Many people use Claude Code to write code, modify projects, and even use it to complete article creation every day, but few people realize that it can also become a dedicated AI image production tool. I recently tried something Claude Code + Codex + GPT Image The workflow integrates all image generation tasks such as article covers, pictures, and social media materials into one work environment. The entire process does not require frequent switching of web pages, and there is no need to manually download and organize pictures, truly realizing the integration from content creation to visual design.
In the past, my workflow has always had an obvious breakpoint. The article can be handed over to Claude Code to write, and the code can be handed over to it to modify, but when it comes to the image section, you must open ChatGPT or other image generation tool, enter the prompt words, wait for the image to be generated, download, and then move the file to the project directory. Although it only takes a few minutes each time, what is really wasted is not the time, but the interruption of attention caused by constantly switching tools. For people who need to create a large amount of content every day, this repeated switch will continue to disrupt the creative rhythm.
Later, I discovered that as long as it cooperates with Codex CLI, Claude Code can fully undertake the scheduling work of image generation. The entire process can be understood as the collaboration of three roles: Claude Code is responsible for understanding article content, analyzing requirements, and planning visual solutions;Codex CLI is responsible for executing commands, calling tools, and managing files; and GPT Image is responsible for the final image generation. After combining the three, you can directly complete various visual materials such as article covers, information maps, small red book first pictures, X long text accompanying pictures, Newsletter illustrations, etc. in Claude Code without leaving the current project.
If you want to experience this process, you first need to install Codex CLI. may be used npm install -g @openai/codex, or implemented through Homebrew brew install --cask codex Complete the installation and run codex login Log in to your ChatGPT account. After the authorization is completed, Codex can call the corresponding capabilities without requiring additional configuration of API Key (the details are subject to the official support method).
完成配置后，真正使用时反而非常简单。例如，你只需要告诉 Claude Code：“使用 Codex 调用 GPT Image，为这篇关于 AI 工作流的文章制作一张科技感封面，并保存到当前项目目录。”随后，Claude Code 会自动分析文章主题、整理视觉方向、生成适合的图片提示词、调用 GPT Image 完成生成，并把图片保存到指定位置。整个过程几乎不需要人工干预，你关注的是最终效果，而不是每一步具体如何执行。
不过，我认为这套工作流真正有价值的地方，并不是自动生成图片，而是可以进一步沉淀为一个属于自己的 AI 图片生成 Skill。很多人习惯保存大量 Prompt，每次生成图片都要重新复制粘贴，告诉 AI 自己喜欢什么风格、需要什么尺寸、图片保存在哪里。这种方式虽然有效，但长期来看效率并不高。相比之下，把这些规则固化到一个固定流程中，让 Claude Code 自动调用，就能真正形成可复用的工作流。
这个 Skill 可以包含自己的品牌风格、常用配色、封面布局、图片比例、提示词模板、文件命名规则以及素材保存位置。以后只需要输入一句“生成这篇文章的配图”，Claude Code 就会按照既定规范完成整个流程，而不需要重复描述所有细节。随着使用次数增加，你的审美、工作习惯和输出规范都会逐渐沉淀下来，AI 输出的结果也会越来越符合你的需求。
这也是 Skill 与普通 Prompt 最大的区别。Prompt 更像一次性的指令，每次都需要重新描述需求；而 Skill 更像一套标准化工作流程，把重复工作全部交给 AI 自动完成。真正高效的 AI 应用，并不是不断优化 Prompt，而是不断优化自己的工作流。
在我看来，这套方案最大的优势还在于明确了不同 AI 的分工。Claude Code 擅长理解文章、分析结构、规划视觉内容；Codex 擅长执行命令、调用工具和管理文件；GPT Image 则专注于图片生成。每个工具负责自己最擅长的事情，整个流程既清晰又高效。
例如，当你完成一篇 AI 教程后，只需要告诉 Claude Code：“根据这篇文章生成 5 张配图。”它可以自动拆解内容，规划封面、流程图、案例展示、数据卡片和结尾引导图，再统一交给 Codex 调用 GPT Image 批量完成生成。这种一次规划、多张输出的方式，比每次只生成一张图片更符合内容创作者的真实需求。同一篇文章还可以同时生成适用于 X、小红书、视频封面、Newsletter 和博客等多个平台的素材，大幅减少重复劳动。
AI 时代，真正拉开差距的已经不是谁拥有更多工具，而是谁拥有更成熟的工作流。Claude Code 不只是一个代码助手，也可以成为整个内容创作流程的中控台。当你把自己的规则、审美和工作方式沉淀成一套可重复使用的 Skill 后，AI 才真正从一个聊天工具，变成能够持续帮助你完成工作的生产力伙伴。