基于 Chromium Manifest V3；当前 Beta 通过“加载已解压的扩展程序”安装。
安装或更新 Beta
当前工作区提供的是解压缩版本，适合开发测试。请直接选择完整的 lumaprompt-extension 文件夹，不要只选择其中某个文件。
- 打开扩展管理页
Edge 输入
edge://extensions/；Chrome 输入chrome://extensions/。 - 开启开发人员模式
在扩展管理页打开“开发人员模式”，显示本地加载和重新加载按钮。
- 加载扩展目录
点击“加载解压缩的扩展程序”，选择项目中的
lumaprompt-extension文件夹。 - 固定到工具栏
在浏览器扩展菜单中固定 EidoPrompt，方便打开弹窗、设置和自定义 API。
- 刷新普通网页
首次安装或点击“重新加载”更新后，需要刷新已打开的网页，新的悬浮分析脚本才会生效。
不能运行的页面
浏览器内部页面、扩展商店、新标签页及部分受保护页面禁止注入扩展脚本，因此不会出现网页图片分析入口。这是浏览器限制。
首次配置
首次打开扩展会进入邮箱验证码登录界面。验证成功后，Supabase Auth 会自动登录或创建账户；分析额度以 Supabase 云端余额和流水为准，本机只缓存用于界面显示的摘要。
- 邮箱登录/注册
填写邮箱并点击“获取验证码”，输入收到的 6 位验证码后登录；无需设置密码，也不需要单独注册。
- 选择界面语言
进入“设置”，选择简体中文或 English；该选择同时用于界面和新的图片分析。
- 确认图片分析开关
首页“图片分析”开关控制网页图片旁的分析按钮。关闭后仍可在扩展弹窗中导入本地图片。
- 决定是否保存历史
“保存悬浮窗历史”默认开启，最多保留 30 条；关闭时会清空已保存的悬浮窗历史。
- 选择分析模式
默认使用 EidoPrompt 平台模型并按成功任务消耗一次额度；也可以配置自己的视觉 API，测试成功后改用自定义模型且不扣平台次数。
分析一张图片
EidoPrompt 提供多个主动入口。无论从哪里开始，只有你触发任务后才会读取并准备图片。
把指针移动到普通网页图片上，点击图片旁出现的“分析”，结果会在当前页面的悬浮窗中打开。
在网页图片上右键，选择“用 EidoPrompt 悬浮分析图片”，适合悬浮入口没有出现时使用。
先从网页图片打开悬浮窗，再用导入菜单选择文件、拖入图片，或读取剪贴板图片。支持 JPG、PNG、WebP、GIF。
点击悬浮窗底部截图按钮，在当前可见网页中拖出至少 24 × 24 像素的选区；按 Esc 退出截图模式。
结果可以继续编辑
悬浮窗结果支持在“中 / EN”之间切换、直接编辑和复制。区域截图只捕获当前标签页的可见区域，不是整页长截图。
同时处理多张图片
多图任务会进入同一队列，而不是用后一张覆盖前一张。队列中的任务会显示等待、分析中、完成或失败状态。
- 一次选择多张本地图片
在弹窗或悬浮窗的文件选择器中多选图片，也可以把多张图片一起拖入。
- 连续分析网页图片
在不同图片旁依次点击“分析”；每张图片都会作为独立任务加入悬浮历史与解析队列。
- 从历史面板切换任务
点击悬浮窗底部的历史按钮查看等待、分析中、完成或失败状态，选择任意卡片即可切换当前预览和结果。
- 复制并继续创作
任务完成后逐条检查提示词；本地历史最多保留 30 条，超过上限时保留最新记录。
每张图片都是独立任务
多图不会融合为同一条提示词；系统会依据网络与视觉服务状态自动调整并发。每个文件最大 10MB，GIF 也不承诺理解完整动画。
配置视觉 API
点击扩展右上角 API 图标进入“自定义 API”。默认已可使用平台模型；三项信息填写并通过视觉测试后，弹窗和网页悬浮窗会自动切换到你的 API，之后不消耗 EidoPrompt 次数。
API Base URL / Endpoint
填写兼容 Chat Completions 图片输入格式的 Base URL 或完整 Endpoint。火山方舟可填写 https://ark.cn-beijing.volces.com/api/v3，扩展会补全请求路径。
视觉模型名称
填写服务商提供的、明确支持图片输入的模型 ID。文本模型即使可以响应，也无法完成真实图片识别。
API Key
填写原始密钥，不要手动添加 Bearer。使用权限与额度受限的专用 Key；当前字段保存在浏览器本地，没有额外应用级加密。
视觉测试
点击“视觉测试”确认地址、模型和密钥可用。成功后配置会自动保存并切换到自定义模式；测试只发送扩展生成的红、黄、蓝测试图，不使用你的图片。输入会即时写入会话草稿，弹窗关闭重开后仍可继续。
优先使用 HTTPS
当前 Beta 允许自定义 HTTP Endpoint，但 HTTP 可能让图片和 API Key 在传输中缺少加密保护。请优先使用可信服务的 HTTPS 地址；清空三项配置并保存即可返回平台模型。
常见问题
没有配置 API 还能使用吗？
可以。默认使用 EidoPrompt 平台模型，每次成功分析消耗一次账户额度；购买或赠送的次数都可使用。
图片什么时候会发送到外部服务？
只有你主动触发分析后才会上传：平台模式经 Supabase Edge Function 转发到火山方舟，自定义模式直接发送到你验证成功的视觉服务。单纯浏览或悬停不会上传图片。
多图会合并成一条提示词吗？
不会。每张图片都是独立任务和独立结果；并发数量会依据网络与视觉服务状态自动调整。
历史记录保存在哪里？
保存在当前浏览器配置文件的扩展本地存储中，默认最多 30 条，不会由当前 Beta 同步到 EidoPrompt 服务器。
Chrome 和 Edge 都能使用吗？
当前版本采用 Chromium Manifest V3，主要针对 Microsoft Edge 和 Chrome。Canvas、CSS 背景图、过小或被遮挡的图片可能不会出现分析按钮。
故障排查
网页没有“分析”按钮
确认“图片分析”开关已打开，重新加载扩展并刷新网页；在普通网页的较大图片上重试。内部页、Canvas 和 CSS 背景图可能无法显示。
剪贴板读取失败
先点击悬浮窗中的导入按钮并允许剪贴板权限，也可以在悬浮窗内直接按 Ctrl+V。
网页图片读取失败
图片站点可能要求登录 Cookie 或启用了防盗链。尝试右键入口；仍失败时，改用本地导入或区域截图。
视觉测试失败
依次检查 Endpoint、支持图片的模型 ID、原始 API Key、账户额度和服务商状态。401/403 常见于鉴权，404 常见于路径或模型错误，429 表示额度或频率受限。
结果过于泛化
先重新分析并确认网络稳定；如果使用自定义 API，确认所选模型确实支持图片输入，必要时重新进行视觉测试或清空配置切回平台模型。
更新后界面仍是旧版
在扩展管理页点击“重新加载”，再刷新所有正在测试的网页。浏览器已注入的旧内容脚本不会自动替换。
需要彻底重置
先退出登录并清空历史和 API 字段；卸载扩展可移除本地设置、令牌和额度资料。删除 Supabase 云端账户需通过隐私页联系方式提出请求。
提交有效反馈
请通过下面的邮箱联系我们，并附上浏览器版本、EidoPrompt 版本、问题页面类型、复现步骤和错误提示；不要发送 API Key 或包含隐私的原图。
反馈邮箱478839209@qq.com
DOCUMENTATION · BETA V0.4.160
Start with installation
finish with your first prompt
This guide covers the real workflow in the current EidoPrompt Beta: load the extension, finish local setup, analyze web or local images, work through a multi-image queue, and diagnose common problems.
Built on Chromium Manifest V3. The current Beta is loaded as an unpacked extension.
JPG, PNG, WebP, and GIF are accepted, up to 10 MB for each imported file.
History stays in the current browser profile and can be disabled or cleared.
Install or update the Beta
The current workspace contains an unpacked build for development testing. Select the complete lumaprompt-extension folder, not an individual file inside it.
- Open extension management
Use
edge://extensions/in Edge orchrome://extensions/in Chrome. - Enable Developer mode
Turn on Developer mode to expose the local-load and reload controls.
- Load the extension folder
Choose “Load unpacked,” then select the project’s
lumaprompt-extensiondirectory. - Pin EidoPrompt
Pin it from the browser extension menu for quick access to the popup, settings, and Custom API.
- Refresh normal web pages
After initial installation or Reload, refresh pages that were already open so the new content script can run.
Restricted pages
Browser-internal pages, extension stores, new-tab pages, and some protected surfaces prohibit extension injection, so the in-page Analyze control cannot appear there.
First-time setup
The first launch opens email-code sign-in. After verification, Supabase Auth signs you in or creates the account automatically. Supabase cloud balances and ledger entries are authoritative; the browser keeps only a display summary.
- Sign in / sign up by email
Enter your email, request a code, and submit the six-digit code you receive. No password or separate registration is required.
- Choose the interface language
Open Settings and select Simplified Chinese or English. The choice also becomes the default for new analyses.
- Check Image Analysis
The Image Analysis switch controls Analyze buttons beside web images. Local imports remain available from the popup when it is off.
- Choose whether to save history
Overlay history is on by default and keeps up to 30 items. Disabling it clears the saved overlay-history array.
- Choose an analysis mode
The EidoPrompt managed model is the default and uses one credit per successful task. You can instead verify your own vision API; custom mode uses no EidoPrompt credits.
Analyze one image
EidoPrompt provides several user-triggered entry points. An image is read and prepared only after you start a task.
Move over an image on a normal website and click Analyze beside it. Results open in an overlay on the same page.
Right-click a web image and choose the EidoPrompt overlay action, useful when the hover control is unavailable.
Open the overlay from a web image, then choose files, drop images, or read a clipboard image from its import menu. JPG, PNG, WebP, and GIF are supported.
Use the overlay capture control and drag a region of at least 24 × 24 pixels on the visible page. Press Esc to leave capture mode.
Results remain editable
Overlay results can be switched between 中 / EN, edited directly, and copied. Region capture uses only the visible tab area, not a full scrolling page.
Process multiple images
New images enter one queue instead of replacing the previous task. Each item shows a waiting, analyzing, complete, or failed state.
- Select multiple local files
Multi-select in the popup or overlay file picker, or drop several images together.
- Analyze several web images
Click Analyze beside different images in sequence; each becomes an independent task in overlay history and the analysis queue.
- Switch from the history panel
Open history to see waiting, analyzing, complete, and failed states, then select any card to restore its preview and result.
- Copy and continue
Review each completed prompt. Local history retains the newest 30 items when its limit is reached.
Every image is a separate task
Multiple images are not fused into one prompt; concurrency adapts to network and provider conditions. Each file is limited to 10 MB, and full GIF-animation understanding is not promised.
Configure a vision API
Open Custom API from the icon at the top-right. The managed model already works by default; once all three custom values pass Vision Test, popup and overlay automatically switch to your API without using EidoPrompt credits.
API Base URL / Endpoint
Enter a Base URL or full endpoint compatible with Chat Completions image input. For Volcengine Ark, https://ark.cn-beijing.volces.com/api/v3 can be used and EidoPrompt completes the request path.
Vision model name
Use the exact provider model ID that explicitly supports image input. A text-only model cannot perform real image recognition.
API Key
Enter the raw key without adding Bearer. Prefer a dedicated key with restricted permission and quota; it is stored locally without additional application-level encryption.
Vision test
Run Vision Test to verify the endpoint, model, and key. Success automatically saves and enables custom mode. The test sends only a generated red, yellow, and blue image, not one of your images. Input is written immediately to a session draft and restores after the popup reopens.
Prefer HTTPS
The current Beta permits custom HTTP endpoints, but HTTP can leave images and API Keys without transport encryption. Clear all three fields and save to return to the managed model.
Frequently asked questions
Can I use EidoPrompt without an API?
Yes. The EidoPrompt managed model is the default and each successful analysis uses one purchased or granted credit.
When is an image sent to an external service?
Only after you actively trigger analysis: managed mode forwards it through a Supabase Edge Function to Volcengine Ark, while custom mode sends it directly to your verified service. Browsing or hovering alone does not upload an image.
Are several images merged into one prompt?
No. Every image is an independent task with its own result. Concurrency adjusts to network and provider conditions.
Where is history stored?
It remains in extension-local storage in the current browser profile, up to 30 entries by default, with no current EidoPrompt server synchronization.
Does it work in both Chrome and Edge?
The current Manifest V3 build targets Microsoft Edge and Chrome. Canvas, CSS background images, very small images, and obscured images may not expose an Analyze button.
Troubleshooting
No Analyze button on a page
Check Image Analysis, reload the extension, refresh the page, and retry on a sufficiently large image. Internal pages, canvas, and CSS backgrounds may not expose it.
Clipboard reading fails
Open the overlay import control and allow clipboard permission, or press Ctrl+V directly inside the overlay.
A web image cannot be read
The host may require login cookies or block hotlinking. Try the context menu, then use local import or region capture if the request still fails.
Vision Test fails
Check the endpoint, image-capable model ID, raw API Key, quota, and service status. 401/403 usually means authentication, 404 a path/model issue, and 429 a quota or rate limit.
The result is too generic
Retry after confirming the network is stable. In custom mode, confirm the model accepts image input, rerun Vision Test, or clear the configuration to return to the managed model.
The UI is still an older build
Choose Reload on the extension-management page, then refresh every test page. Already-injected content scripts are not replaced automatically.
A complete reset is needed
Sign out and clear history and API fields first. Uninstalling removes local settings, tokens, and credit profiles. Request deletion of the Supabase cloud account through the privacy-page contact details.
Send useful feedback
Contact us at the address below. Include browser version, EidoPrompt version, page type, reproduction steps, and the exact error. Never send an API Key or a private source image.
Contact email478839209@qq.com