本文来源:https://mp.weixin.qq.com/s/UvxLreRKl8aR8dja2dGIZQ
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
开发者 Maxime Rivest 做了这么一件事:他拿触控笔在自己的墨水屏平板上写下一行字,向汤姆·里德尔的日记本提问——「你是谁」。
字迹淡了,消失了。
几秒钟后,一行陌生的手写字迹从纸面同一位置浮现出来,回答了他的问题。读完之后,这行字又缓缓隐去,就像从未出现过一样。
这并非电影特效,而是他亲手写的代码,运行在一台 reMarkable Paper Pro 上。项目已经开源,仓库地址是 github.com/MaximeRivest/Riddle。
看到这条动态,我的第一反应大概和你一样:这到底是 AI 的功劳,还是设备本身就有的本事?普通人能不能把它照搬到自己的墨水屏上?会不会一顿折腾,把好好的平板搞成砖?
这几个问题的答案,其实都藏在一个数字里:reMarkable Paper Pro 这块屏,局部刷新约 260 毫秒,整屏重绘却要一两秒。这个细节直接决定了后面能不能做出「字迹自己浮现」的效果。往下我先讲屏幕,再讲 AI,最后说说你自己敢不敢动手刷机。
一、这效果到底是 AI 做的还是屏幕做的

先说结论:两者各占一部分,但真正决定「能不能做出来」的,其实是屏幕。

reMarkable Paper Pro 用的是一块 11.8 英寸彩色墨水屏,E Ink Gallery 3 工艺,官方标称 4096 级灰阶色彩,局部刷新压到 260 毫秒。这块屏最关键的特性,是局部刷新和渐变过渡足够细腻——能让一段文字淡出,也能让另一段文字浮现,而不是像老式电子纸那样一刷就整屏闪烁。伏地魔日记那种字迹自己消失、又自己冒出来的感觉,靠的就是这个物理层面的显示能力。换成普通黑白墨水屏,或者刷新慢、只能整屏重绘的设备,这个效果基本做不出来,AI 再厉害也救不了。
不过光有屏幕还不够,得有人把这块屏的显示能力和一套交互逻辑接起来。Maxime Rivest 做的事,是拿到设备的 root 权限,接管系统底层的书写驱动。用户用笔写字时,他的程序在中间拦截这段手写输入,转成文本发给 AI;拿到 AI 的文字回复后,再把回复渲染成手写笔迹的样子,写回屏幕的同一位置,并配合墨水屏的局部刷新,做出那种「浮现—消退」的动画感。
翻译成人话:AI 在这套流程里只干「把问题变成答案」这一件事。真正让效果成立的,是「拦截书写 — 调用 AI — 手写体反写回屏幕」这一整条工程链路,而这条链路的地基,是那块支持渐变显示、260 毫秒局部刷新的墨水屏。
二、为什么说 AI 不是重点,产品思路才是
平心而论,这件事并没有用到什么前沿的模型能力。把背后的 AI 换成任何一个能生成文字、能模拟手写风格的模型,效果都不会有明显差异。它不需要多强的推理,也不需要多长的上下文,就是一问一答,回答风格贴近日记本的语气就行。
那这个项目到底牛在哪儿?
我的判断是,它牛在把两个原本不相干的东西绑在了一起:墨水屏「字迹渐显渐隐」的物理特性,和《哈利波特》里「日记本会自己回话」的虚构场景。这两样单拿出来都不新鲜——墨水屏的渐变显示早就有了,伏地魔日记这个梗也被讨论过无数次。但从没人把它们做成一个能实际操作的原型,让你真的拿笔去写、真的看着字浮现又消退。
说实话,这件事的关键不在于 AI 又秀了什么新技能。关键在于 Maxime Rivest 先想清楚了这块屏幕最适合讲一个什么故事,再倒推出技术思路。顺序反过来,大概率做不出这个效果。AI 在这里只是接口后面那个负责生成文字的模块,换掉它,体验几乎不受影响;但要是抽掉「墨水屏的渐变特性」,这个项目就彻底立不住了。
这其实提示了一个更大的方向。墨水屏设备这两年一直在卷参数:刷新率、灰阶数、彩色还原度,谁的芯片跑分更高。但参数堆起来,用户买不买账,看的不是参数表。文石 BOOX 的 Tab Ultra 把刷新做到跟平板差不多顺滑,Kindle Scribe 标称 16 级灰阶,这些数字堆得再漂亮,真正决定体验的,是有没有人想清楚这块屏幕该配什么样的交互、什么样的场景。伏地魔日记这个项目就是个例子:它用的硬件参数市面上早就有,也没人拿它当卖点炒过,直到有人把它跟一个具体故事绑在一起。
三、普通人现在能不能自己动手复现

先说清楚一件事:这玩意儿装不了。你没法在 App Store 或者 reMarkable 自己的应用商店里搜到它、点一下「安装」就完事。它就是一份挂在 GitHub 上的开源代码,写稿这几天我看了下,star 数刚过 800。想跑起来,得先拿到设备的 root 权限,直接接管系统的书写驱动。

目前这个项目只在 reMarkable Paper Pro 的特定系统版本上跑得起来。别的墨水屏设备,包括其他品牌的电子纸平板,甚至 reMarkable 自己的其他系统版本,能不能直接照搬,Maxime Rivest 本人的说明里也没有打包票。
更现实的问题是风险。root 权限意味着你要绕过厂商的官方限制,接管驱动意味着你在动系统底层的东西。reMarkable Paper Pro 新机含 Marker 笔的官方售价约 579-629 美元起,市面上常见的 599 美元报价其实是官方翻新版的折扣价;一旦自己刷机,保修基本作废。操作顺利,你能感受到伏地魔日记那种沉浸感;操作不顺利,轻则系统卡死重启,重则设备变砖。这跟点一下「安装」按钮完全是两码事,更像是极客圈子里的一次实验性折腾。
所以我的建议很朴素:感兴趣的话,先去看代码,理解它是怎么拦截书写、怎么调用 AI、怎么把回复渲染成手写体的,这个思路本身就值回票价。真机尝试这件事,我建议再等等,看有没有人做出更安全的封装版本,或者 reMarkable 官方会不会把这类交互吸收进正式能力里,那时候再上手,风险小得多。
结语
这件事最容易被讲成「AI 又整了个新花活」,这么理解就把重点看歪了。

这套东西里,AI 换掉基本不影响体验,墨水屏的渐变显示特性一拿掉,这个项目直接散架。我的判断是,接一个 AI 接口太容易了,随便一个应届生半天就能调通 OpenAI 的 API。稀缺的是那种愿意花时间琢磨显示屏的物理脾气、再给它配一个恰到好处故事的人。这样的人,一个十人团队里未必凑得出一个。
对关注墨水屏和电子纸生态的人,我建议盯紧一件事:会不会有更多团队沿着「材质特性加场景想象力」这条路往下走。这个项目本身能不能变成 reMarkable 官方能力,其实没那么关键。墨水屏能做的渐变、局部刷新、低功耗常显,这些特性被真正用出彩的案例,掰着手指头都数得过来。
对普通用户,现在合理的动作是:去 GitHub 看看那 800 多个 star 背后的代码,理解它怎么拦截书写、怎么调用 AI、怎么把回复渲染成手写体,这个思路本身就值回票价。真机 root 这件事,我的建议是再等半年,等有人出个不会让设备变砖的稳定封装版本,那时候上手,风险小得多。

Developer Maxime Rivest did something like this: he used a stylus to write a line on his e-ink tablet, asking Tom Riddle's diary — "Who are you".
The writing faded and disappeared.
A few seconds later, a line of unfamiliar handwriting emerged from the same spot on the paper, answering his question. After he finished reading, the line slowly faded away, as if it had never appeared.
This wasn't a movie special effect; it was code he wrote himself, running on a reMarkable Paper Pro. The project is already open source; the repo is at github.com/MaximeRivest/Riddle.
Seeing this demo, my first reaction was probably the same as yours: is this thanks to AI, or is it the device itself? Can ordinary people port it to their own e-ink screens? Could tinkering with it turn a perfectly good tablet into a brick?
The answers to these questions all hinge on one number: the reMarkable Paper Pro's screen takes about 260 milliseconds for a local refresh, but a full-screen redraw takes one to two seconds. This detail directly determines whether the "writing emerges by itself" effect can be achieved. Below, I'll start with the screen, then AI, and finally discuss whether you'd dare to flash your own device.
1. Is this effect done by AI or the screen
First, the conclusion: both play a part, but what really determines whether it can be done is the screen.
The reMarkable Paper Pro uses an 11.8-inch color e-ink screen with E Ink Gallery 3 technology, officially rated at 4096 levels of grayscale color. Its key feature is that local refresh and gradient transitions are fine enough—letting a block of text fade out and another fade in, rather than flashing the whole screen like old electronic paper. The feeling of Voldemort's diary writing disappearing and reappearing on its own relies on this physical display capability. On an ordinary black-and-white e-ink screen, or a device with slow refresh and only full-screen redraw, the effect basically can't be done; no matter how powerful AI is, it can't save it.
But the screen alone isn't enough; someone has to connect its display capability with a set of interaction logic. What Maxime Rivest did was gain root access to the device and take over the system-level handwriting driver. When the user writes with the pen, his program intercepts the handwriting input midstream, converts it to text, and sends it to an AI; after receiving the AI's text reply, it renders it to look like handwriting and writes it back to the same spot on the screen, using the e-ink screen's local refresh to create that "emerging-fading" sense of animation.
In plain terms: in this workflow, AI only does the one thing of turning the question into an answer. What really makes the effect work is the whole engineering chain—"intercept handwriting → call AI → write handwriting back to the screen"—and the foundation of that chain is the e-ink screen that supports gradient display and 260-millisecond local refresh.
2. Why AI isn't the point; the product thinking is
To be fair, this didn't use any cutting-edge model capability. Swap the AI behind it for any model that can generate text and simulate a handwriting style, and the effect wouldn't differ noticeably. It doesn't need strong reasoning or a long context—just a question and an answer, with the reply in the diary's tone.
So where exactly is this project impressive?
My judgment is that it's impressive because it ties together two previously unrelated things: the e-ink screen's physical trait of "handwriting gradually appearing and disappearing," and the fictional scene in Harry Potter where "the diary writes back by itself." Neither is new on its own—e-ink gradient display has existed for a long time, and the Voldemort diary meme has been discussed countless times. But no one had turned them into an actually operable prototype where you really pick up a pen, write, and watch the characters appear and fade.
Honestly, the key here isn't that AI showed off some new skill. The key is that Maxime Rivest first figured out what story this screen is best suited to tell, then worked backward to the technical approach. Reverse the order, and the effect probably wouldn't happen. AI here is just the module behind an interface responsible for generating text; replace it, and the experience is barely affected. But remove the e-ink screen's gradient trait, and the project completely falls apart.
This actually points to a larger direction. E-ink devices have been competing on specs for the past two years: refresh rate, grayscale levels, color accuracy, whose chip scores higher. But piling up specs doesn't determine whether users buy in; what matters is whether anyone has figured out what kind of interaction and scenarios this screen should serve. For example, the Onyx BOOX Tab Ultra makes scrolling almost as smooth as a tablet, and the Kindle Scribe claims 16 levels of grayscale—these numbers can be as pretty as they want, but what really decides the experience is whether someone has thought through what story fits the screen. The Voldemort diary project is an example: its hardware specs have been available on the market for a while, and no one hyped them as a selling point, until someone tied them to a concrete story.
3. Can ordinary people reproduce it themselves right now
First, one thing needs to be clear: this thing isn't installable. You can't find it in the App Store or reMarkable's own app store and tap "Install" to finish. It's just an open-source repo on GitHub; as of writing, its star count has just passed 800. To run it, you first need root access on the device and to directly take over the system's handwriting driver.
Currently, this project only runs on specific system versions of the reMarkable Paper Pro. Whether other e-ink devices, including other brands of e-paper tablets or even other system versions of reMarkable's own hardware, can be directly ported, Maxime Rivest's own notes don't guarantee.
The more realistic issue is risk. Root access means bypassing the manufacturer's official restrictions; taking over the driver means touching low-level system stuff. A new reMarkable Paper Pro with the Marker pen starts at an official price of about $579–$629; the common $599 price you see is actually a discount for the official refurbished version. Once you flash it yourself, the warranty is basically void. If it goes smoothly, you get the immersive Voldemort-diary feeling; if it doesn't, at best the system freezes and reboots, at worst the device becomes a brick. This is nothing like tapping an "Install" button—it's more like an experimental hack in the geek community.
So my advice is simple: if you're interested, go read the code first and understand how it intercepts handwriting, calls the AI, and renders the reply as handwriting—that idea alone is worth the price of admission. As for trying it on a real device, I'd suggest waiting to see if someone makes a safer packaged version, or if reMarkable's official team absorbs this kind of interaction into a formal feature; at that point, the risk will be much lower.
Conclusion
The easiest way to tell this story is "AI pulled off another new trick"; that's missing the point.
In this setup, swapping out the AI barely affects the experience; remove the e-ink screen's gradient display trait and the project collapses. My judgment is that hooking up an AI API is too easy—any new grad can get the OpenAI API working in half a day. What's scarce is the kind of person willing to spend time understanding the physical temperament of a display and then pairing it with a perfectly matched story. You might not find one such person in a ten-person team.
For those following the e-ink and e-paper ecosystem, I suggest keeping an eye on one thing: whether more teams will keep going down the path of "material traits plus scene imagination." Whether this project itself becomes an official reMarkable feature isn't that crucial. The cases where e-ink's capabilities—gradient display, local refresh, low-power always-on—are truly used brilliantly are few enough to count on one hand.
For ordinary users, the sensible move right now is: go to GitHub and look at the code behind those 800-plus stars, understand how it intercepts handwriting, calls the AI, and renders the reply as handwriting—that idea alone is worth the price of admission. As for rooting a real device, my advice is to wait another six months for someone to release a stable packaged version that won't brick the device; at that point, the risk will be much lower.