Spot it. Save it. Free, local-first Chrome extension for images, videos and prompts. Explore 2,000+ curated cases and connect your library to AI Agents via MCP. 提示词导演
JavaScript ai-artai-videochrome-extensionclaude-codecodexhiggsfieldjimenglocal-firstmcpmidjourneymoodboardprompt-engineeringprompt-libraryprompt-managerreference-libraryseedancevisual-bookmarksweb-clipper
让 GitHub 首页清楚展示采集与 Agent 创作价值
wchao6891109adf7f8c1f
diff --git a/README.md b/README.mdindex bd6258a..ec2407a 100644--- a/README.md+++ b/README.md@@ -2,42 +2,68 @@<p align="center"><strong>English</strong> · <a href="README.zh-CN.md">简体中文</a></p>-<h1 align="center">Keep the image. Keep the prompt.</h1>+<h1 align="center">Turn great references into your next creation.</h1>-<p align="center"><strong>好画面,连同提示词一起收藏。</strong><br>Collect creative references from Higgsfield, Jimeng, X and more.<br>Keep images, videos, prompts and sources together in your own visual library.</p>+<p align="center"><strong>One-click capture. A visual prompt library. Ready for your Agent.</strong><br>Collect from Jimeng, Higgsfield, X, HeyBox, WeChat articles, Feishu docs and more.<br>Keep the image, the prompt and the context — then put them to work in your next project.</p><p align="center"><a href="https://github.com/wchao6891/PromptDirector/releases/latest"><strong>Download the latest release</strong></a> · <a href="https://wchao6891.github.io/PromptDirector-Curated/">Explore curated cases</a> · <a href="#install-and-update">Installation guide</a></p>-<p align="center">Free & open source · Local first · No account required · Optional AI analysis</p>+<p align="center">Free & open source · Local first · No account required · Local MCP Agent access</p>-## Bring the best of your favorite platforms together+## Found a great reference? Start capturing in one click.-Found a reference worth keeping? Save more than a bookmark. PromptDirector is a Chrome extension with capture adaptations for creative websites and social platforms. It brings the media, text and source available on a page into one case.+**Keep the work and the creative clues behind it.** PromptDirector is a Chrome extension for visual creators. Collect images, videos, text, prompts and sources into a library you can browse visually, organize by project and bring into your next creative conversation.-| Find inspiration on | Keep it in your library |+**Open a work, post or article → Capture → Preview and select → Save cases.**++| Where inspiration lives | What you can bring with you || --- | --- |-| **Higgsfield** | Collect work images, videos and creative descriptions to reference a shot or visual approach later. |-| **Jimeng / 即梦** | Collect work media and the original prompt, model and author information provided by the page. |-| **X / Twitter** | Extract post text and image or video references; save prompts shared by the author alongside the work. |-| **More creative websites** | Adaptations also cover LiblibAI, LibTV, Krea, Pinterest and Behance. On other pages, select text, pick images or capture a screenshot. |+| **Jimeng · Higgsfield · LiblibAI · Krea · LibTV** | Work images, videos and publicly available prompts or creative details. Keep the visual and its context together. |+| **HeyBox / 小黑盒 · WeChat articles** | Save a whole article or split clearly structured multi-case posts into selectable image-and-prompt groups. Keep a single HeyBox image gallery with its prompt from the post body. |+| **Feishu documents** | Capture document bodies, long documents and images within tables, preserving cell placement and merged cells. |+| **X / Twitter · Xiaohongshu · Weibo · Reddit** | Collect the current post's text and accessible media, including prompts shared by the author. |+| **Pinterest · Behance · ArtStation · Huaban · ZCOOL** | Bring visual work into one place to browse, search and organize. |+| **More webpages and local files** | General webpage capture, text and image selection, screenshots and local imports help gather scattered materials. |++### One article. A collection of individual references.++For articles with clear case headings or numbering, preview the detected groups under **Cases in article**, select the cases you want and save them individually. Review each image-and-prompt group before saving, or keep the whole article for context.++### Keep the context that makes a reference useful.++Multi-image works stay grouped, articles retain media order and Feishu tables retain their layout. Media, text, original prompts, sources and later analysis have their own place, ready to revisit or use with a model.++Capture, browsing and organization require no AI service. Pages must be accessible and original prompts must be public; login state, lazy loading and site changes can affect results. Preview before saving. Videos are saved locally when a complete file is accessible; supported standalone players and HLS streams can play in the library. See the [capture guide](docs/CAPTURE_SUPPORT.md).-**Open a work or post → Start capture with one click → Preview and adjust → Save.**+## Your eye chooses. Your Agent takes it further.-Capture depends on content the page provides and makes accessible. Original prompts must be shared by their authors; descriptions and AI analysis are not original prompts. Login state, page changes and media access restrictions can affect the result.+**Connect your library to Codex, Claude Code or WorkBuddy.** Search references, read prompts, retrieve originals, capture URLs and save creative materials back to the library through your Agent conversation.-Capture does not call AI. Videos are saved locally when possible; supported standalone players and HLS streams can play inside the library. See the [capture guide](docs/CAPTURE_SUPPORT.md) for playback and access boundaries.+1. Select **Copy connection request** in the extension's **Settings → Agent connection**.+2. Paste it into an Agent with local MCP support so it can install, configure and verify the connector.+3. Enable the connection in Chrome when prompted, then start using your library.-## Make your collection useful+Try requests like:-| What you want to do | How PromptDirector helps |+> “Find character-poster references in my library and list a few options for me to choose.”+>+> “Read the selected cases' original images and prompts, then use them to draft this campaign's visual direction.”+>+> “Save the approved prompt and generated images back to PromptDirector.”++**Google Chrome on Windows, macOS and Linux.** Configuration entry points are provided for Codex, Claude Code and WorkBuddy; other local MCP clients can use the generic configuration. The Agent installs a separate local connector, and the first connection requires browser authorization. Cloud-only workspaces cannot use this local flow. Creative and generation capabilities depend on your Agent and its tools. [Agent setup guide →](connector/INSTALL.md)++## Build a library that grows with your creative practice++| What your next project needs | What your library provides || --- | --- |-| **Find a reference visually** | Browse an image-first library, search text and filter by tags to rediscover useful work. |-| **Organize a project** | Arrange cases in nested projects, combine related cases and manage references in batches. |-| **Understand what works** | Connect your own AI service to analyze text, images or videos when you want more context and searchable notes. |-| **Keep your materials together** | Add local images, videos, PDFs, Markdown, TXT, HTML and timestamped video notes. |-| **Share and preserve your work** | Export selected cases or a project subtree as a sharing package. Full folder backups support recovery on another computer; encrypted folder sync supports recovery across your devices. |+| **A reference you can recognize at a glance** | An image-first case wall, text search and tag filters. Browse the work and choose for yourself. |+| **A focused collection for each project** | Nested projects, related cases and batch management to keep a creative task's materials together. |+| **Experience you can reuse** | The Composer reads selected cases and drafts editable prompts, reusable Skills and tag suggestions for your review. |+| **Room for more than images** | Local images, videos, PDFs, Markdown, TXT, HTML and timestamped video notes. |+| **A collection you can share and preserve** | Export cases or project subtrees, create full folder backups, recover on another computer and use encrypted folder sync. |-The Composer can chat, find cases on request, read selected material, create editable prompts and draft reusable Skills or case tags for your review. It can also check the curated catalog for available versions without downloading packages. Search results stay in the conversation until you choose references; only selected or explicitly requested images are sent to a model. Capturing, organizing and browsing your library do not require an AI service.+AI analysis is optional and uses your own service. Selected references help communicate the visual direction you want; only selected or explicitly requested original images are sent to a model. You can also use PromptDirector entirely as a visual reference library.## Start with curated references@@ -90,7 +116,3 @@ Contributions to capture adaptations, interactions, tests and documentation areSee the [development guide](docs/DEVELOPMENT.md) to build and test contributions.Code is licensed under [Apache License 2.0](LICENSE). See [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md) for dependencies. The code license does not grant rights to third-party case media.--## Agent access--In settings, select **Copy connection request** under Agent connection and send it to an Agent with local MCP support. The Agent follows the [setup guide](connector/INSTALL.md) to install the connector, configure it and verify access; enable the connection in Chrome when requested. Supports **Google Chrome on Windows, macOS and Linux**. Agents can search cases, read prompts and original files, capture requested URLs and save creative materials. The connector is a separate local component installed by the Agent; cloud-only workspaces cannot use this local setup flow.diff --git a/README.zh-CN.md b/README.zh-CN.mdindex 06099f6..e99a882 100644--- a/README.zh-CN.md+++ b/README.zh-CN.md@@ -2,46 +2,68 @@<p align="center"><a href="README.md">English</a> · <strong>简体中文</strong></p>-<h1 align="center">好画面,连同提示词一起收藏。</h1>+<h1 align="center">把好案例,变成你的创作底气。</h1>-<p align="center">从 Higgsfield、即梦、X 等平台收集创作参考,建立自己的视觉灵感库。<br>图片、视频、提示词与来源放在一起,下一次创作时找得到、用得上。</p>+<p align="center"><strong>多网站一键采集 · 图文提示词成套收藏 · Agent 随时调用</strong><br>从即梦、Higgsfield、X,到小黑盒、微信公众号、飞书文档。<br>把你看中的画面、提示词和经验存进自己的案例库,让下一次创作从好参考开始。</p><p align="center"><a href="https://github.com/wchao6891/PromptDirector/releases/latest"><strong>下载 GitHub 最新版</strong></a> · <a href="https://wchao6891.github.io/PromptDirector-Curated/">浏览精选案例</a> · <a href="#安装与更新">安装说明</a></p>-<p align="center">免费开源 · 本地优先 · 无需账号 · AI 分析可选</p>+<p align="center">免费开源 · 本地优先 · 无需账号 · 支持本机 MCP Agent</p>-## 集各家所长,存进自己的库+## 刷到好案例,一键开始收藏-刷到喜欢的作品,不必只留一个链接。PromptDirector 是一个 Chrome 扩展,为创作网站和社交平台做专门采集适配,把页面提供的素材、文字与来源整理成一个案例。+**好画面值得留下,做出它的提示词也一样。** PromptDirector 是为视觉创作者打造的 Chrome 扩展:把网页中的图片、视频、文字、提示词和来源一起整理入库。收藏时保留创作线索,需要时看图选择参考,让积累真正进入下一次创作。-| 在哪里发现灵感 | 可以怎样收藏 |+**打开作品、帖子或文章 → 一键采集 → 预览勾选 → 保存案例。**++| 灵感来自哪里 | 值得带走什么 || --- | --- |-| **Higgsfield** | 收集作品图片、视频与创作说明,把喜欢的镜头和视觉表达留下来。 |-| **即梦 Jimeng** | 收集作品媒体,以及页面提供的原始提示词、模型和作者信息。 |-| **X / Twitter** | 从帖子中提取正文和图片、视频线索,把作者分享的提示词与作品一起保存。 |-| **更多创作与设计网站** | 已为 LiblibAI、LibTV、Krea、Pinterest、Behance 等站点加入适配;其他页面也可选择文字、图片或截图采集。 |+| **即梦 · Higgsfield · LiblibAI · Krea · LibTV** | 作品图片、视频,以及页面公开的提示词和创作信息。把喜欢的画面与创作线索放在一起。 |+| **小黑盒 · 微信公众号** | 收藏整篇图文,也可将结构清晰的多案例文章拆开,勾选其中的图片与对应提示词,分别保存。小黑盒单案例图片墙可与正文提示词成组收藏。 |+| **飞书文档** | 保存正文、长文档和表格内图文,保留单元格位置与合并关系,让案例回到对应的位置。 |+| **X / Twitter · 小红书 · 微博 · Reddit** | 收集当前帖子正文和可获取的媒体,把作者分享的提示词与作品一起留下。 |+| **Pinterest · Behance · ArtStation · 花瓣 · 站酷** | 收集作品页面的视觉参考,集中浏览、搜索和整理。 |+| **更多网页与本地资料** | 通用网页采集,加上选区、选图、截图和本地文件导入,把分散的资料汇到一个库里。 |-**打开作品或帖子 → 一键发起采集 → 预览、调整 → 保存入库。**+### 一篇文章,收藏成一组能单独使用的案例-采集以页面实际提供、当前可访问的内容为准。原始提示词需要作者公开;作品说明与 AI 分析会作为不同来源的内容保留。登录状态、页面变化或媒体访问限制可能影响采集结果。+遇到十几个案例放在一起的教程或分享帖,可以在「文章内案例」中预览分组,只勾选想要的案例,分别入库;也可以保留整篇文章。支持明确案例标题或编号的图文结构,拆分后仍可核对每组图片和提示词。-采集不调用 AI。视频优先保存本地,可用的独立播放器与 HLS 流可在库内播放。播放与访问边界详见[采集说明](docs/CAPTURE_SUPPORT.md)。+### 好参考,连同上下文一起留下-## 收藏之后,继续用起来+多图作品保留成组关系,文章保留图文顺序,飞书表格保留布局。图片、正文、原始提示词、来源与后续分析各有位置,方便回看,也方便交给模型使用。-| 你想做的事 | PromptDirector 如何帮你 |-| --- | --- |-| **看图找灵感** | 用图片优先的资料墙浏览案例,搜索文字、筛选标签,找回记不清名字的参考。 |-| **按项目整理** | 把案例放进可嵌套的项目树,组合相关案例,批量管理同一创作任务的资料。 |-| **看懂为什么好** | 按需连接自己的 AI 服务,分析文字、画面或视频,为参考补充理解与检索线索。 |-| **把资料放在一起** | 收录本地图片、视频、PDF、Markdown、TXT、HTML,以及视频时间点笔记。 |-| **分享与长期保存** | 把选定案例或项目子树打成分享包;完整资料夹备份支持换电脑恢复,加密文件夹同步可在自己的设备间恢复资料。 |+采集、浏览和整理无需 AI 服务。页面需要可访问,原始提示词需要作者公开;登录、懒加载和网站变化可能影响结果,保存前可预览核对。视频在可取得完整文件时保存本地,支持的独立播放器与 HLS 流可在库内播放。详见[网页与媒体采集说明](docs/CAPTURE_SUPPORT.md)。++## 你的眼光选参考,你的 Agent 接着创作++**把案例库交给 Codex、Claude Code、WorkBuddy 使用。** 找参考、读提示词、取原件、采集网址、回存创作材料,在 Agent 对话里串起来。你在案例墙里挑中的画面,也能成为下一次对话中明确可用的创作依据。++1. 在插件「设置 → Agent 连接」点击 **复制连接指令**。+2. 粘贴给支持本机 MCP 的 Agent,让它安装连接器、配置并验证。+3. 按提示在插件里启用授权,开始调用案例库。-创作台可直接对话、按需找案例、读取资料、生成可编辑提示词,也可提炼 Skill 或标签建议,供你查看、编辑后保存。还可以查询精选目录的可用版本,无需下载案例包。查询候选留在对话里,选定后才加入参考;原图只在你选中或明确指定时发送给模型。采集、整理和浏览资料无需配置 AI 服务。+连接之后,可以这样下需求:-## 让 Agent 使用案例库+> “从我的案例库找几组国风人物海报参考,列出来让我选。”+>+> “读取我选中的案例原图和提示词,参考它们写这次活动的视觉方案。”+>+> “把这次确认的提示词和生成图片存回 PromptDirector。”++**Google Chrome + Windows / macOS / Linux。** Codex、Claude Code、WorkBuddy 提供配置入口,其他本机 MCP 客户端可使用通用配置。连接器是由 Agent 安装的独立本机组件,首次连接需要浏览器授权;纯云端工作台不适用此流程。具体创作和生成能力取决于所用 Agent 及其工具。[查看 Agent 安装说明 →](connector/INSTALL.md)++## 越收藏,越有自己的创作资料库++| 你在创作中需要的 | 这里已经准备好 |+| --- | --- |+| **看一眼,就知道想用哪张** | 图片优先的案例墙,配合文字搜索与标签筛选,让人直接看图做判断。 |+| **一个项目,一套参考** | 可嵌套项目树、案例组合和批量管理,把同一任务的灵感放在一起。 |+| **把经验用在下一次** | 创作台可读取选定案例,生成可编辑提示词、提炼可复用 Skill 和标签建议,由你审阅保存。 |+| **图片之外,也装得下资料** | 本地图片、视频、PDF、Markdown、TXT、HTML,以及视频时间点笔记。 |+| **分享给别人,也留给未来的自己** | 导出案例或项目子树分享包,完整资料夹备份、换电脑恢复和加密文件夹同步。 |-在设置的「Agent 连接」中点击「复制连接指令」,发给支持本机 MCP 的 Agent,由它按[统一安装说明](connector/INSTALL.md)安装连接器、配置并验证;需要时在插件中点击启用授权。支持 **Windows、macOS 和 Linux 上的 Google Chrome**。连接后可搜索案例、读取提示词和原件、按指令采集网页,并把创作材料存回库中。连接器是独立的本机组件,由 Agent 完成安装;纯云端工作台不适用此本机流程。+AI 分析按需开启,使用你自己的服务;选中的参考可以帮助模型理解这次想要的画面。原图只在你选中或明确指定时发送给模型。你也可以只把它当作一套好用的视觉案例库,随时查、随时看。## 从精选案例开始diff --git a/docs/CAPTURE_SUPPORT.md b/docs/CAPTURE_SUPPORT.mdindex 05ef5d2..abc1f64 100644--- a/docs/CAPTURE_SUPPORT.md+++ b/docs/CAPTURE_SUPPORT.md@@ -8,6 +8,8 @@| 平台 | 提取和播放路径 ||---|---|+| 小黑盒、微信公众号 | 文章正文与图片;有明确案例标题或编号的多案例图文可在“文章内案例”中分组预览、勾选并分别保存,也可保存整篇文章。小黑盒单案例图片墙与正文提示词保持一组。 |+| 飞书文档 | 提取正文和长文档内容,保留表格内图文、单元格位置及合并关系;无法取得原件的画布内容保留可见快照。未完成的扫描会提示核对。 || X | 当前帖正文、媒体和可选引用组;未取得直链的视频,在本次主动采集期间刷新当前帖一次,读取该帖正常加载的媒体响应。优先保存完整 MP4,保留独立播放地址;不会嵌入整条帖子。 || 即梦、Higgsfield、LiblibAI、LibTV、Krea | 专门的当前作品或公开内容提取,结合页面媒体与明确的原始提示词。Krea 区分成品与风格参考;LibTV 保留可读取的节点素材,不恢复可运行工程。 || ArtStation、Pinterest、Behance | 作品结构与页面媒体结合提取;ArtStation 的可访问官方视频子页面可读取实际视频文件。Pinterest 的已解析图片使用原件候选。 |