duration_seconds }) => {
     try {
@@ -2216,14 +2239,32 @@ server.registerTool('generate-sound-effect', {
         return createErrorResponse(error, 'generate-sound-effect');
     }
 });
-server.registerTool('generate-music', {
-    title: 'Generate Music',
-    description: 'Generate the music. Include background music or song.',
+server.registerTool('generate-music-or-mv', {
+    title: '创作音乐（Music）或音乐视频（Music Video）',
+    description: '生成音乐，包括MV（music video）、BGM 或 歌曲',
     inputSchema: {
         prompt: zod_1.z.string().describe('The prompt to generate.'),
+        singerPhoto: zod_1.z
+            .string()
+            .optional()
+            .describe('The singer photo to use. 只有type为music_video的时候才生效，也可以不传，模型会自动生成'),
+        mvOrientation: zod_1.z
+            .enum(['portrait', 'landscape'])
+            .optional()
+            .describe('The orientation of the music video. Defaults to portrait.')
+            .default('portrait'),
+        mvOriginalSong: zod_1.z
+            .string()
+            .optional()
+            .describe('用于生成mv的音乐. 只有type为music_video的时候才生效，也可以不传，模型会自动创作'),
+        mvGenSubtitles: zod_1.z
+            .boolean()
+            .optional()
+            .default(false)
+            .describe('是否生成mv的字幕. 默认为false，只有type为music_video的时候才生效'),
         type: zod_1.z
-            .enum(['bgm', 'song'])
-            .describe('The type of music. Defaults to background music.')
+            .enum(['bgm', 'song', 'music_video'])
+            .describe('The type of music. Defaults to BGM. ⚠️ 如果 type 是 music_video，会直接生成音频和视频，**不需要**额外专门生成歌曲')
             .default('bgm'),
         model: zod_1.z
             .enum(['doubao', 'minimax'])
@@ -2239,9 +2280,11 @@ server.registerTool('generate-music', {
             .boolean()
             .default(false)
             .describe('Whether to skip copyright check.'),
-        saveToFileName: zod_1.z.string().describe('The filename to save.'),
+        saveToFileName: zod_1.z
+            .string()
+            .describe('The filename to save. 如果type是music video，应该是mp4文件，否则应该是mp3文件'),
     },
-}, async ({ prompt, type, model, duration, skipCopyCheck, saveToFileName }, context) => {
+}, async ({ prompt, singerPhoto, mvOrientation, mvOriginalSong, mvGenSubtitles, type, model, duration, skipCopyCheck, saveToFileName, }, context) => {
     try {
         // 验证session状态
         const currentSession = await validateSession('generate-music');
@@ -2252,24 +2295,54 @@ server.registerTool('generate-music', {
         if (type === 'bgm' && duration > 120) {
             throw new Error('BGM duration must be at most 120 seconds.');
         }
-        const finalPrompt = `${prompt.trim()} ${type === 'bgm' ? `纯音乐无歌词，时长${duration}秒` : `时长${duration}秒，使用${model}模型`}`;
-        const res = await ai.generateMusic({
-            prompt: finalPrompt,
-            skipCopyCheck,
-            onProgress: async (metaData) => {
-                try {
-                    await sendProgress(context, metaData.Result?.Progress ?? ++progress, metaData.Result?.Progress ? 100 : undefined, JSON.stringify(metaData));
-                }
-                catch (progressError) {
-                    console.warn('Failed to send progress update:', progressError);
-                }
-            },
-        });
+        let res;
+        if (type === 'music_video') {
+            const singer_photo = singerPhoto
+                ? await getMaterialUri(currentSession, singerPhoto)
+                : undefined;
+            const original_song = mvOriginalSong
+                ? await getMaterialUri(currentSession, mvOriginalSong)
+                : undefined;
+            res = await ai.generateZeroCutMusicVideo({
+                // prompt: `${prompt.trim()} 音乐时长${duration}秒`,
+                prompt,
+                singerPhoto: singer_photo,
+                orientation: mvOrientation,
+                genSubtitles: mvGenSubtitles,
+                originalSong: original_song,
+                duration,
+                resolution: '720p',
+                onProgress: async (metaData) => {
+                    try {
+                        await sendProgress(context, metaData.Result?.Progress ?? ++progress, metaData.Result?.Progress ? 100 : undefined, JSON.stringify(metaData));
+                    }
+                    catch (progressError) {
+                        console.warn('Failed to send progress update:', progressError);
+                    }
+                },
+                waitForFinish: false,
+            });
+        }
+        else {
+            const finalPrompt = `${prompt.trim()} ${type === 'bgm' ? `纯音乐无歌词，时长${duration}秒` : `时长${duration}秒，使用${model}模型`}`;
+            res = await ai.generateMusic({
+                prompt: finalPrompt,
+                skipCopyCheck,
+                onProgress: async (metaData) => {
+                    try {
+                        await sendProgress(context, metaData.Result?.Progress ?? ++progress, metaData.Result?.Progress ? 100 : undefined, JSON.stringify(metaData));
+                    }
+                    catch (progressError) {
+                        console.warn('Failed to send progress update:', progressError);
+                    }
+                },
+            });
+        }
         if (!res) {
-            throw new Error('Failed to generate BGM: no response from AI service');
+            throw new Error('Failed to generate Music: no response from AI service');
         }
         if (res.url) {
-            console.log('BGM generated successfully, saving to materials...');
+            console.log('Music generated successfully, saving to materials...');
             const uri = await saveMaterial(currentSession, res.url, validatedFileName);
             const { url, duration: bgmDuration, captions, ...opts } = res;
             // 保存captions到本地
@@ -2307,15 +2380,29 @@ server.registerTool('generate-music', {
                 ],
             };
         }
+        else if (res.taskUrl) {
+            return {
+                content: [
+                    {
+                        type: 'text',
+                        text: JSON.stringify({
+                            success: true,
+                            message: '该视频生成任务正在运行中，它是异步任务，且执行时间较长，你应立即调用工具 wait-for-task-finish 来等待任务结束，如如 wait-for-task-finish 工具调用超时，你应立即再次重新调用直到任务结束。',
+                            taskUrl: res.taskUrl,
+                        }),
+                    },
+                ],
+            };
+        }
         else {
-            console.warn('BGM generation completed but no URL returned');
+            console.warn('Music generation completed but no URL returned');
             return {
                 content: [
                     {
                         type: 'text',
                         text: JSON.stringify({
                             success: false,
-                            error: 'No BGM URL returned from AI service',
+                            error: 'No Music URL returned from AI service',
                             response: res,
                             timestamp: new Date().toISOString(),
                         }),
@@ -2325,7 +2412,7 @@ server.registerTool('generate-music', {
         }
     }
     catch (error) {
-        return createErrorResponse(error, 'generate-bgm');
+        return createErrorResponse(error, 'generate-music');
     }
 });
 server.registerTool('generate-scene-tts', {
@@ -2352,7 +2439,9 @@ server.registerTool('generate-scene-tts', {
             .string()
             .optional()
             .describe('跳过校验的理由，如果skipConsistencyCheck设为true，必须要传这个参数'),
-        saveToFileName: zod_1.z.string().describe('The filename to save.'),
+        saveToFileName: zod_1.z
+            .string()
+            .describe('The filename to save. 应该是mp3文件'),
         speed: zod_1.z
             .number()
             .min(0.5)
@@ -2923,7 +3012,7 @@ server.registerTool('voice-design', {
         previewText: zod_1.z.string().describe('The preview text to design the voice.'),
         saveToFileName: zod_1.z
             .string()
-            .describe('The file name to save the designed voice.'),
+            .describe('The file name to save the designed voice. 应该是mp3文件'),
     },
 }, async ({ prompt, previewText, saveToFileName }) => {
     try {
@@ -3399,7 +3488,7 @@ server.registerTool('audio-video-sync', {
             .describe('The reference photo face for lip sync.'),
         saveToFileName: zod_1.z
             .string()
-            .describe('The filename to save the audio-video-synced video.'),
+            .describe('The filename to save the audio-video-synced video. 应该是mp4文件'),
     },
 }, async ({ lipSync, lipSyncType, lipSyncPadAudio, videoFileName, audioFileName, audioInMs, refPhotoFileName, saveToFileName, }, context) => {
     try {
@@ -3577,6 +3666,7 @@ server.registerTool('generate-video-by-ref', {
             'veo3.1',
             'veo3.1-pro',
             'vidu',
+            'vidu-uc',
             'pixv',
         ])
             .default('lite')
@@ -3588,7 +3678,7 @@ server.registerTool('generate-video-by-ref', {
             .describe('Whether to mute the video (effective for sora2 and veo3.1).'),
         saveToFileName: zod_1.z
             .string()
-            .describe('The filename to save the generated video.'),
+            .describe('The filename to save the generated video. 应该是mp4文件'),
         sceneIndex: zod_1.z
             .number()
             .min(1)
@@ -3794,7 +3884,54 @@ server.registerTool('generate-video-by-ref', {
         if (promptPrefix) {
             promptPrefix += '\n';
         }
-        const finalPrompt = `${promptPrefix}${prompt}`;
+        let finalPrompt = `${promptPrefix}${prompt}`;
+        if (type === 'pixv') {
+            const completion = await ai.getCompletions({
+                model: 'Doubao-Seed-1.6',
+                messages: [
+                    {
+                        role: 'system',
+                        content: `你根据主体信息，优化用户指令，使描述中的内容正确引用主体名称。
+
+具体方式为，将用户指令中引用主体信息中主体名称的部分，用 “@主体名” 的形式替代，注意它和前后内容之间也需要用**空格**分隔。
+           
+## 例子
+
+### 输入：
+
+主体信息
+[
+  {"type": "subject", "fileName": "dog.png", "ref_name": "狗"},
+  {"type": "background", "fileName": "room.png", "ref_name": "房间"}
+]
+
+用户指令
+一只狗在房间里玩耍
+
+### 输出：
+一只 @狗 在 @房间 里玩耍
+
+---
+
+## 要求与约束
+
+只输出替换主体名后的用户指令，不要输出其他任何额外内容
+`,
+                    },
+                    {
+                        role: 'user',
+                        content: `## 主体信息
+
+${JSON.stringify(referenceImages)}
+
+## 用户指令
+              
+${prompt.trim()}`,
+                    },
+                ],
+            });
+            finalPrompt = completion.choices[0]?.message?.content.trim();
+        }
         // 调用 referencesToVideo 函数
         const result = await currentSession.ai.referencesToVideo({
             prompt: finalPrompt,
@@ -3823,7 +3960,7 @@ server.registerTool('generate-video-by-ref', {
                         type: 'text',
                         text: JSON.stringify({
                             success: true,
-                            message: '该视频生成任务正在运行中，它是异步任务，且执行时间较长，你应立即调用工具 wait-for-task-finish 来等待任务结束，如该工具调用超时，你应立即再次重新调用直到任务结束。',
+                            message: '该视频生成任务正在运行中，它是异步任务，且执行时间较长，你应立即调用工具 wait-for-task-finish 来等待任务结束，如如 wait-for-task-finish 工具调用超时，你应立即再次重新调用直到任务结束。',
                             taskUrl: result.taskUrl,
                         }),
                     },
@@ -3898,7 +4035,7 @@ server.registerTool('extend-video-duration', {
             .describe('Optional end frame image file name in materials directory to guide the video extension.'),
         saveToFileName: zod_1.z
             .string()
-            .describe('The filename to save the extended video.'),
+            .describe('The filename to save the extended video. 应该是mp4文件'),
     },
 }, async ({ videoFileName, duration, resolution, prompt, type = 'turbo', endFrame, saveToFileName, }, context) => {
     try {
@@ -3994,11 +4131,11 @@ server.registerTool('use-template', {
             .describe('Optional materials to use in the template.'),
         saveToFileName: zod_1.z
             .string()
-            .describe('The filename to save the generated material.'),
+            .describe('The filename to save the generated material. 根据用户具体需求，应该是mp4或png文件'),
     },
 }, async ({ user_request, saveToFileName, materials }) => {
     try {
-        const currentSession = await validateSession('generate-video-by-template');
+        const currentSession = await validateSession('use-template');
         const ai = currentSession.ai;
         const data = await ai.listTemplates('all');
         const templates = data.map(item => ({
@@ -4013,7 +4150,7 @@ server.registerTool('use-template', {
             messages: [
                 {
                     role: 'system',
-                    content: `你根据用户需求，从以下模板中选择一个匹配的模板，返回模板ID：\n\n${JSON.stringify(templates)}\n\n**约束**：只输出模板ID，不需要其他解释，如果没有匹配的模版，输出"无匹配模版"`,
+                    content: `你根据用户需求，分析需求与模板描述（description）和触发器（trigger）的匹配程度，从以下模板中选择一个匹配的模板，返回模板ID：\n\n${JSON.stringify(templates)}\n\n**约束**：只输出模板ID，不需要其他解释，如果没有匹配的模版，输出"无匹配模版"`,
                 },
                 {
                     role: 'user',