Glance brings the FaceID-like experience of your iPhone to a Mac near you. Unlock your Mac with a glance — no typing, no reaching for the TouchID key. Everything runs on-device using Apple's Vision and Core ML frameworks, so your face data and your Mac password never touch the internet. The UI is built into your Macbook's notch with fluid dynamic island like animations.
output.mp4
本项目基于 Jonathan Zhou（@jonnyoo） 开源的 Glance 修改。感谢原作者提供面部识别、锁屏解锁和刘海交互的基础实现。
- 原版官网： tryglance.app
- 原作者 GitHub： @jonnyoo
- 原版仓库： jonnyoo/glance
这是基于原版开发的修改分支，以下新增功能与修复由本分支提供，不代表原作者的官方发布。
下表列出本分支相对于开发起点的主要改动；原版后续更新可能有所不同。
| 项目 | 本分支的新增或调整 |
|---|---|
| 自定义识别动画 | 接入 ShaderCN 的 33 种光球动画，通过原生 Metal 在本地渲染。识别中对应 Thinking，成功对应 Speaking，失败对应 Idle。保留原有动画样式。 |
| 动画设置 | 在设置中选择光球、预览状态、调整尺寸、颜色、驱动强度和各动画参数；各状态独立保存，支持重置和复制配置。 |
| 连续面部采集 | 改为缓慢转头一圈连续采集，无需单独拍摄或导入正脸照片。系统自动从摄像头帧中选择参考帧和各角度样本。 |
| 异步整理与补帧 | 采集后显示雾化背景和 Thinking 动画，继续采集 4 秒真实补充帧，再后台筛选清晰度、对齐质量和身份一致性。补采预算为 12 秒，模型处理可能增加耗时；不足的角度可单独补齐。 |
| 会话与安全存储修复 | 统一钥匙串存储查询，兼容旧存储；同步设置页的会话状态，处理配置完成后的密钥缺失、会话无法解锁问题。增加保留旧加密数据的重新配置入口，并持久保存当前存储区标识。 |
| 熄屏与唤醒识别 | 补齐显示器休眠、系统唤醒和屏保退出事件的处理，合并重复事件；增加锁屏状态就绪重试和摄像头预热等待，减少亮屏后漏触发识别。 |
| 中英文切换 | 新增 设置 → 通用 → 语言，支持简体中文和英文即时切换并保存选择，覆盖主要设置、采集提示和动画参数名称。 |
| 回归验证 | 增加采集、凭据、钥匙串存储选择、唤醒事件和多语言检查，运行方法见 验证说明。 |
本分支仍使用普通摄像头与本地 Vision / Core ML 识别，不具备 Apple Face ID 的深度传感器能力。补帧来自真实摄像头画面，不生成虚构的面部样本。旧密钥确实丢失时，重新配置会保留旧数据，但无法解密或恢复旧数据。
代码已通过编译和逻辑回归检查；Touch ID、摄像头采集及熄屏解锁的完整流程仍需使用正常签名的版本进行实机验证。
Warning
MacBooks don't come equipped with the depth sensors that make iPhone FaceID trustworthy and secure. An iPhone builds a 3D map of your face; a MacBook webcam sees a flat 2D image. That means:
- Glance defeats, with reasonable confidence, a printed photo and a photo on a phone screen (heavy liveness detection must be turned on)
- Glance does not reliably defeat a video of you
- macOS has no API that lets a third-party app authorize a login, so Glance unlocks by typing your stored password on the lock screen
Glance is a convenience feature, not a security upgrade. Only continue if you accept the tradeoff.
Requirements:
- macOS 15 Sequoia or later
- Apple Silicon or Intel Mac
以下下载链接为原作者发布的官方版本，不包含本分支的上述改动。使用本分支请参照 Building from source 自行构建。
Open the .dmg file and drag Glance to /Applications, then open it.
- 应用图标： 使用完整 macOS AppIcon 资源集，构建产物包含
AppIcon.icns，兼容 macOS 15 及以后版本。 - 顶部弹窗： 固定窗口尺寸并绕过菜单栏自动避让，显示或切换屏幕后重新校准顶部位置。
- 失败重试： 默认在首次失败后每隔 2 秒重试，最多追加 3 次（共 4 次尝试）。等待和重试期间保持面板展开、摄像头运行，不重新播放入场动画。可在通用设置中关闭自动重试或调整次数；每次尝试重新验证身份和活体，解锁或休眠后停止。
- VTA（Voice to ChatGPT）： 在 设置 → 通用 → VTA 开启并授权麦克风、语音识别和辅助功能。默认按住 F6 录音，可绑定其他按键或组合键；松开后使用 macOS 本机语音识别转写，再尝试在 Mac 版 ChatGPT 新建聊天并发送。支持中文和英文，单次最长 60 秒。
- VTA 动效： 顶部弹出录音与音量动效；转写和发送时显示 Thinking，成功对应 Speaking，错误对应 Idle。关闭浮层、禁用 VTA 或锁屏会取消任务。
VTA 使用已安装并登录的 ChatGPT 应用，不需要 OpenAI API Key。录音不落盘，本机不支持所选语言的离线识别时会提示错误；只有转写文本会在发送时交给 ChatGPT。桌面发送依赖 ChatGPT 的辅助功能控件；找不到新聊天、输入或发送控件时会停止并保留文本，可在 VTA 设置中复制或手动重试。应用更新可能改变这些控件，需要实机验证兼容性。
| Permission | Why |
|---|---|
| Camera | To see your face. Frames are processed in memory and never written to disk. |
| Accessibility | To type your password into the lock screen. |
| Touch ID | Gates the key that encrypts your face data and password. |
| Microphone & Speech Recognition (optional VTA) | Record while the shortcut is held and transcribe locally. |
| Accessibility (VTA) | Detect the held shortcut and operate New chat / Send in the ChatGPT Mac app. |
- Launch the app and follow onboarding. After authenticating secure storage, slowly move your head in a circle; there is no separate front-photo step. Glance collects the eight surrounding angles continuously and automatically selects a reference frame. The capture view blurs and shows a Thinking orb while the camera continues acquiring real supplemental frames. A four-second refinement interval is followed by background quality ranking and targeted retries, with a twelve-second acquisition budget (model processing can add time). Any missing angles can be filled without repeating the whole turn. Only 512-number embeddings are saved; temporary camera images are discarded.
- Enter your Mac password once, encrypted behind Touch ID.
- When your Mac locks or wakes from sleep, the animation appears in the notch and starts searching for a face.
- If it's you — and the liveness checks agree you're a real person — Glance types the password and you're in.
| Feature | Description |
|---|---|
| Face unlock | Triggers on wake, on lock, or on pressing space at the lock screen. Pick any combination. |
| Multiple identities | Enroll several people, or several versions of yourself — with glasses, a beard, different lighting. Toggle any of them off without deleting. |
| Liveness checks | Watches for the motion and reflections that separate a real face from a photo. Light or Heavy strictness, or off. |
| Notch UI | A closed pill that expands into a scan animation with success and failure states. Hover to retry — or turn animations off entirely and Glance stays invisible. |
| Camera & display | Choose which camera to use, including different cameras for the built-in display vs. an external monitor. |
| Auto-locking sessions | The Touch ID session re-locks itself after an idle period you choose, so an unattended Mac doesn't stay authorized forever. |
| Trackpad haptics | Hovering over the notch will trigger haptics |
| Notchless Mac support | Macs without a notch will be replaced with a pill-shape, dynamic island style design. |
| Your data, your call | Edit or delete your enrolment or stored password at any time. The encrypted files are removed immediately. |
Glance is designed to keep biometric data and credentials on-device.
Glance never stores camera images. During enrollment, each captured face is converted into a 512-dimensional embedding using an ArcFace-based Core ML model. The original frame is then discarded.
Embeddings are stored locally and encrypted with AES-GCM.
Your Mac password is stored as encrypted data and is never written to disk in plaintext. The encryption key is a 256-bit AES key stored in the macOS Keychain, protected by userPresence — requiring Touch ID or your device password.
The key is only held in memory while an authorized Glance session is active.
If an older encrypted store exists but its key is inaccessible (for example after a signing change), setup and locked Settings pages offer Recover secure setup. After explicit confirmation and system authentication, Glance selects a separate new storage namespace. Previous encrypted face files and Keychain items remain untouched; this does not recover the lost key or decrypt the old data. Keep the original signed app/key if you need the old enrollment. The active vault identifier is persisted alongside the encrypted face files. New Keychain items explicitly use the data-protection backend; queries also support legacy items.
Choose General → Language to switch immediately between 简体中文 and English. The selection persists across launches.
Display sleep, system wake and screensaver exit feed a coalesced wake trigger. Glance waits up to three seconds for authoritative lock/display readiness and gives the camera a separate warm-up budget before timing recognition. A session must already be authenticated and On wake enabled; Glance does not display authentication prompts on the lock screen.
Glance won't type your password simply because a face matches. An unlock requires all of the following:
- A valid Glance session is authorized.
- The Mac is actually at the lock screen.
- An enabled identity matches above the configured similarity threshold.
- Liveness checks accept the detected face.
- Accessibility permission is available to enter the password.
Face recognition and liveness detection run independently and must both succeed before the password is entered.
Face recognition, face enrollment, and liveness detection run entirely on-device using Vision and Core ML. Glance does not send face data, camera frames, or credentials to a server.
Five independent cues over a rolling ~2s window, in two roles:
- Deny cues are evidence of a spoof — screen glare, or a device-shaped rectangle framing the face. Either one fails the scan outright and overrides anything else.
- Confirm cues are evidence of a real face — flat-vs-3D landmark geometry, nose parallax across head turns, blinks. Any one is enough, and their absence is never a failure, since a live person can sit still and not blink.
Light detection only include deny cues. Heavy detection includes both deny and confirm cues.
Face Lab is a hidden debug console to test face recognition and liveness detection with real values.
To open it: Settings → About, then click the app icon 5 times. A debug section should appear in the sidebar.
- macOS 15+
- Xcode 26+
- Download or clone this repository using its GitHub Code button, then open the directory containing
glance.xcodeproj. Cloningjonnyoo/glanceinstead gives you the original upstream version, without this branch's changes. - Open in Xcode:
open glance.xcodeproj- Run the project:
- Click
runor pressCmd + R.
Not currently accepting PRs. Feel free to fork this project.
App feedback goes to tryglance.app/feedback.
- Jonathan Zhou — @jonnyoo — original creator of Glance, whose open-source work this branch builds on.
- ShaderCN / XorDev — the orb animations adapted for native Metal rendering; attribution and terms are preserved in ShaderOrb-LICENSE.txt.
- The Boring Notch — for the notch window physics.
- InsightFace — the ArcFace model doing the recognition.
- Alcove — big design inspiration.
MIT © Jonathan Zhou
The bundled ShaderCN / XorDev shader assets have separate non-commercial, attribution-required terms. See ShaderOrb-LICENSE.txt; the MIT license does not replace those asset-specific terms.