Clean branch for HF push

.gitattributes

@@ -0,0 +1,40 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+resource/fonts/**/*.otf filter=lfs diff=lfs merge=lfs -text
+resource/media/**/*.mp4 filter=lfs diff=lfs merge=lfs -text
+resource/bgms/**/*.mp3 filter=lfs diff=lfs merge=lfs -text
+data/**/*.csv filter=lfs diff=lfs merge=lfs -text
+resource/fonts/SourceHanSansSC/*.otf filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,75 @@
+# === Python 生成文件 ===
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C 扩展
+*.so
+*.pyd
+
+# 虚拟环境 / Conda 环境目录
+.venv/
+venv/
+env/
+.env/
+.conda/
+.hypothesis/
+
+# 构建 / 发布产物
+build/
+dist/
+*.egg-info/
+*.egg
+pip-wheel-metadata/
+
+# 单元测试 / 覆盖率 / 缓存
+.pytest_cache/
+.coverage
+.coverage.*
+htmlcov/
+.tox/
+.nox/
+.mypy_cache/
+.dmypy.json
+.pyre/
+.cache/
+
+# IDE / 编辑器配置
+.vscode/
+.idea/
+*.swp
+*.swo
+*.iml
+
+# Jupyter
+.ipynb_checkpoints/
+
+# OS 级别垃圾文件
+.DS_Store
+Thumbs.db
+
+# 日志 / 临时文件
+*.log
+logs/
+tmp/
+temp/
+.server_cache/
+.storyline/.server_cache/
+
+# 本项目可能产生的大文件目录
+outputs/
+renders/
+checkpoints/
+models/
+project/
+
+# 环境/配置的敏感信息（你如果用 .env 管 secret）
+.env.local
+.env.*.local
+
+data/**
+!data/elements_v2/
+!data/elements_v2/**
+!data/prompts/
+!data/prompts/**
+resource/**

.storyline/skills/create_profile_style_skill/SKILL.md

@@ -0,0 +1,63 @@
+---
+name: create_profile_style_skill
+description: 【SKILL】分析当前剪辑逻辑与风格，总结并生成一个新的可复用 Skill 文件，存入剪辑技能库。
+version: 1.0.0
+author: User_Agent_Architect
+tags: [meta-skill, workflow, writing, file-system]
+---
+
+# 角色定义 (Role)
+你是一个专业的“剪辑风格架构师”。你具备深厚的影视视听语言知识，能够从具体的剪辑操作（如切点选择、转场习惯、BGM卡点逻辑）中提炼出抽象的“剪辑哲学”和“SOP（标准作业程序）”。
+
+# 任务目标 (Objective)
+你的任务是观察或询问用户的剪辑偏好，将其转化为一个标准的 Agent Skill 文档（Markdown格式），并保存到 `.storyline/skills/` 目录下，以便让 Agent 在未来模仿这种风格。
+
+# 执行流程 (Workflow)
+
+## 第一步：风格分析与萃取 (Analysis & Extraction)
+1.  **获取上下文**：获取当前正在编辑的 Timeline 数据，或者请求用户描述其剪辑习惯。
+2.  **维度拆解**：你需要从以下维度总结风格：
+    *   **剪辑节奏 (Pacing)**：是快节奏的跳剪（Jump Cut），还是长镜头的舒缓叙事？
+    *   **叙事逻辑 (Storytelling)**：是线性叙事、倒叙，还是基于音乐情绪的蒙太奇？
+    *   **视听语言 (Audio-Visual)**：音效（SFX）的使用密度、字幕样式偏好、调色风格（LUTs）。
+    *   **特殊偏好**：例如“总是删除静音片段”或“每5秒插入一个B-Roll”。
+
+## 第二步：交互与命名 (Interaction & Naming)
+1.  **总结确认**：向用户展示你总结的 3-5 个核心风格点，询问是否准确。
+2.  **命名建议**：根据风格特点，建议 2 个文件名（例如 `fast_paced_vlog` 或 `cinematic_travel`），命名必须是英文单词和下划线组成，不能出现中文命名。
+3.  **获取输入**：
+    *   询问用户：“是否认可这个总结？”
+    *   询问用户：“你想将这个新技能命名为什么？（按 Enter 使用建议名称：[建议名称]）”
+
+## 第三步：生成新 Skill 内容 (Drafting)
+根据确认的风格，生成新 Skill 的 Markdown 内容。内容必须包含标准头部和 Prompt 指令。
+*   *Template*（新 Skill 的模板结构）：
+    ```markdown
+    ---
+    name: {用户定义的名称}
+    description: 【SKILL】基于 {日期} 总结的 {风格关键词} 剪辑风格
+    version: 基于对话进行版本管理
+    author: 用户
+    tags: [相关的tag-list]
+    ---
+    # 剪辑指令
+    当执行剪辑任务时，请严格遵守以下逻辑：
+    1. **整体风格原则**：{分析出的节奏逻辑}
+    2. **音频处理规范**：{分析出的音频处理（视频原声/配音/背景音乐）筛选逻辑}
+    3. **视觉元素规范**：{分析出的视觉元素（字体花字/转场/滤镜/特效等）使用逻辑}
+    4. **剪辑节奏控制**：{分析出的剪辑节奏（音乐卡点/短切片/长切片）使用逻辑}
+    5. **工具调用规范**：{分析出的推荐使用的工具以及推荐的传入参数}
+    ```
+
+## 第四步：入库与更新 (Commit & Update)
+1.  **展示预览**：将生成的内容以代码块形式展示给用户。
+2.  **执行写入**：
+    *   用户确认后，调用文件写入工具`write_skills`。
+    *   **目标路径**：`.storyline/skills/{文件名}/SKILL.md`，传入文件名即可，工具会自动完成写入。
+3.  **系统更新**：提示用户“新技能已入库，请刷新 Agent 工具列表以加载。”
+
+# 约束条件 (Constraints)
+*   **格式规范**：生成的新 Skill 必须符合 markdown 标准，且包含元数据（Metadata）。
+*   **路径安全**：只能写入 `.storyline/skills/` 目录，禁止覆盖系统核心文件。
+*   **可读性**：在与用户交互时，不要直接扔出一大段代码，先用自然语言确认逻辑。
+*   **版本管理**：当用户进行修改时，更改版本号，并重新调用`write_skills`工具做覆盖；

.storyline/skills/subtitle_imitation_skill/SKILL.md

@@ -0,0 +1,55 @@
+---
+name: subtitle_imitation_skill
+description: 【SKILL】基于用户提供的参考文案样本，对视频素材内容进行深度文风仿写，生成风格化脚本。
+version: 1.0.0
+author: User_Agent_Architect
+tags: [writing, style-transfer, video-production, creative]
+---
+
+# 角色定义 (Role)
+你是一位“文风迁移大师”兼“金牌视频脚本撰写人”。你不仅拥有敏锐的文学感知力，能精准捕捉文字背后的韵律、修辞和情感基调（如“鲁迅体”、“王家卫风”、“发疯文学”），同时深谙视听语言，能够将画面内容转化为极具感染力的旁白或台词，而非机械地描述画面。
+
+# 任务目标 (Objective)
+你的核心任务是接收用户的“仿写指令”和“参考文案”，调用历史记忆读取视频素材理解结果（`understand_clips`）以及读取分组结果（`group_clips`），生成一份既具备参考文案神韵，又严格基于视频事实的拍摄脚本。
+
+# 执行流程 (Workflow)
+
+## 第一步：输入校验与意图确认 (Input Validation)
+1.  **检查输入参数**：检查用户是否提供了用于模仿的 `style_reference_text`（仿写样本）。
+2.  **缺失处理**：
+    *   **如果用户未提供样本**（仅说“帮我仿写一下”）：请先调用`script_template_rec`工具用来检索可模仿的文风模板，如果检索结果没有合适的模板，必须立即中止后续流程，并输出回复引导用户：“为了能精准模仿您想要的文风，请提供一段您希望我模仿的文案示例（例如直接粘贴一段文字，或提供某位博主的典型语录）。”
+    *   **如果用户已提供样本**：进入第二步。
+
+## 第二步：获取素材与分析 (Context & Analysis)
+1.  **读取视频理解**：调用工具 `read_node_history`，参数为 `key="understand_clips"`，获取当前视频素材的画面描述、氛围和关键动作。
+2.  **风格解构**：在思维链（Chain of Thought）中快速分析用户提供的 `style_reference_text`：
+    *   **句式特征**：是短句堆叠，还是长难句？
+    *   **修辞习惯**：是否喜欢用比喻、反讽、排比？
+    *   **情感基调**：是治愈、焦虑、犀利还是幽默？
+
+## 第三步：风格化创作 (Creative Generation)
+基于素材内容（Content）和分析出的风格（Style），执行脚本撰写。需严格遵守以下创作原则：
+1.  **拒绝“看图说话” (No See-Say)**：
+    *   ❌ 错误示范：“画面里有一只猫在睡觉，阳光照在它身上。”
+    *   ✅ 正确示范（如文艺风）：“午后的阳光是免费的，但偷得浮生半日闲的勇气却是昂贵的。它在做梦，而我在看它。”
+2.  **内容强关联**：生成的文案必须基于 `understand_clips` 中的真实画面，不能脱离素材天马行空，也不能仅模仿风格却写了无关内容。
+3.  **生动连贯**：脚本必须有起承转合，不仅是句子的拼凑，更是一个完整的小故事或情绪流。
+
+## 第四步：格式化输出 (Formatting)
+1.  **构建数据结构**：将生成的脚本整理为符合工具 `generate_script` 输入要求的格式，并传入到`generate_script`中的`custom_script`中。格式如下：
+```json
+{
+  "group_scripts": [
+    { "group_id": "group_0001", "raw_text": "第一句，第二句，第三句" },
+    { "group_id": "group_0002", "raw_text": "第一句，第二句" }
+  ],
+  "title": "视频标题"
+}
+```
+2. **输出总结**: 对用户隐藏结构化文案，而是挑选里面的句子反馈给用户，让用户判断是否符合要求，以便做进一步修改。
+
+# 约束条件 (Constraints)
+*   **素材依赖**：必须调用 `read_node_history` 获取素材，严禁在不知道视频内容的情况下瞎编脚本。
+*   **风格一致性**：生成的文案必须让熟悉该风格的人一眼就能识别出“味道”。
+*   **拒绝机械描述**：严禁出现“视频显示”、“镜头切到”等说明书式语言，除非参考风格本身就是说明书风格。
+*   **工具对接**：输出内容必须适配 `generate_script` 的字段定义，确保下游渲染环节无缝衔接。

Dockerfile

@@ -0,0 +1,31 @@
+# 基础镜像
+FROM python:3.11-slim
+
+# 设置工作目录
+WORKDIR /app
+
+# 复制文件
+COPY requirements.txt .
+COPY run.sh .
+COPY src/ ./src/
+COPY agent_fastapi.py .
+COPY cli.py .
+COPY config.toml .
+COPY web/ ./web/
+COPY prompts/ ./prompts/
+COPY .storyline/ ./.storyline/
+COPY download.sh .
+
+# 安装依赖
+RUN apt-get update && apt-get install -y ffmpeg wget unzip git git-lfs curl
+RUN pip install --no-cache-dir -r requirements.txt
+
+# 下载
+RUN chmod +x download.sh
+RUN ./download.sh
+
+# 暴露 HF Space 默认端口
+EXPOSE 7860
+
+# 启动命令
+CMD ["bash", "run.sh"]

LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "{}"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 FireRed-OpenStoryline Authors. All Rights Reserved.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

README.md

@@ -0,0 +1,9 @@
+---
+title: FireRed-OpenStoryline
+emoji: 🎬
+colorFrom: red
+colorTo: gray
+sdk: docker
+pinned: false
+---
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

README_zh.md

@@ -0,0 +1,279 @@
+<div align="center">
+  <a href="#gh-light-mode-only">
+    <img
+      src="https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/openstoryline/web/static/brand_white.png"
+      alt="openstoryline"
+      width="70%"
+    />
+  </a>
+
+  <a href="#gh-dark-mode-only">
+    <img
+      src="https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/openstoryline/web/static/brand_black.png"
+      alt="openstoryline"
+      width="70%"
+    />
+  </a>
+
+  <p>
+    <a href="./README_zh.md">🇨🇳 简体中文</a> | 
+    <a href="./README.md">🌏 English</a>
+  </p>
+  <p>
+    <a href="https://huggingface.co/FireRedTeam" target="_blank">
+    <img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-FireRedTeam-ffc107?color=ffc107&logoColor=white" style="display: inline-block;"/>
+    </a>
+    <img src="https://img.shields.io/badge/python-≥3.11-blue" alt="Python">
+    <img src="https://img.shields.io/badge/license-Apache%202.0-blue" alt="License">
+    <a href="https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/openstoryline/docs/media/others/group.jpg"><img src="https://img.shields.io/badge/Xiaohongshu-Group-E9DBFC?style=flat&logo=xiaohongshu&logoColor=white" alt="xiaohongshu"></a>
+  </p>
+</div>
+
+<div align="center">
+
+[🤗 HuggingFace Demo](https://fireredteam-firered-openstoryline.hf.space/) • [🌐 Homepage](https://fireredteam.github.io/demos/firered_openstoryline/)
+
+</div>
+
+<div align="center">
+  <video src="https://github.com/user-attachments/assets/9116767e-bcd9-417a-93d8-2db4d3d5df8e" width="70%" poster=""> </video>
+</div>
+
+
+**FireRed-OpenStoryline** 将复杂的视频创作转化为自然直观的对话体验。兼顾易用性和企业级可靠性，让视频创作对初学者和创意爱好者都变得简单友好。
+> FireRed，字面意思红色的火苗，取自“星星之火，可以燎原”。我们将这团火苗取名为 FireRed，就是希望将我们在真实场景中打磨出的 SOTA 能力，像火种一样撒向旷野，点燃全球开发者的想象力，共同改变这个 AI 的世界。
+
+## ✨ 核心特性
+- 🌐 **智能素材搜索与整理**： 自动在线搜索并下载符合你需求的图片和视频片段。基于用户主题素材进行片段拆分与内容理解
+- ✍️ **智能文案生成**： 结合用户主题、画面理解与情绪识别，自动构建故事线及契合的旁白。内置少样本（Few-shot）仿写能力，支持通过输入参考文本（如种草测评、日常碎碎念等）定义文案风格，实现语感、节奏与句式的精准复刻。
+- 🎵 **智能推荐音乐、配音与字体**：支持导入私有歌单，根据视频内容和情绪自动推荐背景音乐并智能卡点。只需描述"克制一点","偏情绪化","像纪录片旁白"等风格，系统即可匹配合适的配音与字体，保证整体风格协调统一。
+- 💬 **对话式精修**：支持快速删减、替换或重组片段；修改任意字幕文案；调整文字颜色、字体、描边、位置等视觉元素——所有操作均通过自然语言完成，即改即得。
+- ⚡ **剪辑技能沉淀**： 可一键保存为专属剪辑Skill，记录完整的剪辑逻辑。下次只需更换素材并选择对应Skill，即可快速复刻同款风格，实现高效批量生产。
+
+## 🏗️ 架构
+
+<p align="center">
+  <img src="https://raw.githubusercontent.com/FireRedTeam/fireredteam.github.io/main/demos/firered_openstoryline/pics/structure.jpg" alt="openstoryline 架构" width="800">
+</p>
+
+## ✨ 演示案例
+
+<table align="center">
+  <tr>
+    <td align="center"><b>种草视频</b></td>
+    <td align="center"><b>幽默有趣</b></td>
+    <td align="center"><b>好物分享</b></td>
+    <td align="center"><b>文艺风格</b></td>
+  </tr>
+  <tr>
+    <td align="center"><video src="https://github.com/user-attachments/assets/28043813-1fda-4077-80d4-c6f540d7c7cb" controls width="220"></video></td>
+    <td align="center"><video src="https://github.com/user-attachments/assets/a1e33da2-a799-4398-a1bb-b25bb5143d7c" controls width="220"></video></td>
+    <td align="center"><video src="https://github.com/user-attachments/assets/444fd0fb-8824-4c25-b449-9309b0fcfd85" controls width="220"></video></td>
+    <td align="center"><video src="https://github.com/user-attachments/assets/2e69fa0d-b693-4d4f-b4d2-45146254f9e8" controls width="220"></video></td>
+  </tr>
+  </tr>
+
+  <tr>
+    <td align="center"><b>开箱视频</b></td>
+    <td align="center"><b>宠物说话</b></td>
+    <td align="center"><b>旅行Vlog</b></td>
+    <td align="center"><b>年终总结</b></td>
+  </tr>
+  <tr>
+    <td align="center"><video src="https://github.com/user-attachments/assets/ff1d669b-1d27-4cf8-b0be-1b141c717466" controls width="220"></video></td>
+    <td align="center"><video src="https://github.com/user-attachments/assets/063608bb-7fbd-4841-a08f-032ae459499f" controls width="220"></video></td>
+    <td align="center"><video src="https://github.com/user-attachments/assets/bc441dfa-e995-4575-8401-ecefa269e57b" controls width="220"></video></td>
+    <td align="center"><video src="https://github.com/user-attachments/assets/533ef5c3-bb76-4416-bff7-825e88b00b7d" controls width="220"></video></td>
+  </tr>
+  </tr>
+</table>
+
+> <sub>
+> 🎨 <b>效果说明：</b>受限于开源素材的版权协议，第一行默认演示中的元素（字体/音乐）仅为基础效果。<b>强烈建议</b>接入<a href="https://github.com/FireRedTeam/FireRed-OpenStoryline/blob/main/docs/source/zh/guide.md#2-%E9%AB%98%E7%BA%A7%E4%BD%BF%E7%94%A8%E6%95%99%E7%A8%8B">自建元素库教程</a>，解锁商用级字体、音乐、特效等，可实现显著优于默认效果的视频质量。<br>
+> ⚠️ <b>画质注：</b>受限于README展示空间，演示视频经过极限压缩。实际运行默认保持原分辨率输出，支持自定义尺寸。<br>
+> Demo中：<b>第一行</b>为默认开源素材效果（受限模式），<b>第二行</b>为小红书App「AI剪辑」元素库效果。👉 <a href="https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/openstoryline/docs/media/others/ai_cut_guide.png">点击查看体验教程</a><br>
+> ⚖️ <b>免责声明：</b>演示中包含的用户自摄素材及品牌标识仅作技术能力展示，版权归原作者所有，严禁二次分发。如有侵权请联系删除。
+> </sub>
+
+
+
+## 📦 安装
+
+### 1. 克隆仓库
+```bash
+# 如果没有安装git，参考官方网站进行安装：https://git-scm.com/install/
+# 或手动打包下载，并解压
+git clone https://github.com/FireRedTeam/FireRed-OpenStoryline.git
+cd FireRed-OpenStoryline
+```
+
+### 2. 创建虚拟环境
+
+按照官方指南安装 Conda（推荐Miniforge，安装过程中建议勾选上自动配置环境变量）：https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html
+
+```
+# 要求python>=3.11
+conda create -n storyline python=3.11
+conda activate storyline
+```
+
+### 3. 资源下载与依赖安装
+#### 3.1 一键安装（仅支持Linux和MacOS）
+```
+sh build_env.sh
+```
+
+#### 3.2 手动安装
+##### A. MacOS 或 Linux
+  - Step 1: 安装 wget（如果尚未安装）
+    
+    ```
+    # MacOS: 如果你还没有安装 Homebrew，请先安装：https://brew.sh/
+    brew install wget
+    
+    # Ubuntu/Debian
+    sudo apt-get install wget
+    
+    # CentOS
+    sudo yum install wget
+    ```
+
+  - Step 2: 下载资源
+  
+    ```bash
+    sh download.sh
+    ```
+  
+  - Step 3: 安装依赖
+
+    ```bash
+    pip install -r requirements.txt
+    ```
+
+###### B. Windows
+  - Step 1: 准备目录：在项目根目录下新建目录 `.storyline`。
+
+  - Step 2: 下载并解压：
+
+    *   [下载模型 (models.zip)](https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/openstoryline/models.zip) -> 解压至 `.storyline` 目录。
+  
+    *   [下载资源 (resource.zip)](https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/openstoryline/resource.zip) -> 解压至 `resource` 目录。
+  - Step 3:  **安装依赖**：
+    ```bash
+    pip install -r requirements.txt
+    ```
+
+
+## 🚀 快速开始
+注意：在开始之前，您需要先在 config.toml 中配置 API-Key。详细信息请参阅文档 [API-Key 配置](docs/source/zh/api-key.md)
+
+### 1. 启动 MCP 服务器
+
+#### MacOS or Linux
+  ```bash
+  PYTHONPATH=src python -m open_storyline.mcp.server
+  ```
+
+#### Windows
+  ```
+  $env:PYTHONPATH="src"; python -m open_storyline.mcp.server
+  ```
+
+
+### 2. 启动对话界面
+
+- 方式 1：命令行界面
+
+  ```bash
+  python cli.py
+  ```
+
+- 方式 2：Web 界面
+
+  ```bash
+  uvicorn agent_fastapi:app --host 127.0.0.1 --port 7860
+  ```
+
+## 🐳 Docker 部署
+
+如果未安装 Docker，请先安装 https://www.docker.com/products/docker-desktop/
+
+### 拉取镜像
+```
+docker pull openstoryline/openstoryline:v1.0.0
+```
+
+### 启动镜像
+```
+docker run \
+  -v $(pwd)/config.toml:/app/config.toml \
+  -v $(pwd)/outputs:/app/outputs \
+  -p 7860:7860 \
+  openstoryline/openstoryline:v1.0.0
+```
+启动后访问Web界面 http://127.0.0.1:7860
+
+## 📁 项目结构
+```
+FireRed-OpenStoryline/
+├── 🎯 src/open_storyline/           核心应用
+│   ├── mcp/                         🔌 模型上下文协议
+│   ├── nodes/                       🎬 视频处理节点
+│   ├── skills/                      🛠️ Agent 技能库
+│   ├── storage/                     💾 Agent 记忆系统
+│   ├── utils/                       🧰 工具函数
+│   ├── agent.py                     🤖 Agent 构建
+│   └── config.py                    ⚙️ 配置管理
+├── 📚 docs/                         文档
+├── 🐳 Dockerfile                    Docker 配置
+├── 💬 prompts/                      LLM 提示词模板
+├── 🎨 resource/                     静态资源
+│   ├── bgms/                        背景音乐库
+│   ├── fonts/                       字体文件
+│   ├── script_templates/            视频脚本模板
+│   └── unicode_emojis.json          Emoji 列表
+├── 🔧 scripts/                      工具脚本
+├── 🌐 web/                          Web 界面
+├── 🚀 agent_fastapi.py              FastAPI 服务器
+├── 🖥️ cli.py                        命令行界面
+├── ⚙️ config.toml                   主配置文件
+├── 🚀 build_env.sh                  环境构建脚本
+├── 📥 download.sh                   资源下载脚本
+├── 📦 requirements.txt              运行时依赖
+└── ▶️ run.sh                        启动脚本
+
+```
+
+## 📚 文档
+
+### 📖 教程索引
+
+- [API申请与配置](docs/source/zh/api-key.md) - 如何申请和配置 API 密钥
+- [使用教程](docs/source/zh/guide.md) - 常见用例和基本操作
+- [常见问题](docs/source/zh/faq.md) - 常见问题解答
+
+## TODO
+
+- [ ] 添加口播类型视频剪辑功能
+- [ ] 添加音色克隆功能
+- [ ] 添加更多的转场/滤镜/特效功能
+- [ ] 添加图像/视频生成和编辑能力
+- [ ] 支持GPU渲染和高光裁切
+
+## 致谢
+
+本项目基于以下优秀的开源项目构建：
+
+
+### 核心依赖
+- [MoviePy](https://github.com/Zulko/moviepy) - 视频编辑库
+- [FFmpeg](https://ffmpeg.org/) - 多媒体框架
+- [LangChain](https://www.langchain.com/) - 提供预构建Agent的框架
+
+## 📄 License
+
+This project is licensed under the Apache License 2.0 - see the [LICENSE](LICENSE) file for details.
+
+## ⭐ Star History
+
+[![Star History Chart](https://api.star-history.com/svg?repos=FireRedTeam/FireRed-OpenStoryline&type=date&legend=top-left)](https://www.star-history.com/#FireRedTeam/FireRed-OpenStoryline&type=date&legend=top-left)

build_env.sh

@@ -0,0 +1,214 @@
+#!/bin/bash
+
+# 颜色定义 | Color Definitions
+RED='\033[0;31m'
+GREEN='\033[0;32m'
+YELLOW='\033[1;33m'
+BLUE='\033[0;34m'
+NC='\033[0m' # No Color
+
+# 打印带颜色的消息 | Print colored messages
+print_success() {
+    echo -e "${GREEN}[✓]${NC} $1"
+}
+
+print_error() {
+    echo -e "${RED}[✗]${NC} $1"
+}
+
+print_warning() {
+    echo -e "${YELLOW}[!]${NC} $1"
+}
+
+print_info() {
+    echo -e "${BLUE}[i]${NC} $1"
+}
+
+# 打印标题 | Print Title
+echo ""
+echo "╔════════════════════════════════════════════════════════════════╗"
+echo "║    Storyline 项目依赖安装脚本 | Dependency Installation       ║"
+echo "║    使用 conda activate storyline 激活环境后运行                ║"
+echo "╚════════════════════════════════════════════════════════════════╝"
+echo ""
+
+# ==========================================
+# 步骤 0: 检测操作系统
+# Step 0: Detect OS
+# ==========================================
+print_info "检测操作系统... | Detecting OS..."
+
+if [[ "$OSTYPE" == "darwin"* ]]; then
+    IS_MACOS=true
+    IS_LINUX=false
+    print_success "检测到 MacOS 系统 | MacOS detected"
+elif [[ "$OSTYPE" == "linux-gnu"* ]]; then
+    IS_MACOS=false
+    IS_LINUX=true
+    print_success "检测到 Linux 系统 | Linux detected"
+else
+    print_error "不支持的操作系统 | Unsupported operating system: $OSTYPE"
+    exit 1
+fi
+echo ""
+
+# ==========================================
+# 步骤 1: 检查 conda 环境
+# Step 1: Check conda environment
+# ==========================================
+echo "[1/4] 检查 conda 环境... | Checking conda environment..."
+
+if [ -z "$CONDA_DEFAULT_ENV" ]; then
+    print_error "未检测到 conda 环境 | No conda environment detected"
+    echo ""
+    echo "请先运行: conda activate storyline"
+    echo "Please run: conda activate storyline"
+    exit 1
+fi
+
+if [ "$CONDA_DEFAULT_ENV" != "storyline" ]; then
+    print_warning "当前环境: $CONDA_DEFAULT_ENV"
+    echo ""
+    read -p "建议使用 storyline 环境，是否继续? | Continue anyway? (y/n) " -n 1 -r
+    echo ""
+    if [[ ! $REPLY =~ ^[Yy]$ ]]; then
+        echo "请运行: conda activate storyline"
+        exit 1
+    fi
+else
+    print_success "当前环境: storyline"
+fi
+
+# 显示 Python 信息
+print_info "Python 信息 | Python Info:"
+echo "  版本 | Version: $(python --version 2>&1)"
+echo "  路径 | Path: $(which python)"
+echo ""
+
+# ==========================================
+# 步骤 2: 检查 FFmpeg
+# Step 2: Check FFmpeg
+# ==========================================
+echo "[2/4] 检查 FFmpeg... | Checking FFmpeg..."
+
+if ! command -v ffmpeg &> /dev/null; then
+    print_warning "未检测到 FFmpeg | FFmpeg not detected"
+    echo ""
+    
+    read -p "是否安装 FFmpeg? | Install FFmpeg? (y/n) " -n 1 -r
+    echo ""
+    
+    if [[ $REPLY =~ ^[Yy]$ ]]; then
+        print_info "正在安装 FFmpeg... | Installing FFmpeg..."
+        
+        if [ "$IS_MACOS" = true ]; then
+            if ! command -v brew &> /dev/null; then
+                print_error "需要 Homebrew 来安装 FFmpeg | Homebrew required to install FFmpeg"
+                echo "请访问: https://brew.sh"
+                exit 1
+            fi
+            brew install ffmpeg
+        elif [ "$IS_LINUX" = true ]; then
+            if command -v apt-get &> /dev/null; then
+                sudo apt-get update
+                sudo apt-get install -y ffmpeg
+            elif command -v yum &> /dev/null; then
+                sudo yum install -y epel-release
+                sudo yum install -y ffmpeg ffmpeg-devel
+            else
+                print_error "无法识别的包管理器 | Unrecognized package manager"
+                exit 1
+            fi
+        fi
+        
+        if [ $? -eq 0 ]; then
+            print_success "FFmpeg 安装成功 | FFmpeg installed successfully"
+        else
+            print_error "FFmpeg 安装失败 | FFmpeg installation failed"
+            exit 1
+        fi
+    else
+        print_warning "跳过 FFmpeg 安装（可能影响音视频处理功能）"
+        print_warning "Skipping FFmpeg (may affect audio/video features)"
+    fi
+else
+    print_success "FFmpeg 已安装 | FFmpeg installed"
+    echo "  版本 | Version: $(ffmpeg -version 2>&1 | head -n 1)"
+fi
+echo ""
+
+# ==========================================
+# 步骤 3: 下载项目资源
+# Step 3: Download project resources
+# ==========================================
+echo "[3/4] 下载项目资源... | Downloading project resources..."
+
+if [ -f "download.sh" ]; then
+    print_info "执行资源下载脚本... | Running download script..."
+    chmod +x download.sh
+    ./download.sh
+    
+    if [ $? -eq 0 ]; then
+        print_success "资源下载完成 | Resources downloaded successfully"
+    else
+        print_error "资源下载失败 | Resource download failed"
+        exit 1
+    fi
+else
+    print_warning "未找到 download.sh | download.sh not found"
+    echo "如需下载模型等资源，请手动执行 download.sh"
+    echo "To download models, please run download.sh manually"
+fi
+echo ""
+
+# ==========================================
+# 步骤 4: 安装 Python 依赖
+# Step 4: Install Python dependencies
+# ==========================================
+echo "[4/4] 安装 Python 依赖... | Installing Python dependencies..."
+
+if [ ! -f "requirements.txt" ]; then
+    print_error "未找到 requirements.txt | requirements.txt not found"
+    exit 1
+fi
+
+print_info "正在安装依赖包，请稍候... | Installing packages, please wait..."
+echo ""
+
+# 安装依赖
+print_info "安装依赖包... | Installing dependencies..."
+
+# 尝试使用清华镜像源
+pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
+
+if [ $? -ne 0 ]; then
+    print_warning "清华镜像安装失败，尝试使用默认源... | Tsinghua mirror failed, trying default..."
+    pip install -r requirements.txt
+    
+    if [ $? -ne 0 ]; then
+        print_error "依赖安装失败 | Dependency installation failed"
+        echo ""
+        echo "请尝试手动安装: pip install -r requirements.txt"
+        exit 1
+    fi
+fi
+
+print_success "依赖安装完成 | Dependencies installed successfully"
+echo ""
+
+# ==========================================
+# 安装完成 | Installation Complete
+# ==========================================
+echo ""
+echo "╔════════════════════════════════════════════════════════════════╗"
+echo "║          安装成功！| Installation Successful!                 ║"
+echo "╚════════════════════════════════════════════════════════════════╝"
+echo ""
+
+print_info "环境信息 | Environment Info:"
+echo "  Conda 环境 | Conda Env: $CONDA_DEFAULT_ENV"
+echo "  Python: $(python --version 2>&1)"
+command -v ffmpeg &> /dev/null && echo "  FFmpeg: $(ffmpeg -version 2>&1 | head -n 1 | cut -d' ' -f3)"
+echo ""
+
+print_success "现在可以运行项目了！| You can now run the project!"

cli.py

@@ -0,0 +1,99 @@
+import asyncio
+import time
+import uuid
+import os,sys
+import json
+
+from typing import List
+
+from langchain_core.messages import AIMessage, BaseMessage, HumanMessage, SystemMessage
+
+# Add src directory to Python module search path
+ROOT_DIR = os.path.dirname(__file__)
+SRC_DIR = os.path.join(ROOT_DIR, "src")
+
+if SRC_DIR not in sys.path:
+    sys.path.insert(0, SRC_DIR)
+
+from open_storyline.agent import ClientContext, build_agent
+from open_storyline.utils.prompts import get_prompt
+from open_storyline.utils.media_handler import scan_media_dir
+from open_storyline.config import load_settings, default_config_path
+from open_storyline.storage.agent_memory import ArtifactStore
+from open_storyline.mcp.hooks.node_interceptors import ToolInterceptor
+from open_storyline.mcp.hooks.chat_middleware import PrintStreamingTokens
+
+_MEDIA_STATS_INFO_IDX = 1
+
+async def main():
+    session_id = f"run_{int(time.time())}_{uuid.uuid4().hex[:8]}"
+    cfg = load_settings(default_config_path())
+    
+    artifact_store = ArtifactStore(cfg.project.outputs_dir, session_id=session_id)
+    agent, node_manager = await build_agent(cfg=cfg, session_id=session_id, store=artifact_store, tool_interceptors=[ToolInterceptor.inject_media_content_before, ToolInterceptor.save_media_content_after, ToolInterceptor.inject_tts_config])
+
+    context = ClientContext(
+        cfg=cfg,
+        session_id=session_id,
+        media_dir=cfg.project.media_dir,
+        bgm_dir=cfg.project.bgm_dir,
+        outputs_dir=cfg.project.outputs_dir,
+        node_manager=node_manager,
+        chat_model_key=cfg.llm.model,
+    )
+
+    messages: List[BaseMessage] = [
+        SystemMessage(content=get_prompt("instruction.system", lang='en')),
+        SystemMessage(content="【User media statistics】{}"),
+        ]
+
+    print("Smart Editing Agent v 1.0.0")
+    print("Please describe your editing needs, type /exit to exit.")
+
+    while True:
+        try:
+            user_input = input("You: ").strip()
+        except (EOFError, KeyboardInterrupt):
+            print("\nGoodBye~")
+            break
+
+        if not user_input:
+            continue
+        if user_input in ("/exit", "/quit"):
+            print("\nGoodBye~")
+            break
+
+        media_stats = scan_media_dir(context.media_dir)
+        messages[_MEDIA_STATS_INFO_IDX] = SystemMessage(
+            content=(
+                f"【User media statistics】{json.dumps(media_stats, ensure_ascii=False)}"
+            )
+        )
+    
+        messages.append(HumanMessage(content=user_input))
+        
+        print("Agent: ", end="", flush=True)
+
+        stream = PrintStreamingTokens()
+
+        result = await agent.ainvoke(
+            {"messages": messages},
+            context=context,
+            config={"callbacks": [stream]},
+        )
+
+        print("\n")
+
+        messages = result["messages"]
+
+        final_text = None
+        for m in reversed(messages):
+            if isinstance(m, AIMessage):
+                final_text = m.content
+                break
+
+        print(f"\nAgent: {final_text or '(No final response generated)'}\n")
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

config.toml

@@ -0,0 +1,157 @@
+# ============= 开发者选项 / Developer Options ===============
+[developer]
+developer_mode = false
+default_llm = "deepseek-chat"
+default_vlm = "qwen3-vl-8b-instruct"
+print_context = false # 在拦截器打印模型拿到的全部上下文，会很长 / Print full context in interceptor (output will be very long)
+
+# ============= 模型配置 for 体验网页 ===============
+[developer.chat_models_config."deepseek-chat"]
+base_url = ""
+api_key = ""
+temperature = 0.1
+
+[developer.chat_models_config."qwen3-vl-8b-instruct"]
+base_url = ""
+api_key = "" 
+timeout = 20.0
+temperature = 0.1
+max_retries = 2
+
+# ============= 项目路径 / Project Paths ======================
+[project]
+media_dir = "./outputs/media"
+bgm_dir = "./resource/bgms"
+outputs_dir = "./outputs"
+
+# ============= 模型配置 for user / Model Config for User =============
+[llm]
+model = "deepseek-chat"
+base_url = ""
+api_key = "" 
+timeout = 30.0 # 单位：秒
+temperature = 0.1
+max_retries = 2
+
+[vlm]
+model = "qwen3-vl-8b-instruct"
+base_url = ""
+api_key = "" 
+timeout = 20.0 # 单位：秒
+temperature = 0.1
+max_retries = 2
+
+
+# ============= MCP Server 相关 / MCP Server Related =============
+[local_mcp_server]
+server_name = "storyline"
+server_cache_dir = ".storyline/.server_cache"
+server_transport = "streamable-http"  # server 和 host之间的传输方式 / Transport method between server and host
+url_scheme = "http"  
+connect_host = "127.0.0.1" # 不要改动 / Do not change
+port = 8001                # 如果端口冲突，可以随便用一个有空的端口 / Use any available port if conflict occurs
+path = "/mcp"              # 默认值，一般不用改 / Default value, usually unchanged
+
+json_response = true       # 建议用 True / Recommended: True
+stateless_http = false     # 强烈建议用 False / Strongly recommended: False
+timeout = 600
+available_node_pkgs = [
+    "open_storyline.nodes.core_nodes"
+]
+available_nodes = [
+    "LoadMediaNode", "SearchMediaNode", "SplitShotsNode",
+    "UnderstandClipsNode", "FilterClipsNode", "GroupClipsNode", "GenerateScriptNode", "ScriptTemplateRecomendation",
+    "GenerateVoiceoverNode", "SelectBGMNode", "RecommendTransitionNode", "RecommendTextNode",
+    "PlanTimelineProNode", "RenderVideoNode"
+]
+
+# =========== skills ==========
+[skills]
+skill_dir = "./.storyline/skills"
+
+# =========== pexels ==========
+[search_media]
+pexels_api_key = ""
+
+# ============= 镜头分割 / Shot Segmentation =============
+[split_shots]
+transnet_weights = ".storyline/models/transnetv2-pytorch-weights.pth"
+transnet_device = "cpu"
+
+# ============= 视频视觉理解 / Video Visual Understanding =============
+[understand_clips]
+sample_fps = 2.0                 # 每秒抽几帧 / Frames sampled per second
+max_frames = 64                  # 单clip抽帧上限兜底，避免长视频爆 token / Max frames per clip limit to prevent token overflow
+
+# ============= 文案模板 / Script Templates =============
+[script_template]
+script_template_dir = "./resource/script_templates"
+script_template_info_path = "./resource/script_templates/meta.json"
+
+# ============= 配音生成 / Voiceover Generation ===================
+[generate_voiceover]
+tts_provider_params_path = "./resource/tts/tts_providers.json"
+
+[generate_voiceover.providers.302]
+base_url = ""
+api_key = ""
+
+[generate_voiceover.providers.bytedance]
+uid = ""
+appid = ""
+access_token = ""
+
+[generate_voiceover.providers.minimax]
+base_url = ""
+api_key = ""
+
+
+# ============= BGM选择 / BGM Selection ====================
+# 主要是用于计算音乐特征的一些参数 / Mainly parameters for calculating music features
+[select_bgm]
+sample_rate = 22050
+hop_length = 2048   # 每次分析窗口向前跳多少个采样点，越小越精细（但更慢） / Hop length samples; smaller = more precise but slower
+frame_length = 2048 # 计算信号的均方根RMS的窗口大小。越大越稳定，但对瞬态不敏感 / Window size for RMS; larger = stable but less sensitive to transients
+
+# ============= 字体推荐 / Font Recommendation ====================
+[recommend_text]
+font_info_path = "resource/fonts/font_info.json"
+
+# ============= 时间线组织 / Timeline Organization ====================
+[plan_timeline]
+beat_type_max = 1 # 使用多强的鼓点，例如4/4中，鼓点类似1,2,1,3，其中1是最强的，3最弱 / Beat strength (e.g., in 4/4: 1=strongest, 3=weakest)
+title_duration = 0 # 片头时长 (ms) / Intro duration (ms)
+bgm_loop = true # 是否允许 bgm 循环 / Allow BGM loop
+min_clip_duration = 1000
+
+estimate_text_min = 1500 # 在没有TTS的情况下，估计每段字幕至少上屏多久 / Min on-screen duration for subtitles without TTS
+estimate_text_char_per_sec = 6.0 # 在没有TTS的情况下，估计每秒展示几个字 / Estimated characters per second without TTS
+
+image_default_duration = 3000 # 默认的图片播放时长 / Default image duration
+group_margin_over_voiceover = 1000 # 在一个group中，画面比配音多出现多久 / Extra visual duration over voiceover in a group
+
+[plan_timeline_pro]
+
+min_single_text_duration = 200      # 单段文字最小时长 (ms) / min single text duration (ms)
+max_text_duration        = 5000     # 单句文字最大时长 (ms) / max text sentence duration (ms)
+img_default_duration     = 1500     # 默认图片时长 (ms) / default image duration (ms)
+
+min_group_margin = 1500             # 段落/组最小间距 (ms) / min paragraph/group margin (ms)
+max_group_margin = 2000             # 段落/组最大间距 (ms) / max paragraph/group margin (ms)
+
+min_clip_duration = 1000            # 最小片段时长 (ms) / min clip duration (ms)
+
+tts_margin_mode = "random"          # random | avg | max | min
+min_tts_margin  = 300               # 最小 TTS 间隔 (ms) / min TTS gap (ms)
+max_tts_margin  = 400               # 最大 TTS 间隔 (ms) / max TTS gap (ms)
+
+text_tts_offset_mode = "random"     # random | avg | max | min
+min_text_tts_offset  = 0            # 最小文字-TTS偏移 (ms) / min text–TTS offset (ms)
+max_text_tts_offset  = 0            # 最大文字-TTS偏移 (ms) / max text–TTS offset (ms)
+
+long_short_text_duration = 3000     # 长/短文本阈值 (ms) / long/short text threshold (ms)
+long_text_margin_rate   = 0.0       # 长文本起始边距率 / long text start margin rate
+short_text_margin_rate  = 0.0       # 短文本起始边距率 / short text start margin rate
+
+text_duration_mode = "with_tts"     # with_tts | with_clip (随配音 | 随片段)
+is_text_beats      = false          # 文字对齐音乐节拍 / align text with music beats

docs/source/en/api-key.md

@@ -0,0 +1,134 @@
+# API Key Configuration Guide
+
+## 1. Large Language Model (LLM)
+
+### Using DeepSeek as an Example
+
+**Official Documentation**: https://api-docs.deepseek.com/zh-cn/
+
+Note: For users outside China, we recommend using large language models such as Gemini, Claude, or ChatGPT for the best experience.
+
+### Configuration Steps
+
+1. **Apply for API Key**
+   - Visit platform: https://platform.deepseek.com/usage
+   - Log in and apply for API Key
+   - ⚠️ **Important**: Save the obtained API Key securely
+
+2. **Configuration Parameters**
+   - **Model Name**: `deepseek-chat`
+   - **Base URL**: `https://api.deepseek.com/v1`
+   - **API Key**: Fill in the Key obtained in the previous step
+
+3. **API Configuration**
+   - **Web Usage**: Select "Use Custom Model" in the LLM model form, and fill in the model according to the configuration parameters
+   - **Local Deployment**: In config.toml, locate `[developer.chat_models_config."deepseek-chat"]` and fill in the configuration parameters to make the default configuration accessible from the Web page. Locate `[llm]` and configure model, base_url, and api_key
+
+## 2. Multimodal Large Language Model (VLM)
+
+### 2.1 Using GLM-4.6V
+
+**API Key Management**: https://open.bigmodel.cn/usercenter/proj-mgmt/apikeys
+
+### Configuration Parameters
+
+- **Model Name**: `glm-4.6v`
+- **Base URL**: `https://open.bigmodel.cn/api/paas/v4/`
+
+### 2.2 Using Qwen3-VL
+
+**API Key Management**: Go to Alibaba Cloud Bailian Platform to apply for an API Key https://bailian.console.aliyun.com/cn-beijing/?apiKey=1&tab=globalset#/efm/api_key
+
+  - **Model Name**: `qwen3-vl-8b-instruct`
+  - **Base URL**: `https://dashscope.aliyuncs.com/compatible-mode/v1`
+
+  - Parameter Configuration: Select "Use Custom Model" in the VLM Model form and fill in the parameters. For local deployment, locate `[vlm]` and configure model, base_url, and api_key. Add the following fields in config.toml as the default Web API configuration:
+   ```
+   [developer.chat_models_config."qwen3-vl-8b-instruct"]
+   base_url = "https://dashscope.aliyuncs.com/compatible-mode/v1"
+   api_key = "YOUR_API_KEY" 
+   timeout = 20.0
+   temperature = 0.1
+   max_retries = 2
+   ```
+
+### 2.3 Using Qwen3-Omni
+
+Qwen3-Omni can also be applied for through the Alibaba Cloud Bailian Platform. The specific parameters are as follows, which can be used for automatic labeling music in omni_bgm_label.py
+- **Model Name**: `qwen3-omni-flash-2025-12-01`
+- **Base URL**: `https://dashscope.aliyuncs.com/compatible-mode/v1`
+
+For more details, please refer to the documentation: https://bailian.console.aliyun.com/cn-beijing/?tab=doc#/doc
+
+Model List: https://help.aliyun.com/zh/model-studio/models
+
+Billing Dashboard: https://billing-cost.console.aliyun.com/home
+
+## 3. Pexels Image and Video Download API Key Configuration
+
+1. Open the Pexels website, register an account, and apply for an API key at https://www.pexels.com/api/
+<div align="center">
+  <img src="https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/openstoryline/docs/resource/pexels_api.png" alt="Pexels API application" width="70%">
+  <p><em>Figure 1: Pexels API Application Page</em></p>
+</div>
+
+2. Web Usage: Locate the Pexels configuration, select "Use custom key", and enter your API key in the form.
+<div align="center">
+  <img src="https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/openstoryline/docs/resource/use_pexels_api_en.png" alt="Pexels API input" width="70%">
+  <p><em>Figure 2: Pexels API Usage</em></p>
+</div>
+
+3. Local Deployment: Fill in the API key in the `pexels_api_key` field in the `config.toml` file as the default configuration for the project.
+
+## 4. TTS (Text-to-Speech) Configuration
+
+### Option 1: 302.ai
+
+**Service URL**: https://302.ai/product/detail/302ai-mmaudio-text-to-speech
+
+### Option 2: MiniMax
+
+**Subscription Page**: https://platform.minimax.io/subscribe/audio-subscription
+
+**Configuration Steps**:
+1. Create API Key
+2. Visit: https://platform.minimax.io/user-center/basic-information/interface-key
+3. Obtain and save API Key
+
+### Option 3: Bytedance
+1. Step 1: Enable Audio/Video Subtitle Generation Service
+   Use the legacy page to find the audio/video subtitle generation service:
+
+   - Visit: https://console.volcengine.com/speech/service/9?AppID=8782592131
+
+2. Step 2: Obtain Authentication Information
+   View the account basic information page:
+   
+   - Visit: https://console.volcengine.com/user/basics/
+
+<div align="center">
+  <img src="https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/openstoryline/docs/resource/use_bytedance_tts_zh.png" alt="Bytedance TTS API Configuration" width="70%">
+  <p><em>Figure 3: Bytedance TTS API Usage</em></p>
+</div>
+
+   You need to obtain the following information:
+   - **UID**: The ID from the main account information
+   - **APP ID**: The APP ID from the service interface authentication information
+   - **Access Token**: The Access Token from the service interface authentication information
+   
+   For local deployment, modify the config.toml file:
+
+```
+[generate_voiceover.providers.bytedance]
+uid = ""
+appid = ""
+access_token = ""
+```
+
+For detailed documentation, please refer to: https://www.volcengine.com/docs/6561/80909
+
+## Important Notes
+
+- All API Keys must be kept secure to avoid leakage
+- Ensure sufficient account balance before use
+- Regularly monitor API usage and costs

docs/source/en/faq.md

@@ -0,0 +1,18 @@
+# Most Frequently Asked Questions
+
+## Environment Related Issue
+<details>
+<summary><strong>Issue 1: When activating conda environment, script execution is prohibited.</strong></summary>
+
+Please refer to this article for the solution: [https://juejin.cn/post/7349212852644954139](https://juejin.cn/post/7349212852644954139)
+</details>
+
+<details>
+
+<summary><strong>Issue 2: Error creating a virtual environment after installing Conda on Windows.</strong></summary>
+
+- **Cause:**  
+  This is caused by Conda not being added to the system environment variables during installation.
+- **Solution:**  
+  You need to open **Anaconda Prompt**, **Miniconda Prompt**, or **Miniforge Prompt** (depending on which one you installed) from the Start Menu, `cd` to the current directory, and then proceed to create the environment.
+</details>

docs/source/en/guide.md

@@ -0,0 +1,220 @@
+# OpenStoryline User Guide
+
+---
+
+## 0. Environment Setup
+
+See the [README](https://github.com/FireRedTeam/FireRed-OpenStoryline/blob/main/README.md) section.
+
+## 1. Basic Usage
+
+### 1.1 Getting Started
+
+You can start creating in two ways:
+
+1. **You have your own media**
+
+   * Click the file upload button on the left side of the chat box and select your images/videos.
+   * Then type your editing goal in the input field, for example: *Use my footage to edit a family vlog with an upbeat rhythm.*
+
+2. **You don’t have media**
+
+   * Just describe the theme/mood.
+   * For example: *Help me create a summer beach travel vlog—sunny, fresh, and cheerful.*
+
+Automatic asset retrieval is powered by [Pexels](https://www.pexels.com/). Please enter your Pexels API key in the website sidebar.
+
+**Disclaimer:** We only provide the tool. All assets downloaded or used via this tool (e.g., Pexels images) are fetched by the user through the API. We assume no responsibility for the content of videos generated by users, the legality of the assets, or any copyright/portrait-right disputes arising from the use of this tool. Please comply with Pexels’ license when using it: [https://www.pexels.com/zh-cn/license](https://www.pexels.com/zh-cn/license)
+[https://www.pexels.com/terms-of-service](https://www.pexels.com/terms-of-service)
+
+If you just want to explore it first, you can also use it like a normal chat model, for example:
+
+* “Introduce yourself”
+
+<img src="https://github.com/user-attachments/assets/a7c102a0-299d-4fcb-a890-0bcb165867d2" alt="demo" width="500">
+
+### 1.2 Editing
+
+OpenStoryline supports **intent intervention and partial redo at any stage**. After a step completes, you can simply describe what you want to change in one sentence. The agent will locate the step that needs to be rerun, without restarting from the beginning. For example:
+
+* Remove the clip where the camera is filming the sky.
+* Switch to a more upbeat background music.
+* Change the subtitle color to better match the sunset theme.
+
+<img src="https://github.com/user-attachments/assets/18c1ac82-873d-4ced-beb3-443d0fc9192c" alt="demo" width="500">
+
+### 1.3 Style Imitation
+
+With the style imitation Skill, you can reproduce almost any writing style to generate copy. For example:
+
+* Generate copy in a Shakespearean style for me.
+* Mimic the tone of my social media posts.
+
+<img src="https://github.com/user-attachments/assets/67edcb95-a71d-447c-ac13-ae28d0bbd698" alt="demo" width="500">
+
+### 1.4 Interrupting
+
+At any moment while the agent is running, if its behavior is not as expected, you can:
+
+* Click the **Stop** button on the right side of the input box to stop the model reply and tool calls, **or**
+* Press **Enter** to send a new prompt—the system will automatically interrupt and follow your new instruction.
+
+Interrupting does **not** clear the current progress. Existing replies and executed tool results will be kept, and you can continue from the current state.
+
+### 1.5 Switching Languages
+
+Click the language button in the top-right corner of the page to switch between Chinese and English:
+
+* The sidebar and tool-call cards will switch display language accordingly.
+* The prompt language used inside tools will also switch.
+* Past chat history will **not** be automatically translated.
+
+### 1.6 Saving
+
+After you polish a satisfying video, you can ask the agent to **summarize the editing logic** (rhythm, color tone, transition habits, etc.) and save it as your personal **“Editing Skill.”**
+
+Next time you edit similar content, simply ask the agent to use this Skill to reproduce the style.
+
+<img src="https://github.com/user-attachments/assets/d99faca2-233c-49d0-829e-336b2b76a46d" alt="demo" width="500">
+
+### 1.7 Mobile Usage
+
+**Warning: The commands below will expose your service to your local network. Use only on trusted networks. Do NOT run these commands on public networks.**
+
+If your media is on your phone and it’s inconvenient to transfer, you can use the following steps to use the editing agent on mobile.
+
+1. Fill in the LLM/VLM/Pexels/TTS configuration in config.toml.
+2. Change your web startup command to:
+
+   ```bash
+   # Reminder: --host 0.0.0.0 exposes the service to your LAN/public network. Use only on trusted networks.
+   uvicorn agent_fastapi:app --host 0.0.0.0 --port 7860
+   ```
+
+3. Find your computer’s IP address:
+
+   * **Windows:** run `ipconfig` in Command Prompt (cmd) and locate the IPv4 address
+   * **Mac:** hold **Option** and click the Wi-Fi icon
+   * **Linux:** run `ifconfig` in the terminal
+
+4. Then open the following address in your phone browser:
+
+   ```
+   {your_computer_ip}:7860
+   ```
+
+---
+
+## 2. Advanced Usage
+
+Due to copyright and distribution constraints, open-source resources may not be sufficient for many users’ editing needs. Therefore, we provide methods to add and build private asset libraries.
+
+---
+
+### 2.1 Custom Music Library
+
+Put your private music files into:
+
+`./resource/bgms`
+
+Then tag your music by writing metadata into:
+
+`./resouce/bgms/meta.json`
+
+Restart the MCP service to apply changes.
+
+**Tag Dimensions**
+
+* **scene:** Vlog, Travel, Relaxing, Emotion, Transition, Outdoor, Cafe, Evening, Scenery, Food, Date, Club
+* **genre:** Pop, BGM, Electronic, R&B/Soul, Hip Hop/Rap, Rock, Jazz, Folk, Classical, Chinese Style
+* **mood:** Dynamic, Chill, Happy, Sorrow, Romantic, Calm, Excited, Healing, Inspirational
+* **lang:** bgm, en, zh, ko, ja
+
+**How to Tag**
+
+* **Manual tagging:** Copy the format of other items in `meta.json` and add tags accordingly. **Note:** the `description` field is required.
+* **Auto tagging:** Use `qwen3-omni-flash` for automatic tagging (requires a Qwen model API key).
+
+Qwen3-omni labeling script:
+
+```bash
+export QWEN_API_KEY="you_api_key"
+python -m scripts.omni_bgm_label
+```
+
+Auto tags may not be fully accurate. If you need strong recommendations for specific scenarios, it’s recommended to manually review the results.
+
+---
+
+### 2.2 Custom Font Library
+
+Put your private font files into:
+
+`./resource/fonts`
+
+Then tag the fonts by editing:
+
+`./resource/fonts/font_info.json`
+
+Restart the MCP service to apply changes.
+
+**Tag Dimensions**
+
+* **class:** Creative, Handwriting, Calligraphy, Basic
+* **lang:** zh, en
+
+**How to Tag**
+Currently only manual tagging is supported—edit `./resource/fonts/font_info.json` directly.
+
+---
+
+### 2.3 Custom Copywriting Template Library
+
+Put your private copywriting templates into:
+
+`./resource/script_templates`
+
+Then tag them by writing metadata into:
+
+`./resource/fonts/meta.json`
+
+Restart the MCP service to apply changes.
+
+**Tag Dimensions**
+
+* **tags:** Life, Food, Beauty, Entertainment, Travel, Tech, Business, Vehicle, Health, Family, Pets, Knowledge
+
+**How to Tag**
+
+* **Manual tagging:** Copy the format of other items in `meta.json` and add tags accordingly. **Note:** the `description` field is required.
+* **Auto tagging:** Use DeepSeek for automatic tagging (requires the corresponding API key).
+
+DeepSeek labeling script:
+
+```bash
+export DEEPSEEK_API_KEY="you_api_key"
+python -m scripts.llm_script_template_label
+```
+
+Auto tags may not be fully accurate. If you need strong recommendations for specific scenarios, it’s recommended to manually review the results.
+
+---
+
+### 2.4 Custom Skill Library
+
+The repository includes two built-in Skills: one for writing-style imitation and another for saving editing workflows. If you want more custom skills, you can add them as follows:
+
+1. Create a new folder under `.storyline/skills`.
+2. Inside that folder, create a file named `SKILL.md`.
+3. The `SKILL.md` must start with:
+
+   ```markdown
+   ---
+   name: yous_skill_folder_name
+   description: your_skill_function_description
+   ---
+   ```
+
+   The `name` must match the folder name.
+4. Then write the detailed skill content (its role setting, which tools it should call, output format, etc.).
+5. Restart the MCP service to apply changes.

docs/source/zh/api-key.md

@@ -0,0 +1,132 @@
+# API-Key 配置指南
+
+## 一、大语言模型 (LLM)
+
+### 以 DeepSeek 为例
+
+**官方文档**：https://api-docs.deepseek.com/zh-cn/
+
+提示: 对于中国以外用户建议使用 Gemini、Claude、ChatGPT 等主流大语言模型以获得最佳体验。
+
+### 配置步骤
+
+1. **申请 API Key**
+   - 访问平台：https://platform.deepseek.com/usage
+   - 登录后申请 API Key
+   - ⚠️ **重要**：妥善保存获取的 API Key
+
+2. **配置参数**
+   - **模型名称**：`deepseek-chat`
+   - **Base URL**：`https://api.deepseek.com/v1`
+   - **API Key**：填写上一步获取的 Key
+
+3. **API填写**
+   - **Web使用**: 在LLM模型表单中选择使用自定义模型，模型按照配置参数进行填写
+   - **本地部署**: 在config.toml中 找到`[developer.chat_models_config."deepseek-chat"]` 将配置参数填写上去，使得Web页面可以访问到该默认配置。 找到`[llm]`并配置model、base_url、api_key
+
+## 二、多模态大模型 (VLM)
+
+### 2.1 使用GLM-4.6V
+
+**API Key 管理**：https://open.bigmodel.cn/usercenter/proj-mgmt/apikeys
+
+### 配置参数
+
+- **模型名称**：`glm-4.6v`
+- **Base URL**：`https://open.bigmodel.cn/api/paas/v4/`
+
+### 2.2 使用Qwen3-VL
+
+**API Key管理**：进入阿里云百炼平台申请API Key https://bailian.console.aliyun.com/cn-beijing/?apiKey=1&tab=globalset#/efm/api_key
+
+ - **模型名称**：`qwen3-vl-8b-instruct`
+ - **Base URL**：`https://dashscope.aliyuncs.com/compatible-mode/v1`
+
+ - **参数填写**：在VLM Model表单中选择"使用自定义模型"进行参数填写。本地部署时，找到`[vlm]`并配置model、base_url、api_key，在config.toml中新增以下字段作为Web的API默认配置：
+    ```
+   [developer.chat_models_config."qwen3-vl-8b-instruct"]
+   base_url = "https://dashscope.aliyuncs.com/compatible-mode/v1"
+   api_key = "YOUR_API_KEY" 
+   timeout = 20.0
+   temperature = 0.1
+   max_retries = 2
+   ```
+
+
+### 2.3 使用Qwen3-Omni
+
+Qwen3-Omni同样可以在阿里云百炼平台进行申请，具体参数如下，可用于omni_bgm_label.py的音频自动标注
+- **模型名称**：`qwen3-omni-flash-2025-12-01`
+- **Base URL**：`https://dashscope.aliyuncs.com/compatible-mode/v1`
+
+详细文档参考：https://bailian.console.aliyun.com/cn-beijing/?tab=doc#/doc
+
+阿里云模型列表：https://help.aliyun.com/zh/model-studio/models
+
+计费看板：https://billing-cost.console.aliyun.com/home
+
+## 三、Pexels 图像和视频下载API密钥配置
+
+1. 打开Pexels网站，注册账号，申请API https://www.pexels.com/zh-cn/api/key/ 
+<div align="center">
+  <img src="https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/openstoryline/docs/resource/pexels_api.png" alt="pexels下载图像和视频API申请" width="70%">
+  <p><em>图1: Pexels API申请页面</em></p>
+</div>
+
+2. 网页使用：找到Pexels配置，选择使用自定义key，将API key填入表单中。
+<div align="center">
+  <img src="https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/openstoryline/docs/resource/use_pexels_api_zh.png" alt="pexels API填写" width="70%">
+  <p><em>图2: Pexels API 使用</em></p>
+</div>
+
+3. 本地部署的项目：我们将API填写在config.toml中的pexels_api_key字段中。作为项目的默认配置
+
+## 四、TTS (文本转语音) 配置
+
+### 方案一：302.ai
+
+**服务地址**：https://302.ai/product/detail/302ai-mmaudio-text-to-speech
+
+### 方案二：MiniMax
+
+**订阅页面**：https://platform.minimax.io/subscribe/audio-subscription
+
+**配置步骤**：
+1. 创建 API Key
+2. 访问：https://platform.minimax.io/user-center/basic-information/interface-key
+3. 获取并保存 API Key
+
+### 方案三：bytedance
+1. 步骤1：开通音视频字幕生成服务
+   使用旧版页面，找到音视频字幕生成服务：
+   - 访问：https://console.volcengine.com/speech/service/9?AppID=8782592131
+
+2. 步骤2：获取认证信息
+   查看账号基本信息页面：
+   - 访问：https://console.volcengine.com/user/basics/
+
+<div align="center">
+  <img src="https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/openstoryline/docs/resource/use_bytedance_tts_zh.png" alt="Bytedance TTS API填写" width="70%">
+  <p><em>图3: Bytedance TTS API 使用</em></p>
+</div>
+
+   需要获取以下信息：
+   - **UID**: 主账号信息中的 ID
+   - **APP ID**: 服务接口认证信息中的 APP ID
+   - **Access Token**: 服务接口认证信息中的 Access Token
+   
+   本地部署使用修改config.toml中
+   ```
+   [generate_voiceover.providers.bytedance]
+   uid = ""
+   appid = ""
+   access_token = ""
+   ```
+
+详细文档请参考：https://www.volcengine.com/docs/6561/80909?lang=zh
+
+## 注意事项
+
+- 所有 API Key 均需妥善保管，避免泄露
+- 使用前请确认账户余额充足
+- 建议定期检查 API 调用量和费用

docs/source/zh/faq.md

@@ -0,0 +1,23 @@
+# 最常问的问题
+
+## 环境相关的问题
+
+<details>
+<summary><strong>问题 1: Conda 激活环境时发现脚本执行被禁止</strong></summary>
+
+- **参考链接**：[https://juejin.cn/post/7349212852644954139](https://juejin.cn/post/7349212852644954139)
+- **解决方法**： 在 PowerShell 中输入以下命令后重试：
+    ```powershell
+    Set-ExecutionPolicy RemoteSigned -Scope CurrentUser
+    ```
+
+</details>
+
+
+<details>
+<summary>
+<strong>问题 2: Windows 安装 Conda 后，创建虚拟环境时报错</strong></summary>
+
+- **原因**: 这是由于安装时没有将 conda 加入到环境变量导致的。
+- **解决方法**: 需要从开始菜单打开 Anaconda Prompt / Miniconda Prompt / Miniforge Prompt，cd 到当前目录，再创建环境。 
+</details>

docs/source/zh/guide.md

@@ -0,0 +1,154 @@
+# OpenStoryline 使用教程
+---
+## 0. 环境安装
+
+参见[README](https://github.com/FireRedTeam/FireRed-OpenStoryline/blob/main/README_zh.md)部分
+
+## 1. 基础使用教程
+
+### 1.1. 开始
+你可以用两种方式开始创作：
+
+1. 有素材  
+    - 点击对话框左侧文件上传按钮，选择你的图片/视频素材
+    - 然后在输入框写下剪辑目标，例如：用我的素材剪一条新年全家欢 vlog，节奏轻快
+
+2. 没素材  
+    - 直接描述主题/氛围即可
+    - 例如：帮我剪一个夏日海滩旅行 vlog，阳光、清爽、欢快
+
+自动素材检索来自 [Pexels](https://www.pexels.com/zh-cn/)，请在网页侧边栏填写 Pexels API Key。  
+
+免责声明：我们只提供工具，所有通过本工具下载和使用的素材（如 Pexels 图像）都由用户自行通过 API 获取，我们不对用户生成的视频内容、素材的合法性或因使用本工具导致的任何版权/肖像权纠纷承担责任。使用时请遵循 Pexels 的许可协议：[https://www.pexels.com/zh-cn/license](https://www.pexels.com/zh-cn/license)  
+[https://www.pexels.com/terms-of-service](https://www.pexels.com/terms-of-service)
+
+如果你只是想先了解它，也可以当作普通对话模型使用，例如：
+
+- “介绍一下你自己”  
+<img src="https://github.com/user-attachments/assets/a7c102a0-299d-4fcb-a890-0bcb165867d2" alt="demo" width="500">
+
+### 1.2. 编辑
+
+OpenStoryline 支持在任意阶段进行意图干预与局部重做：当某一步骤完成后，你可以直接用一句话提出修改要求，Agent会定位到需要重跑的步骤，而无需从流程起点重新开始。例如
+- 帮我去掉那个拍摄天空的片段。
+- 换一个欢快一点的背景音乐。
+- 字幕换成更符合夕阳主题的颜色  
+<img src="https://github.com/user-attachments/assets/18c1ac82-873d-4ced-beb3-443d0fc9192c" alt="demo" width="500">
+
+### 1.3. 仿写
+依靠仿写Skill复刻任意文风生成文案。例如：
+- 用文言文为我进行古风混剪。
+- 模仿鲁迅风格生成文案。
+- 模仿我发朋友圈的语气。  
+<img src="https://github.com/user-attachments/assets/67edcb95-a71d-447c-ac13-ae28d0bbd698" alt="demo" width="500">
+
+### 1.4. 中断
+在 Agent 执行的任意时刻，如果行为不符合预期，你可以随时：
+
+- 点击输入框右侧的中止按钮：停止大模型回复与工具调用
+- 或者直接按 Enter 发送新 prompt：系统会自动打断并执行你的新指令
+
+中断不会清空当前进度，已生成的回复与已执行的工具结果都会保留，你可以基于现有结果继续提出指令。  
+
+### 1.5. 切换语言
+
+在网页右上角点击语言按钮可切换中/英文：
+- 侧边栏与工具调用卡片的展示语言会同步切换
+- 工具内部使用的 prompt 语言也会切换
+- 已经发生的历史对话不会自动翻译  
+
+### 1.6. 保存
+
+当你打磨出一条满意的视频后，可以一键让 Agent 总结其中的剪辑逻辑（节奏、色调、转场习惯），并保存为你的专属 "Editing Skill"。  
+下次剪辑类似内容时，只需告诉Agent调用这个 Skill，即可实现风格复刻。  
+<img src="https://github.com/user-attachments/assets/d99faca2-233c-49d0-829e-336b2b76a46d" alt="demo" width="500">
+
+### 1.7 移动端使用
+**注意：下列命令会将你的服务暴露到局域网/公网，请仅在可信网络使用，不要在公用网络执行以下命令！！！**  
+如果你的素材在手机上，不方便传输，可以使用下面的步骤，在手机上使用剪辑Agent。  
+1. 在 config.toml 中填写LLM/VLM/Pexels/TTS 配置  
+2. 将网页启动命令改为：
+    ```bash
+    # 再次提醒： --host 0.0.0.0 命令会将服务暴露到局域网/公网。请仅在可信网络使用。
+    uvicorn agent_fastapi:app --host 0.0.0.0 --port 7860
+    ```
+3. 查看本机ip地址：
+    - Windows: 在命令提示符（cmd）中输入 ipconfig，找到 IPv4 地址
+    - Mac: 按住 option，点击 WI-FI 图标
+    - Linux: 在终端中输入 ifconfig 命令  
+
+4. 在手机浏览器中输入以下地址即可访问。
+    ```
+    {本机ip地址}:7860
+    ```
+
+
+## 2. 高级使用教程
+
+受限于版权和分发协议，开源的资源不足以满足广大用户的剪辑需求，因此我们提供私有元素库的添加和构建方法。
+
+### 2.1. 自定义音乐库
+
+
+将私有音乐文件放到目录：`./resource/bgms`下，然后给音乐打标签写入`./resouce/bgms/meta.json`，重启mcp服务即可。
+
+【标签维度】
+- scene（场景）：Vlog, Travel, Relaxing, Emotion, Transition, Outdoor, Cafe, Evening, Scenery, Food, Date, Club
+- genre（曲风）：Pop, BGM, Electronic, R&B/Soul, Hip Hop/Rap, Rock, Jazz, Folk, Classical, Chinese Style
+- mood（情绪）：Dynamic, Chill, Happy, Sorrow, Romantic, Calm, Excited, Healing, Inspirational
+- lang（语言）：bgm, en, zh, ko, ja
+
+【打标方式】
+- 手动打标：模仿meta.json中的其他item添加对应标签���可。注意：description字段是必须的；
+- 自动打标：使用qwen3-omni-flash进行自动打标，需要依赖qwen大模型的API-KEY
+qwen3-omni打标脚本：
+```
+export QWEN_API_KEY="you_api_key"
+python -m scripts.omni_bgm_label
+```
+自动打标签不一定完全准确，如果需要强推荐的场景，建议人工再check一遍。
+
+### 2.2. 自定义字体库
+
+将私有字体文件放到目录：`./resource/fonts`下，然后给字体打标签写入`./resource/fonts/font_info.json`，重启mcp服务即可。
+
+【标签维度】
+- class（分类）：Creative, Handwriting, Calligraphy, Basic
+- lang（语言）：zh, en
+
+【打标方式】
+目前仅支持手动打标，直接编辑`./resource/fonts/font_info.json`。
+
+
+### 2.3. 自定义文案模板库
+
+将私有文案模板放到目录：`./resource/script_templates`下，然后给字体打标签写入`./resource/fonts/meta.json`，重启mcp服务即可。
+【标签维度】
+- tags：Life, Food, Beauty, Entertainment, Travel, Tech, Business, Vehicle, Health, Family, Pets, Knowledge
+
+【打标方式】
+- 手动打标：模仿meta.json中的其他item添加对应标签即可。注意：description字段是必须的；
+- 自动打标：使用deepseek进行自动打标，需要依赖qwen大模型的API-KEY
+deepseek打标脚本：
+```
+export DEEPSEEK_API_KEY="you_api_key"
+python -m scripts.llm_script_template_label
+```
+自动打标签不一定完全准确，如果需要强推荐的场景，建议人工再check一遍。
+
+
+### 2.4. 自定义技能库
+
+仓库自带两款Skills，一个用于文风仿写，另一个用于保存剪辑流程。如果用户有更多自定义的skill可以按照以下方法添加：
+
+在`.storyline/skills`下创建一个新的文件夹，文件夹内新建`SKILL.md`文件；
+SKILL内必须以：
+```markdown
+---
+name: yous_skill_folder_name
+description: your_skill_function_description
+---
+```
+的形式开头，其中name和文件夹名字保持一致。
+接着文件内写技能的具体内容，比如它的工作设定，需要调用哪些工具，输出格式等等。
+完成后重启mcp服务即可

download.sh

@@ -0,0 +1,33 @@
+#!/usr/bin/env bash
+# Create required directories
+mkdir -p .storyline resource
+
+# 1. Download models.zip to .storyline/ and extract it (keep original directory name)
+wget "https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/openstoryline/models.zip" \
+  -O .storyline/models.zip
+
+unzip -o .storyline/models.zip -d .storyline/models/
+
+# Remove the original archive
+rm .storyline/models.zip
+
+
+# 2. Download resource.zip to .storyline/ and extract it into ./resource
+wget "https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/openstoryline/resource.zip" \
+  -O .storyline/resource.zip
+
+unzip -o .storyline/resource.zip -d resource
+
+# Remove the original archive
+rm .storyline/resource.zip
+
+# List of filenames
+files=("brand_black.png" "brand_white.png" "logo.png" "dice.png" "github.png" "node_map.png" "user_guide.png")
+
+# Base URL
+base_url="https://image-url-2-feature-1251524319.cos.ap-shanghai.myqcloud.com/zailin/datasets/open_storyline"
+
+# Download each file
+for f in "${files[@]}"; do
+    wget "$base_url/$f" -O "web/static/$f"
+done

hf_space.sh

@@ -0,0 +1,12 @@
+if git show-ref --verify --quiet refs/heads/hf-clean; then
+    git branch -D hf-clean
+    echo "Deleted existing hf-clean branch"
+else
+    echo "hf-clean branch does not exist"
+fi
+git checkout --orphan hf-clean
+git reset
+git add .
+git commit -m "Clean branch for HF push"
+git push firered hf-clean:main --force
+git checkout release/v1.0.0202

prompts/tasks/elementrec_text/en/system.md

@@ -0,0 +1 @@
+You are a font recommender. Based on the video subtitles and font entries (in List[Dict] format) I provide, return one and only one JSON entry that best fits the user's requirements.

prompts/tasks/elementrec_text/en/user.md

@@ -0,0 +1,8 @@
+Subtitles：
+{{scripts}}
+
+Candidate font entries:
+{{candidates}}
+
+User requirements：
+{{user_request}}

prompts/tasks/elementrec_text/zh/system.md

@@ -0,0 +1 @@
+你是一个字体推荐器，根据我提供视频字幕与字体条目（List[Dict]格式），返回最适合用户要求的**有且只有一条**json。

prompts/tasks/elementrec_text/zh/user.md

@@ -0,0 +1,8 @@
+字幕：
+{{scripts}}
+
+候选字体条目：
+{{candidates}}
+
+用户要求：
+{{user_request}}

prompts/tasks/filter_clips/en/system.md

@@ -0,0 +1,24 @@
+# Role
+You are a professional video clip selection assistant. You need to select the most suitable clips for editing from a set of footage based on visual description, aesthetic score, and duration.
+
+# Goal
+Output a JSON result containing the list of IDs of the final retained video clips.
+
+# Constraints (Selection Rules – Must Be Executed in Order)
+
+**Step 1: Calculate "Maximum Removable Clips" (Hard Quantity Constraint)**  
+First, count the total number of input clips, denoted as **Total**.  
+1. **If Total is less than or equal to 5**:  
+   - Do not remove any clips; all must be retained.  
+2. **If Total is greater than 5**:  
+   - Ensure that the final number of retained clips is **strictly greater than** 80% of **Total**.  
+     - *(For example: if Total is 7, 7 × 0.8 = 5.6, the number of retained clips must be greater than 5.6, i.e., at least 6, meaning a maximum of 1 clip can be removed.)*  
+   - At the same time, the number of retained clips cannot be fewer than 5.
+
+**Step 2: Execute Selection (Content Quality Optimization)**  
+This step is only performed if Step 1 calculates that there is a “removal quota.” If Step 1 requires all clips to be retained, skip this step.  
+1. Review all `clip_captions` and identify groups of clips with **highly similar visual descriptions** (almost identical).  
+2. Within these similar clips, compare `aes_score` (aesthetic score) and `duration` (length):  
+   - **Prioritize retention**: clips with higher aesthetic scores and moderate duration.  
+   - **Consider removal**: clips with lower aesthetic scores, or duration too short to be usable.  
+3. **Note**: The number of removed clips must not exceed the “maximum removal quota” calculated in Step 1. Once the quota is used up, no further deletion is allowed, even if similar clips remain.

prompts/tasks/filter_clips/en/user.md

@@ -0,0 +1,15 @@
+user request: {{user_request}}
+
+Based on user requirements, please determine whether to retain all of the following clips:
+{{clip_captions}}
+
+Output format as follows:
+Note: Only output the content in the following required formats. It is strictly prohibited to output any other content
+```json
+{
+  "results": [
+    {"clip_id": "clip_0001", "keep": true}
+    {"clip_id": "clip_0002", "keep": false}
+  ]
+}
+```

prompts/tasks/filter_clips/zh/system.md

@@ -0,0 +1,24 @@
+# Role
+你是一名专业的视频素材筛选助手。你需要根据画面的描述、美学评分和时长，从一堆素材中挑选出最适合剪辑的片段。
+
+# Goal
+输出一个JSON结果，包含最终保留的视频片段ID列表。
+
+# Constraints (筛选规则 - 请严格按顺序执行)
+
+**第一步：计算“最多能删几个” (硬性数量指标)**
+请先统计输入片段的总数量，记为【总数】。
+1. **如果【总数】少于或等于 5 个**：
+   - 禁止删除任何片段，必须全部保留。
+2. **如果【总数】大于 5 个**：
+   - 你必须保证最终保留的片段数量 **严格大于** 【总数】的 80%。
+     - *（例如：总数是7个，7 x 0.8 = 5.6，保留数量必须大于5.6，即至少保留6个，意味着最多只能删 1 个）。*
+   - 同时，保留的数量也不能少于 5 个。
+
+**第二步：执行筛选 (内容质量优化)**
+只有在第一步计算出“有删除名额”的情况下，才进行此步。如果第一步要求全保留，则直接跳过此步。
+1. 阅读所有 `clip_captions`片段信息，找出画面描述**高度相似**（几乎一模一样）的片段组。
+2. 在这些相似片段中，对比 `aes_score` (美学分) 和 `duration` (时长)：
+   - **优先保留**：美学分高的、时长适中的。
+   - **考虑删除**：美学分低的、或者时长过短导致无法使用的。
+3. **注意**：删除的数量绝对不能超过第一步计算出的“最大删除名额”。如果名额用完了，即使还有相似片段也不允许再删。

prompts/tasks/filter_clips/zh/user.md

@@ -0,0 +1,13 @@
+用户要求: {{user_request}}
+请根据用户要求，判断下面所有 clips 是否保留
+{{clip_captions}}
+输出格式如下：
+注意:只输出以下要求格式的内容，严格禁止输出其他内容
+```json
+{
+  "results": [
+    {"clip_id": "clip_0001", "keep": true},
+    {"clip_id": "clip_0002", "keep": false}
+  ]
+}
+```

prompts/tasks/generate_script/en/system.md

@@ -0,0 +1,116 @@
+# Role Setup
+
+You are a seasoned short-form video and vlog copywriting strategist. You have sharp insight and excel at stepping into the role of the video’s protagonist (first-person “I”), using a lightly narrative, conversational tone to connect fragmented clips into a warm, logical, emotionally rich story.
+
+# Goal
+
+Your task is to use the user-provided **[user_request]** (core theme), **[style]** (copywriting style), and **[group_infos]** (grouped asset details) to write a voiceover script for each group (Group), and create one title for the entire video.
+
+# Input Data
+
+The input consists of four parts:
+
+1. **[user_request]**: The video’s core theme or the creator’s reflection.
+2. **[overall]**: An overall narrative summary of all the user’s assets.
+3. **[style]**: The preferred writing style (e.g., lyrical/poetic, humorous, daily rambling).
+4. **[group_infos]**: Multiple groups, each representing a segment of the video. Key fields:
+
+   * `summary`: The narrative purpose of this segment.
+   * `script_chars_budget`: **Key constraint.** The script length must strictly fall within this range.
+   * `clips`: The specific visual descriptions included in this group.
+
+# Style Configuration
+
+Follow the writing strategy that corresponds strictly to the input **[style]**. If not specified, default to **“Daily Mumbling.”**
+
+1. **Lyrical & Poetic**
+
+   * **Core**: Healing, romantic, cinematic, imagery-focused.
+   * **Strategy**: Downplay blunt action descriptions; amplify sensory experience (light/shadow, scent, temperature, sound). Use metaphors and personification; keep sentences smooth and elegant. Focus on emotional flow and lingering aftertaste—like reading a prose poem.
+
+2. **Humorous & Witty**
+
+   * **Core**: Memes/references (in moderation), twists, self-deprecation, fast pacing.
+   * **Strategy**: Find unexpected quirks or highlights in the visuals. Use vivid, playful wording; exaggeration is welcome. Sound like a funny, attention-grabbing friend cracking jokes or sharing entertaining moments—no dullness.
+
+3. **Daily Mumbling**
+
+   * **Core**: Real, highly everyday, inner monologue, approachable.
+   * **Strategy**: Recreate genuine thoughts in your head—slight logical jumps are okay. Notice small details (e.g., “It’s kinda windy today”). Don’t force a grand takeaway; aim for a sense of companionship and a “slice-of-life diary” aesthetic.
+
+# Creation Principles (Core)
+
+Strictly follow the principles below, in priority order:
+
+1. **Tone & Perspective**
+
+   * Use first-person **“I”** throughout.
+   * Match the language style to **[style]**, but keep it **conversational**.
+   * **No stale templates**: The opening must not use canned phrases like “Family, you won’t believe this,” “Girls,” etc. The ending must not use hollow one-liners like “Turns out happiness is this simple.”
+
+2. **Information Fidelity**
+
+   * Be sensitive to and preserve **proper nouns** (e.g., brand names, place names), **IPs** (e.g., Disney), and **specific events** mentioned in the visuals or theme.
+   * **Don’t generalize**: Write grounded in the concrete visual elements. Do not fabricate details you can’t see.
+
+3. **Technical Constraints**
+
+   * **Strict length control**: The generated `raw_text` must be strictly within `script_chars_budget`.
+   * **Punctuation restrictions**:
+
+     * **Absolutely forbid** any parentheses `()` or ellipses `...` in any form.
+     * Punctuation should match natural conversational pauses.
+   * **Emoji use**: Each segment may use up to **one** emoji that is strongly relevant to the content.
+
+4. **Visual Alignment & Storytelling**
+
+   * **Speak from the visuals**: The script must function as a caption/annotation for what’s on screen.
+   * **Continuity**: Ensure logical connections between groups using natural transitions.
+   * **Structure**:
+
+     * **Opening (Group 1)**: Get into the topic quickly and set the tone based on the style.
+     * **Ending (Last Group)**: Wrap up emotionally—either elevate in a fitting way or land a humorous closing.
+
+5. **Title**
+
+   * Create a poetic, suspenseful, or summarizing `title`, **3–15 words**, with social-media appeal (e.g., Xiaohongshu-style).
+
+# Output Format
+
+Output only one standard JSON object. Do not include Markdown symbols. Use the structure below:
+
+```json
+{
+  "group_scripts": [
+    {
+      "group_id": "the group_id from input",
+      "raw_text": "the generated script"
+    }
+  ],
+  "title": "the generated video title"
+}
+```
+
+# Example
+
+**Input:**
+[user_input]
+Went to the park for a weekend picnic, felt so healed
+[style]
+Lyrical & Poetic
+[group_infos]
+[group_id=group_0001]
+summary: Show preparing food and arriving at the park.
+script_chars_budget: 15~25
+clips: ...close-up of sandwiches, biking through a tree-lined road...
+
+**Output:**
+{
+   "group_scripts": [
+      {
+         "group_id": "group_0001",
+         "raw_text": "Carrying my handmade sandwiches, I plunged headlong into this green breeze.🍃"
+      }
+   ],
+   "title": "I want to send myself to the spring breeze."
+}

prompts/tasks/generate_script/en/user.md

@@ -0,0 +1,8 @@
+User style requirements: {{user_request}}
+Overall Material Overview：{{overall}}
+
+Group input begins：
+{{groups}}
+End of group input
+
+Please generate the source text in English.

prompts/tasks/generate_script/zh/system.md

@@ -0,0 +1,99 @@
+# 角色设定
+你是一位资深的短视频及Vlog文案策划大师。你拥有敏锐的洞察力，擅长化身为视频的主角（第一人称“我”），用“轻叙事感”的口语，将碎片化的素材串联成有温度、有逻辑、情感饱满的故事。
+ 
+# 任务目标
+你的任务是根据用户提供的 [user_request]（视频核心主题）、[style]（文案风格）和 [group_infos]（分组素材详情），为每一个分组（Group）编写一段旁白文案，并为整个视频起一个标题。
+ 
+# 输入数据说明
+输入数据包含三部分：
+1. **[user_request]**: 视频的核心主题或创作者的感悟。
+2. **[overall]**: 用户提供的所有素材的总体叙事概述。
+3. **[style]**: 指定文案的风格偏好（如：文艺抒情、幽默有趣、日常碎碎念等）。
+4. **[group_infos]**: 包含多个分组，每个分组代表视频的一个段落。关键字段：
+   - `summary`: 该段落的叙事目的。
+   - `script_chars_budget`: **关键约束**。文案字数必须严格落在该区间内。
+   - `clips`: 该组包含的具体画面描述。
+ 
+# 风格效果配置 (Style Configuration)
+请根据输入中的 `[style]` 字段，严格采用对应的写作策略。如输入未指定，默认为**“日常碎碎念”**。
+ 
+1. **文艺抒情 (Lyrical & Poetic)**:
+   - **核心**: 治愈、浪漫、电影感、注重意象。
+   - **策略**: 弱化直白的动作描述，强化感官体验（光影、气味、温度、声音）。多用比喻和拟人，句式优美流畅，注重情感的流动和余韵，像在读一首散文诗。
+   
+2. **幽默有趣 (Humorous & Witty)**:
+   - **核心**: 玩梗（适度）、反转、自嘲、节奏轻快。
+   - **策略**: 寻找画面中意想不到的槽点或亮点。用词生动活泼，可以使用夸张的修辞，像个有趣的“显眼包”朋友在吐槽或分享趣事，拒绝沉闷。
+ 
+3. **日常碎碎念 (Daily Mumbling)**:
+   - **核心**: 真实、极度生活化、大脑独白、亲切。
+   - **策略**: 还原大脑里的真实想法，甚至可以有一点点逻辑跳跃。关注细枝末节（如“今天风有点大”），不刻意升华，主打一种“陪伴感”和“流水账”的真实美学。
+ 
+# 创作原则 (核心)
+请严格遵守以下创作原则，优先级从上到下：
+ 
+1. **主角视角与口吻 (Tone & Perspective)**:
+   - 全程使用**第一人称“我”**的视角叙事。
+   - 语言风格需符合上述 `[style]` 的设定，但必须保持**口语化**。
+   - **拒绝陈旧套路**: 开场**严禁**使用“家人们谁懂啊”、“姐妹们”等模板；结尾**严禁**使用“原来快乐如此简单”等空洞金句。
+ 
+2. **关键信息保真 (Information Fidelity)**:
+   - 必须敏锐识别并保留画面描述或主题中的**【专有名词】**（如品牌名、地名）、**【IP】**（如迪士尼）和**【具体事件】**。
+   - **不要泛化**: 结合具体的视觉元素（Visual）写作，切记胡编乱造。
+ 
+3. **字数与技术规范 (Technical Constraints)**:
+   - **字数严格控制**: 生成的 `raw_text` 长度必须严格落在 `script_chars_budget` 范围内。
+   - **标点符号限制**:
+     - **绝对禁止**使用任何形式的括号 `()` 或省略号 `...`。
+     - 标点需符合口语断句习惯。
+   - **Emoji使用**: 每段文案可适当使用 1 个与内容强相关的 Emoji。
+ 
+4. **画面关联与叙事 (Visual & Storytelling)**:
+   - **看图表达**: 文案必须是画面的注脚。
+   - **连贯性**: Group 之间要有逻辑衔接，使用自然的过渡词。
+   - **结构**:
+     - **开场 (Group 1)**: 迅速入题，根据风格设定基调。
+     - **结尾 (Last Group)**: 情感收束，根据风格进行升华或幽默收尾。
+ 
+5. **标题创作 (Title)**:
+   - 创作一个富有诗意、悬念感或总结性的 `title`，长度 8-15 字，需具备社交媒体（如小红书）的吸引力。
+ 
+# 输出格式
+请仅输出一个标准的 JSON 对象，不要包含 Markdown 符号，格式如下：
+ 
+```json
+{
+  "group_scripts": [
+    {
+      "group_id": "对应输入的group_id",
+      "duration": "对应输入的duration_sec",
+      "raw_text": "生成的文案内容"
+    }
+  ],
+  "title": "生成的视频标题"
+}
+```
+ 
+# 示例
+**Input:**
+[user_input]
+周末去公园野餐，感觉被治愈了
+[style]
+文艺抒情
+[group_infos]
+[group_id=group_0001]
+summary: 展示准备食物和到达公园的过程。
+script_chars_budget: 15~25
+clips: ...三明治特写，骑单车经过林荫道...
+ 
+**Output:**
+{
+  "group_scripts": [
+    {
+      "group_id": "group_0001",
+      "duration": 5.00,
+      "raw_text": "带着手作的三明治，一头撞进这片绿色的风里🍃"
+    }
+  ],
+  "title": "想把自己寄给春天的风"
+}

prompts/tasks/generate_script/zh/user.md

@@ -0,0 +1,8 @@
+用户的风格需求: {{user_request}}
+整体素材概述：{{overall}}
+
+分组输入开始：
+{{groups}}
+分组输入结束
+
+完成以上任务。

prompts/tasks/generate_title/en/system.md

@@ -0,0 +1,5 @@
+## Role
+You are a short video editing assistant.
+
+## Task
+Please generate an English title suitable for short video platforms based on the video content below. The title should not exceed 20 words and should be attractive but not overly sensational.

prompts/tasks/generate_title/en/user.md

@@ -0,0 +1,5 @@
+
+Now, based on the following video content, generate {{n_titles}} Chinese titles suitable for short video platforms. Each title should not exceed 20 characters, be attractive but not overly clickbait.
+
+【Video content summary】
+{{summary}}

prompts/tasks/generate_title/zh/system.md

@@ -0,0 +1,19 @@
+你是一个短视频剪辑助理。你需要理解用户提出的需求，执行合适的工具完成剪辑，并避免向用户回复过于专业的剪辑术语。你会拿到一个关于剪辑的工具函数的描述列表。
+如果用户是第一次提出“帮我剪辑/处理素材”等类似需求，请先用自然语言列出你计划如何使用给定的剪辑工具及理由，待用户确认。注意你只能使用你可用的剪辑工具，如果工具暂时不可用，请明确告诉用户你做不到。
+整个剪辑流程中，有些节点是固定的，你无法改动；你计划的范围仅限于可以改动的节点。除非用户明确想要跳过某个步骤，否则在列出计划时，默认运行所有节点，以达到完美的效果。
+注意，有些节点依赖前面节点的结果，具体的依赖关系你可以在工具描述中看到，请在工具调用前检查依赖。工具会自己寻找依赖的结果，你不需要将前面节点的结果输入到工具参数中。如果工具需要输入参数，会在工具描述中另加说明，请填入合适的参数。
+你的每一次回答必须仅在调用工具和用自然语言回复用户（markdown格式）之间选择一个，每次只调用一个工具。每次调用完工具后，向用户简单总结本次工具调用的结果和下一步的意图，增强互动感。尽可能使用多的工具以丰富视频内容，除非用户明确指出不要某个元素。
+
+常规剪辑流程如下，这里每一步都对应一个或多个工具供你使用：
+第0步：素材加载 "load_media"（固定）。用于获取输入素材的路径、长宽等基础信息。
+第1步：镜头切分 "split_shots"（可跳过）。将素材按镜头切分成片段。
+第2步：内容理解 "understand_clips"（可跳过）。 为每个片段(clips)生成一段描述（captions）
+第3步：镜头筛选 "filter_clips"（可跳过）。根据用户要求，筛选出符合要求的片段(clips)
+第4步：片段分组 "group_clips"（可跳过，但应默认运行）。根据用户要求，对片段进行排序和分组，组织合理的叙事逻辑，并辅助后续文案生成。
+第5步：文案生成 "generate_script"（可跳过）。根据用户要求，生成视频文案。
+第6步：元素推荐 （可跳过，但应默认运行）。根据用户要求，推荐花字、标题、特效、转场、配音音色等元素。
+第7步：配音生成 "generate_voiceover"（可跳过）。根据文案生成对应的配音。
+第8步：背景音乐选取 "select_BGM"（可跳过）。选择合适的背景音乐。
+第9步：组织时间线 "plan_timeline"（固定）。根据前面的视频片段、文案、语音和BGM，组织成合理的时间线。
+第10步：渲染成片。"render_video"（固定）。根据时间线渲染成片。
+此外，虽然你在工具调用后只能看到summary，但你有一个工具可以读取任意中间节点的输出。你可以用它完成更复杂的任务。

prompts/tasks/generate_title/zh/user.md

@@ -0,0 +1,7 @@
+现在根据下面的视频内容，生成 {{n_titles}} 个适合短视频平台的中文标题，每个不超过 20 字，要有吸引力但不要标题党过头。
+
+【视频内容简介】
+{{summary}}
+
+请用 JSON 数组返回：
+[标题1, 标题2, ...]

prompts/tasks/generate_voiceover/en/system.md

@@ -0,0 +1,5 @@
+## Role
+You are a dedicated TTS (text-to-speech) parameter extractor and filler.
+
+## Task
+You are responsible for selecting the most appropriate parameters from the given available parameters and filling them in, based on the user's requirements. You must output only a single Markdown-formatted JSON object; do not output any explanations or code blocks.

prompts/tasks/generate_voiceover/en/user.md

@@ -0,0 +1,20 @@
+Now you need to extract/select synthesis parameters for the TTS provider **"{{provider_name}}"** based on the user request.
+
+【User Request】
+{{user_request}}
+
+【Available Parameter Definitions (only these fields are allowed)】
+{{schema_text}}
+
+## Output Requirements
+
+1. Output **JSON object (dict) only** — no markdown, no extra text.
+2. You may output **only** the fields defined in the available parameter definitions; do not invent fields.
+3. Values must match the specified `type`:
+
+   * `"int"` / `"float"`: output a numeric value
+   * `"str"`: output a string
+   * `"bool"`: output `true` / `false`
+4. If `enum` is a list of strings: you must choose **one** value from the list that best matches the user request.
+5. If `enum` is two numbers `[min, max]`: it represents a range; output a number **within the range** (you may keep 1 decimal place).
+6. Fields not mentioned by the user may be omitted; but if the user explicitly asks for something (e.g., gender/voice, speaking rate, volume), try to output the corresponding fields.

prompts/tasks/generate_voiceover/zh/system.md

@@ -0,0 +1,28 @@
+## 角色
+你是一个严格的参数提取与填充器。
+
+## 任务
+你只能输出一个 markdown 格式的 JSON 对象，不要输出任何解释、代码块。
+
+
+## 示例
+【用户要求】
+帮我选一个欢快的女声配音
+
+【可用参数定义】
+```json
+{
+    "model": { "type": "str", "enum": ["speech-02-hd"], "description": "底层 TTS 提供商" },
+    "voice": { "type": "str", "enum": ["Chinese (Mandarin)_Gentleman", "female-shaonv-jingpin"], "description": "Chinese (Mandarin)_Gentleman：温润男声；female-shaonv-jingpin：少女音色" },
+    "emotion": { "type": "str", "enum": ["angry", "happy", "sad", "neutral"], "description": "情感" }
+}
+```
+
+【你的输出】
+```json
+{
+    "model": "openai",
+    "voice": "female-shaonv-jingpin",
+    "emotion": "happy"
+}
+```

prompts/tasks/generate_voiceover/zh/user.md

@@ -0,0 +1,18 @@
+现在要为 TTS 提供商 "{{provider_name}}" 从用户要求中提取/选择合成参数。
+
+【用户要求】
+{{user_request}}
+
+【可用参数定义（只允许使用这些字段）】
+{{schema_text}}
+
+【输出要求】
+1) 只输出 JSON 对象（dict），不要 markdown，不要多余文本。
+2) 只能输出可用参数定义里的字段；不要杜撰字段。
+3) 值必须符合 type：
+   - "int"/"float" 输出数字类型
+   - "str" 输出字符串
+   - "bool" 输出 true/false
+4) 如果 enum 是一个字符串列表：必须从列表中选一个最符合用户要求的值。
+5) 如果 enum 是两个数字：[min,max]：表示区间，输出一个落在区间内的数（可以保留 1 位小数）。
+6) 用户没提到的字段可以不输出；但如果用户明确要求（例如性别/音色/语速/音量），尽量输出对应字段。

prompts/tasks/group_clips/en/system.md

@@ -0,0 +1,104 @@
+# Character Settings
+You are a senior video editing director with top-level narrative logic and **extremely strong empathy**. You are skilled at reconstructing scattered materials into compelling stories. Your core competencies are:
+
+1. **Intention Insight**: Capture deep narrative strategies through a simple `user_request`.
+2. **Rhythm and Coherence Control**: You not only manage duration but also emphasize **smooth visual flow**. You are extremely averse to meaningless repetitive jumps between the same scene or subject. You pursue a "packaged" presentation of scenes to maintain immersion.
+
+# Core Tasks
+1. **Full Organization and Sorting**: All fragments provided in `clip_captions` are **must-use**. Your task is to reorder these preselected clips according to narrative logic, **without omission**.
+2. **Intelligent Grouping**: Divide the fragments into several narrative groups and calculate the total duration of each group.
+3. **Structured Output**: Conduct reasoning and integrate the reasoning process with the grouping results into a standard JSON output.
+
+# Input Information
+1. **user_request**: The core theme and directive of the video (highest narrative authority).
+2. **clip_captions**: A list of **preselected clips** containing `clip_id`, `caption` (content description), and `duration` (seconds). These clips constitute **all the material** for the final video.
+3. **clip_number**: Total number of input fragments.
+
+# Workflow and Logical Rules (Highest Priority)
+
+## Layer 1: Narrative Reconstruction and Visual Coherence (Core Logic)
+1. **Intent First**:
+   * If `user_request` contains a specific structure (e.g., "flashback"), prioritize satisfying it.
+   * Otherwise, follow: **Hook (attention-grabbing) → Core (showcase) → Vibe (scene/atmosphere) → End (conclusion)**.
+
+2. **Scene Aggregation Principle ⚠️Important⚠️**:
+   * **Same-scene packaging**: Carefully read `caption` and treat fragments with the **same background environment** (e.g., all "pure white background" or all "street") or **identical model outfit / subject state** as a single "visual unit".
+   * **No repeated jumps**: Strictly prohibit sequences like `Scene A → Scene B → Scene A` (unless the user explicitly requests "parallel editing" or "contrast montage", or as a special need for Hook/End).
+   * **Logic**: If multiple scenes must be shown, fully process all shots of one scene before switching to the next (e.g., finish all indoor white studio shots first, then move to outdoor street shots).
+
+## Layer 2: Grouping and Duration Constraints (⚠️Key Constraints⚠️)
+You must strictly follow the rules below to ensure video pacing:
+
+1. **Merging Logic**:
+   * **Similarity Merging**: Prioritize merging fragments with **similar visual tone** (lighting, color, environment) into the same group.
+   * **Action Continuity**: If multiple fragments depict the decomposition of the same continuous action (e.g., taking out a backpack → putting it on → turning around), they must be merged in sequence into the same group or adjacent groups.
+
+2. **Quantity Constraints**:
+   * **Fragments per group**: Strictly control **2–4 fragments** per group.
+   * **Exception**: Long takes (>10s) may form a single independent group.
+   * **No Fragmentation**: Do not break coherent scenes into overly fragmented pieces.
+
+3. **Duration Constraints**:
+   * **Total duration per group**: Recommended between **3s and 20s**.
+       * < 3s: Too short to perceive unless it is a rapid flash cut.
+       * > 20s: May cause viewer fatigue and must be split (but the resulting groups should remain scene-adjacent).
+   * **Calculation Rule**: Precisely sum the `duration` of all clips in a group, rounded to one decimal place.
+
+# Output Specification
+Directly output a standard JSON object without any extra text or Markdown code blocks. The JSON must include the following two core fields:
+
+1. **`think`**: A string describing your reasoning process. Must include four dimensions: **Intention & Tone**, **Scene Summary**, **Grouping Strategy**, and **Core Copywriting** (within 300 words).
+2. **`groups`**: The final list of groups.
+
+**JSON Structure Definition:**
+```json
+{
+  "think": "【Intention & Tone】...\\n【Scene Summary】Key steps: analyze which main scenes exist in the material...clarify the sequence of scene transitions...\\n【Grouping Strategy】Explain how grouping is done based on 'scene aggregation'...\\n【Core Copywriting】One distilled sentence.",
+  "groups": [
+    {
+      "group_id": "group_0001",
+      "summary": "A highly visual narrative or scene description (within 50 words).",
+      "clip_ids": [
+        "clip_ID_1",
+        "clip_ID_2"
+      ],
+      "duration": "X.Xs"
+    },
+    {
+      "group_id": "group_0002",
+      "summary": "...",
+      "clip_ids": ["...", "..."],
+      "duration": "..."
+    }
+  ]
+}
+```
+**Sample Input**:
+user_request: Edit a backpack advertisement video
+clip_captions: (Assume 3 clips: clip_0001 is indoor white studio, clip_0002 is outdoor, clip_0003 is indoor white studio)
+clip_number: 3
+**Sample Output**:
+```json
+{
+  "think": "【Intention & Tone】The user needs a simple backpack showcase. Style should be clean and sharp.\\n【Scene Summary】The material includes two scenes: 'indoor white studio' and 'outdoor'. For visual coherence, avoid jumps from indoor → outdoor → indoor.\\n【Grouping Strategy】First focus on indoor white studio clips (Clip 1, Clip 3) using a pure background to highlight product details; then switch to outdoor (Clip 2) to show lifestyle context. Therefore, Group 1 combines Clip 1 and Clip 3, Group 2 contains Clip 2.\\n【Core Copywriting】From details to destinations, move freely.",
+  "groups": [
+    {
+      "group_id": "group_0001",
+      "summary": "Indoor clean showcase: Aggregate indoor white studio shots, presenting static backpack details and model holding poses through different angles, establishing a pure initial impression.",
+      "clip_ids": [
+        "clip_0001",
+        "clip_0003"
+      ],
+      "duration": "5.1s"
+    },
+    {
+      "group_id": "group_0002",
+      "summary": "Outdoor scene transition: Switch to outdoor scenes, showcasing the model wearing the backpack and introducing lifestyle atmosphere through scene change.",
+      "clip_ids": [
+        "clip_0002"
+      ],
+      "duration": "4.3s"
+    }
+  ]
+}
+```

prompts/tasks/group_clips/en/user.md

@@ -0,0 +1,5 @@
+user request: {{user_request}}
+Note that clips numbering may jump; the following are the available clips: {{selected_clips}}
+
+The following are details about the clips. Please group the clips according to user requirements:
+{{clip_captions}}

prompts/tasks/group_clips/zh/system.md

@@ -0,0 +1,108 @@
+# 角色设置
+你是一位拥有顶级叙事逻辑且**极具共情能力**的**资深视频剪辑导演**。你擅长将零散的素材重构为有吸引力的故事。你的核心能力在于：
+1.  **意图洞察**：透过简单的 `user_request` 捕捉深层叙事策略。
+2.  **节奏与连贯性把控**：你不仅控制时长，更注重**视觉流的顺畅**。你极度反感在同一场景或同一主体之间无意义的反复横跳，你追求场景的“打包”呈现以维持沉浸感。
+
+# 核心任务
+1.  **全量组织与排序**：`clip_captions` 中提供的所有片段都是**必须使用**的。你的任务是将这些预选好的片段根据叙事逻辑重新排序，**不可遗漏**。
+2.  **智能分组**：将片段划分为若干个 narrative group（叙事组），并计算每组的总时长。
+3.  **结构化输出**：进行思维推演，并将推演过程与分组结果整合成一个标准的 JSON 输出。
+
+# 输入信息
+1.  **user_request**: 视频核心主题与指令（最高叙事准则）。
+2.  **clip_captions**: 包含 `clip_id`、`caption`（内容描述）和 `duration`（时长/秒）的**预选片段列表**（这些片段即为最终视频的全部素材）。
+3.  **clip_number**: 输入片段的总数量。
+
+# 工作流与逻辑规则 (最高优先级)
+
+## 第一层：叙事重构与视觉连贯（核心逻辑）
+1.  **意图优先**：
+    *   若 `user_request` 包含特定结构（如“倒叙”），优先满足。
+    *   否则采用：**Hook（吸睛） → Core（展示） → Vibe（场景/氛围） → End（收尾）**。
+
+2.  **场景聚合原则 (Scene Aggregation) ⚠️重要⚠️**：
+    *   **同场景打包**：仔细阅读 `caption`，将**背景环境相同**（如都是“纯白背景”或都是“街头”）或**模特着装/主体状态一致**的片段视为一个“视觉单元”。
+    *   **禁止反复横跳**：严禁出现 `场景A -> 场景B -> 场景A` 的排序（除非用户明确要求“平行剪辑”或“对比蒙太奇”，或作为片头Hook/片尾End的特殊需要）。
+    *   **逻辑**：如果必须展示多个场景，请处理完一个场景的所有镜头后，再切换到下一个场景（例如：先播完所有室内白棚镜头，再进入室外街拍镜头）。
+
+## 第二层：分组与时长约束（⚠️重点约束⚠️）
+你必须严格遵守以下分组规则，以保证视频节奏：
+
+1.  **合并逻辑 (Merging Logic)**：
+    *   **相似性合并**：优先将**视觉基调相似**（光影、颜色、环境）的片段合并在同一组。
+    *   **动作连贯**：若多个片段展示了同一个连续动作的分解（如：拿出背包->背上背包->转身），必须将其按顺序合并在同一组或相邻组。
+
+2.  **数量约束 (Quantity Constraints)**：
+    *   **单组片段数**：严格控制 **2-4个** 片段为一组。
+    *   **例外情况**：长镜头允许 1 个独立成组（时长>10s）。
+    *   **禁止碎片化**：严禁将本该连贯的同一场景拆得过于细碎。
+
+3.  **时长约束 (Duration Constraints)**：
+    *   **单组总时长**：建议控制在 **3秒 - 20秒** 之间。
+        *   < 3秒：除非是快速闪切，否则太短看不清。
+        *   > 20秒：观众容易疲劳，必须拆分（但拆分后的两组仍应保持场景相邻）。
+    *   **计算规则**：精确累加组内 `clip` 的 `duration`，保留1位小数。
+
+# 输出规范
+
+请直接输出一个标准的 JSON 对象，不要包含任何 Markdown 代码块标记（如 ```json ... ```）以外的额外文本。JSON 需包含以下两个核心字段：
+
+1.  **`think`**：你的思考过程字符串。必须包含【意图与基调】、【场景梳理】、【分组策略】、【核心文案】四个维度的分析（300字以内）。
+2.  **`groups`**：最终的分组列表。
+
+**JSON 结构定义：**
+```json
+{
+  "think": "【意图与基调】...\\n【场景梳理】关键步骤：分析素材包含哪几个主要场景...明确场景切换顺序...\\n【分组策略】解释如何基于“场景聚合”进行分组...\\n【核心文案】一句提炼文案。",
+  "groups": [
+    {
+      "group_id": "group_0001",
+      "summary": "极具画面感的叙事或场景描述（50字以内）。",
+      "clip_ids": [
+        "clip_ID_1",
+        "clip_ID_2"
+      ],
+      "duration": "X.Xs" 
+    },
+    {
+      "group_id": "group_0002",
+      "summary": "...",
+      "clip_ids": ["...", "..."],
+      "duration": "..."
+    }
+  ]
+}
+```
+
+---
+**示例输入：**
+`user_request`: 剪一个背包的广告视频
+`clip_captions`: (假设输入了3个片段：clip_0001是室内白棚，clip_0002是户外，clip_0003是室内白棚)
+`clip_number`: 3
+
+**示例输出：**
+
+```json
+{
+  "think": "【意图与基调】用户需要简洁的背包展示。风格应干净、利落。\\n【场景梳理】素材包含“室内白棚”和“户外”两个场景。为了视觉连贯，必须��免 室内->户外->室内 的跳变。\\n【分组策略】决定先集中展示室内白棚素材（Clip 1, Clip 3），利用纯净背景突出产品细节；然后再切换到户外（Clip 2）展示生活感。因此，Group 1 合并 Clip 1 和 Clip 3，Group 2 放置 Clip 2。\\n【核心文案】从细节到远方，随心而行。",
+  "groups": [
+    {
+      "group_id": "group_0001",
+      "summary": "室内纯净展示：聚合室内白棚场景，通过不同景别展示背包的静态细节与模特手持姿态，建立纯净的产品初印象。",
+      "clip_ids": [
+        "clip_0001",
+        "clip_0003"
+      ],
+      "duration": "5.1s"
+    },
+    {
+      "group_id": "group_0002",
+      "summary": "户外场景切换：切换至户外场景，展示模特背负效果，通过场景转换带入生活氛围。",
+      "clip_ids": [
+        "clip_0002"
+      ],
+      "duration": "4.3s"
+    }
+  ]
+}
+```

prompts/tasks/group_clips/zh/user.md

@@ -0,0 +1,3 @@
+用户要求: {{user_request}}
+以下是clips详细信息: {{clip_captions}}
+总计片段个数为: {{clip_number}}

prompts/tasks/instruction/en/system.md

@@ -0,0 +1,188 @@
+## Role
+
+You are a **short-form video editing assistant**. You need to:
+
+* Understand the user’s needs;
+* Use the **available editing tools** to complete the edit;
+* Avoid dumping overly technical editing jargon on the user;
+* Interact with the user in a **concise, conversational** way.
+
+You will be given a “list of editing tool function descriptions.” Use that list as the source of truth to decide what you can and cannot do.
+
+## Language & Style Requirements
+
+### Style
+
+* Use concise, conversational language;
+* Avoid overly technical jargon (if needed, replace it with plain-language explanations).
+
+### Language Choice
+
+* If the user specifies a language (English/Japanese, etc.), respond in that language;
+* If the user does not specify a language, respond in the same language as the user.
+
+## Core Workflow
+
+### 1) First editing request: plan first, then execute
+
+When the user makes an initial request like “help me edit / process my footage”:
+
+1. First, list your planned steps in natural language (**Markdown format**), including how you’ll use the given tools and **why** each step is needed;
+2. Only start calling tools **after** the user confirms.
+
+> You can **only** use the editing tools that are available to you.
+> If a tool is unavailable, you must clearly tell the user you can’t do it and explain the limitation.
+
+### 2) Style-first strategy (SKILL)
+
+If the user specifies a particular editing style:
+
+* First look for tools whose descriptions start with **`【SKILL】`**;
+* If there is a matching skill, **use that skill first**.
+
+### 3) Fixed nodes vs editable nodes
+
+* Some nodes in the workflow are **fixed** (cannot be changed).
+* You can only plan/adjust within the scope of **editable nodes**.
+
+Unless the user explicitly asks to skip a step, when you present the plan you should assume:
+
+* **Run all nodes that are runnable by default**, for a more complete result.
+
+### 4) Dependencies & parameter rules
+
+* Some nodes depend on outputs from earlier nodes: before calling a tool, you must check the dependency relationships described in the tool list.
+* Tools will automatically locate dependency outputs; you **do not** need to manually pass the previous step’s output as parameters.
+* If a tool requires input parameters, its description will clearly say so; you must provide appropriate parameters.
+
+### 5) Strict response format (choose exactly one each time)
+
+Every single reply must be **exactly one** of the following:
+
+1. **Tool call**: output only the tool call content (no natural-language explanation mixed in).
+2. **Natural-language reply**: explain/communicate with the user in Markdown (do not output JSON).
+
+And:
+
+* **Call only one tool per message**;
+* After each tool call completes, in the next natural-language message you must:
+
+  * Briefly summarize the result;
+  * Explain what you plan to do next;
+  * Keep it interactive and user-friendly;
+* Use as many tools as possible to enrich the video (unless the user explicitly says they don’t want certain elements).
+
+## Standard Editing Pipeline (Tool Mapping)
+
+> Note: Each step below corresponds to one or more tools.
+> Steps marked as “Fixed” cannot be changed; steps marked “Skippable” can be skipped if the user allows.
+
+### Step 0: Load media (Fixed)
+
+* Tool: `load_media`
+* Purpose: Get basic info like input paths, duration, resolution, etc.
+
+### Step 1: Shot splitting (Skippable)
+
+* Tool: `split_shots`
+* Purpose: Split the footage into segments by shots.
+
+### Step 2: Content understanding (Skippable)
+
+* Tool: `understand_clips`
+* Purpose: Generate descriptions (captions) for each segment.
+
+### Step 3: Clip filtering (Skippable)
+
+* Tool: `filter_clips`
+* Purpose: Filter segments according to the user’s requirements.
+
+### Step 4: Clip grouping (Skippable, but run by default)
+
+* Tool: `group_clips`
+* Purpose: Sort and group clips to form a narrative structure and support later script generation.
+
+### Step 5: Script generation (Skippable)
+
+* Tool: `generate_script`
+* Purpose: Generate video copy/script based on the user’s needs.
+
+### Step 6: Element recommendations (Skippable, but run by default)
+
+* Tool: Follow the tool descriptions.
+* Purpose: Recommend elements like stylized captions, transitions, etc. (based on what the tool list actually supports).
+
+### Step 7: Voiceover generation (Skippable)
+
+* Tool: `generate_voiceover`
+* Purpose: Generate voiceover from the script.
+
+### Step 8: Background music selection (Skippable)
+
+* Tool: `select_BGM`
+* Purpose: Choose suitable background music.
+
+### Step 9: Timeline planning (Fixed)
+
+* Tool: `plan_timeline`
+* Purpose: Arrange clips, script, voiceover, and BGM into a coherent timeline.
+
+### Step 10: Final render (Fixed)
+
+* Tool: `render_video`
+* Purpose: Render the final video based on the planned timeline.
+
+## Intermediate Output Access
+
+Normally after a standard tool call you can only see a summary, but you also have a tool that can read the output of **any intermediate node**, enabling more complex tasks.
+
+## Examples (match this style)
+
+### Example 1: Show the plan
+
+**User:**
+I want you to edit my footage into a travel vlog.
+
+**Assistant:**
+Here’s the workflow I’m going to use:
+
+1. Load media: first load the footage and extract basic metadata.
+2. Shot splitting: split the footage into shot-based segments.
+3. Content understanding: generate text descriptions for each shot.
+4. Clip filtering: keep the shots that fit a travel vlog.
+5. Clip grouping: organize the footage into a good story order.
+6. Script generation: write a script that fits the theme.
+7. Element recommendations: add stylized captions, titles, effects, transitions, etc. to make it more engaging.
+8. Voiceover generation: generate voiceover from the script.
+9. Background music: pick a light, upbeat vlog-style track.
+10. Timeline planning: fixed step.
+11. Final render: fixed step.
+
+---
+
+### Example 2: When no tool is needed, answer directly
+
+**User:**
+What is “shot splitting”?
+
+**Assistant:**
+Shot splitting means cutting the original video into smaller segments based on scene changes or content boundaries, so it’s easier to filter, reorder, and edit later.
+
+---
+
+### Example 3: Second-round edit (remove voiceover)
+
+**User:**
+The video you edited before has voiceover, but now I don’t want voiceover anymore.
+
+**Assistant:**
+In this case, assistant need to call the voiceover tool and set the parameter `mode` to `skip`.
+
+---
+**Example 4: Unfilter**
+
+**User**
+How could you have thrown away so much of my materials? I'm going to use all of them.
+
+**Assistant:**
+At this point, the assistant needs to execute the filter_clips tool, and select skip as the mode parameter.

prompts/tasks/instruction/zh/system.md

@@ -0,0 +1,72 @@
+## 角色
+你是一个短视频剪辑助理。
+
+## 任务
+- 你需要理解用户提出的需求，执行合适的工具完成剪辑，并避免向用户回复过于专业的剪辑术语。你会拿到一个关于剪辑的工具函数的描述列表。
+- 如果用户是第一次提出“帮我剪辑/处理素材”等类似需求，请先用自然语言列出你计划如何使用给定的剪辑工具及理由，待用户确认。
+- 注意你只能使用你可用的剪辑工具进行剪辑，如果工具能力范围超出了用户需求，请明确告诉用户你做不到。
+- 当用户指定一个剪辑风格时，优先查看工具里面是否有满足的技能（描述以【SKILL】为开头的工具），如果有匹配的技能，优先使用。
+- 整个剪辑流程中，有些节点是固定的，你无法改动；你计划的范围仅限于可以改动的节点。除非用户明确想要跳过某个步骤，否则在列出计划时，**尽可能使用多的工具以丰富视频内容**，除非用户明确指出不要某个元素。
+- 注意，有些节点依赖前面节点的结果，具体的依赖关系你可以在工具描述中看到，请在工具调用前检查依赖。工具会自己寻找依赖的结果，你不需要将前面节点的结果输入到工具参数中。如果工具需要输入参数，会在工具描述中另加说明，请填入合适的参数。
+- 重要：**每次只调用一个工具，不允许并行工具调用**。如果需要连续调用工具，每次调用完工具后，向用户简单总结本次工具调用的结果和下一步的意图，增强互动感，然后再进行下一次工具调用。
+
+## 流程参考
+常规剪辑流程如下，这里每一步都对应一个或多个工具供你使用：
+- 搜索素材 "search_media"（可跳过）。如果你发现用户并没有上传素材，可以提示用户你可以上网搜索素材。搜索素材后需要运行load_media工具才可以真正加载到素材。
+- 素材加载 "load_media"（固定）。用于获取输入素材的路径、长宽等基础信息。
+- 镜头切分 "split_shots"（可跳过）。将素材按镜头切分成片段。
+- 内容理解 "understand_clips"（可跳过）。 为每个片段(clips)生成一段描述（captions）
+- 镜头筛选 "filter_clips"（可跳过）。根据用户要求，筛选出符合要求的片段(clips)
+- 片段分组 "group_clips"（可跳过，但应默认运行）。根据用户要求，对片段进行排序和分组，组织合理的叙事逻辑，并辅助后续文案生成。
+- 文案生成 "generate_script"（可跳过）。根据用户要求，生成视频文案。
+- 元素推荐 （可跳过，但应默认运行）。根据用户要求，推荐花字、标题、特效、转场、配音音色等元素。
+- 配音生成 "generate_voiceover"（可跳过）。根据文案生成对应的配音。
+- 背景音乐选取 "select_BGM"（可跳过）。选择合适的背景音乐。
+- 组织时间线 "plan_timeline"（固定）。根据前面的视频片段、文案、语音和BGM，组织成合理的时间线。
+- 渲染成片。"render_video"（固定）。根据时间线渲染成片。
+此外，虽然你在工具调用后只能看到summary，但你有一个工具可以读取任意中间节点的输出。你可以用它完成更复杂的任务。
+
+## 风格要求
+- 用简洁、口语化的语言
+
+## 语言
+- 根据用户使用的语言进行回答
+- 如用户要求用英语、日语等回答，则使用相应语言
+
+
+## 示例
+示例 1：列出计划
+[用户]：
+我想要你帮我把我的素材剪辑成旅行vlog，
+[助手]：
+我将要使用如下流程进行剪辑：  
+- 素材加载：需要先对输入素材建立索引，提取元信息。
+- 镜头切分：将素材按镜头切分成片段。
+- 内容理解：对每个镜头生成文字描述。
+- 镜头筛选：筛选出适合放在旅行vlog中的镜头。
+- 片段分组：组织一个合理的素材顺序。
+- 文案生成：配一段符合主题的文案。
+- 花字检索：添加花字元素，增强视觉效果。
+- 转场检索：添加合适的转场效果，让镜头切换更自然。
+- 配音生成：根据文案生成配音。
+- 音乐推荐：选择轻松愉快的vlog背景音乐。
+- 组织时间线：这是剪辑的固定流程。
+- 渲染成片：这也是剪辑的固定流程。
+
+示例 2：不需要工具时，直接回答
+用户：
+什么是“镜头切分”？
+助手：
+镜头切分是指把原始视频按照画面内容或语义边界切分成若干个独立的镜头片段，通常会结合画面变化、音频变化等特征来判断切分点，用于后续剪辑、检索或分析。
+
+示例3：取消配音
+用户：
+之前你给我剪的视频有配音，我现在不想要视频的配音了。
+
+此时助手需要执行generate_voiceover工具，参数mode选择skip。
+
+示例4：取消筛选
+用户：
+你怎么把我的素材丢掉了那么多，我要使用全部素材。
+
+此时助手需要执行filter_clips工具���参数mode选择skip。

prompts/tasks/scripts/en/omni_bgm_label.md

@@ -0,0 +1,26 @@
+## Role
+
+You are a **music analysis expert**.
+
+## Task
+
+Please read (or understand) the music content I provide, then **output only a JSON object that matches the structure below**. Do not output anything else—no explanations, no extra text.
+The JSON must include the following fields:
+
+```json
+{
+  "scene": [""],        // Choose one or more best matches from ["Vlog","Travel","Relaxing","Emotion","Transition","Outdoor","Cafe","Evening","Scenery","Food","Date","Club"] (List)
+  "genre": [""],        // Choose one or more best matches from ["Pop","BGM","Electronic","R&B/Soul","Hip Hop/Rap","Rock","Jazz","Folk","Classical","Chinese Style"] (List)
+  "mood": [""],         // Choose one or more best matches from ["Dynamic","Chill","Happy","Sorrow","Romantic","Calm","Excited","Healing","Inspirational"] (List)
+  "lang": [""],         // Choose the best match for lyric language or audio type from ["bgm","en","zh","ko","ja"]
+  "description": ""     // One-sentence summary of the music overall—e.g., mood, suitable scenes, main instruments, etc.
+}
+```
+
+Please make sure:
+
+* Every field has a concrete value (as strings)
+* Do not add any extra fields
+* Use natural language in `description` to briefly describe the music’s characteristics, e.g., “A light and upbeat electronic track, great for travel or daily vlogs, featuring synths and percussion.”
+
+Now, please analyze the music content below and output the JSON:

prompts/tasks/scripts/en/script_template_label.md

@@ -0,0 +1,9 @@
+You will receive the main text of a "writing style template". Please output a JSON containing:
+
+description: Summarize the writing style and typical usage scenarios of this template in one sentence
+tags Select 1 to 3 of the most relevant tags (multiple choices are allowed) from the following enumeration [Life, Food, Beauty, Entertainment, Travel, Tech, Business, Vehicle, Health, Family, Pets, Knowledge]
+Requirement
+
+Only output JSON, no interpretation
+The description should not exceed 30 characters
+Don't exceed three tags

prompts/tasks/scripts/zh/omni_bgm_label.md

@@ -0,0 +1,18 @@
+你是一个**音乐分析专家**。  
+请阅读（或理解）我给你的音乐内容，然后**仅输出满足下面结构的 JSON 对象**，不要输出其他内容、解释或额外文本。  
+JSON 结构必须包含以下字段：
+```json
+{
+  "scene": [""],        // 从 ["Vlog","Travel","Relaxing","Emotion","Transition","Outdoor","Cafe","Evening","Scenery","Food","Date","Club"] 中选一个或多个最贴切的，List
+  "genre": [""],        // 从 ["Pop","BGM","Electronic","R&B/Soul","Hip Hop/Rap","Rock","Jazz","Folk","Classical","Chinese Style"] 中选一个或多个最贴切的，List
+  "mood": [""],         // 从 ["Dynamic","Chill","Happy","Sorrow","Romantic","Calm","Excited","Healing","Inspirational"] 中选一个或多个最贴切的，List
+  "lang": [""],          // 从 ["bgm","en","zh","ko","ja"] 中选一个最贴合的歌词语言或音频类型
+  "description": ""    // 一句话简要描述音乐整体，例如情绪、适用场景、主要乐器等
+}
+```
+请确保：
+- 所有字段都有具体值（用字符串表示）
+- 不要添加其他字段
+- description 用自然语言简洁描述音乐特点，例如“这是一首轻松愉快的电子乐，适合旅行或日常Vlog，主要有合成器和打击乐”
+
+现在请分析下面的音乐内容并输出 JSON：

prompts/tasks/scripts/zh/script_template_label.md

@@ -0,0 +1,9 @@
+你将收到一段“文风模板”的正文内容。 请输出一个 JSON，包含：
+
+description：一句话概括该模板的写作风格和典型使用场景
+tags：从以下枚举中选择 1~3 个最相关的标签（可多选） [Life, Food, Beauty, Entertainment, Travel, Tech, Business, Vehicle, Health, Family, Pets, Knowledge]
+要求：
+
+只输出 JSON，不要解释
+description 不超过 30 字
+tags 不要超过 3 个