I. First, a Thought Many People Have Had but Few Say Out Loud

After someone leaves, what they leave behind undergoes a slow evaporation.

The chat history on the phone is still there at first. Photos lie in some corner of the album, not opened for years. The chair they used to sit in is still in its place, but no one sits in it anymore. The things they said, you remember the gist, but the exact words are gone.

You'd say, I remember them.

But you know in your heart: remembering, and being able to see them again, are two completely different things.

In the past two years, technology has made "seeing them again" less distant. With enough photos, you can rebuild a three-dimensional likeness; with enough text, you can teach a language model their tone of voice. Technically, these things are no longer hard.

The hard part is something else — a question almost no one asks:

Once it's built, where do you put it?

II. First, the Three Routes Available Now, Each With Its Own Problems

I grouped the approaches I could find on the market into three categories by "where the stuff is stored." Here I only describe the mechanics, not judge good or bad; you can decide for yourself.

Route One: Cloud AI Memorial Services

You upload photos, voice, and chat logs to the other party's server, and they generate a conversable likeness for you.

Mechanically, there are three things you must accept:

• All materials must leave your hands. What you upload is your most private thing — the face of a deceased relative, their voice, the words they said. Once uploaded to someone else's server, you can no longer control how it's used, how long it's kept, or who sees it.
• What you signed is a terms of service, not ownership. The content in the account is yours, but the service itself belongs to the other party. If they change terms, remove features, or shut down, your account could be empty overnight.
• It must accept compliance constraints. Providing deep-synthesis services to the public requires a series of obligations like labeling, consent, and filing. This means such features can be adjusted or taken offline at any time due to policy changes — not because they want to, but because they must.

In one sentence: you've put something you cannot bear to lose a second time in a place you can't control.

Route Two: Local Files + Chat Apps

No fuss — photos on a hard drive, chat screenshots in cloud storage, flipped through occasionally.

The problem with this route isn't technical, it's the form. Photos and text are two-dimensional; they are objects "to be looked at," not spaces "to be entered." The act of flipping through an album, versus "going to a place to see them," gives a completely different psychological feeling.

And it depends heavily on your self-discipline. You won't flip through it often, because opening the album is an act that requires resolve.

Route Three: Build Your Own 3D Space

Using some 3D tools, or a game engine, build a room yourself, put models, photos, and sounds in it, and walk in anytime.

This route is right in form — it's "a place," not a folder.

But it has two thresholds: first, you must know modeling, scene arrangement, and lighting; second, what you build is usually just a scene file, scattered in some directory, with no unified management, and it won't grow over time.

III. The Real Question Isn't "Can We Build It," But "Can We Keep It"

Putting these three routes together, you'll find they're actually answering the same question:

How long can this thing last?

Not "can it be built" — technically that's long possible. But will it still be there in thirty, fifty years.

Something stored on someone else's server has a lifespan equal to that company's, or to that business line's. And a company's business line on average doesn't survive more than a few years. You wouldn't want something about your father to depend for its lifespan on some product manager's quarterly review.

Something stored on your own hard drive has a lifespan equal to your willingness to maintain it. That at least is something you control.

So the answer is actually clear: only what's in your own hands can be said to be "forever."

This isn't a sentimental line; it's an engineering conclusion.

IV. Why This Must Run on Your Own Machine

What I built is a browser-based 3D virtual world system; users download it and deploy it on their own computer or server, accessed through a browser. The entire system passes through no third-party server.

This architecture was designed for other purposes originally, but applied to digital memorials, it happens to match on every point.

4.1 Materials don't leave home

Photos, video, sound, text, 3D models — all stored on your own hard drive. You don't need to upload any of it to anyone.

This is a bonus elsewhere; in this scenario it's the premise. You can judge for yourself: should these things pass through someone else's server.

4.2 It lasts as long as you want it to

The system runs on your own machine, depending on no external service. Whether our company is open or closed has nothing to do with your world. Whether the server is renewed has nothing to do with it either.

Its only lifespan limit is the lifespan of that hard drive. And hard drives can be replaced.

4.3 It's a "place," not a folder

This is the essential difference between 3D space and an album.

You can build them a room. Arrange it like the study they had in life — the books they actually read on the shelf, their usual cup on the desk, their photo on the wall. Prepare a 3D likeness of them to place there — standing, sitting, by the window, your choice.

You can walk in anytime, sit for a while.

This action has a low threshold — just open a browser; no client to download, no VR device to wear, accessible on a phone too.

Going from "flipping through an album" to "going to see them" is a completely different psychological act. The former is recollection; the latter is a visit.

4.4 Models, scenes, and materials have one unified place to manage

The underlying layer of my system is a framework, not a finished product. It itself comes with a set of asset-management and scene-organization capabilities:

• 3D models (standard GLB/GLTF format) are stored in the system as assets, scheduled and called uniformly;
• Scenes, buildings, and props can be built yourself, or added layer upon layer;
• The underlying layer is open, not a black box — if you have special needs, you can modify it yourself or have someone modify it.

This means it's not one-time. Put a photo this year, add a video next year, complete their study the year after. It can grow slowly with you.

4.5 Whether to let others in is your decision

The system supports multiple people online at once. You can let family members in too — stay together in the same space for a while.

You can also let no outsider in at all, just yourself.

This switch is in your hands, not the platform's.

V. On "Talking to Him," I Urge You to Think Clearly

This is the paragraph I most want to say in this whole article; please read it carefully.

Technically, getting a 3D likeness to talk with you is achievable: organize the text they left — chat logs, diaries, letters, social posts — into a corpus, pair it with a language model, and connect it to this space. If the model also runs locally, then the entire conversation process sends not a single byte of data out of your home. This is something a cloud solution can never do.

I lay out this route clearly because it's feasible, and my system reserves an AI interface, so you can connect your own model.

But personally I don't recommend making it the default, for three reasons.

First, how much it resembles them depends on how much real material you left. With little material, what it generates will only be a vague shadow of tone. What that shadow says may hurt more than silence. What you hope for is them; what you get is the statistically most probable next piece of text. You should be prepared for this gap in advance.

Second, it may make saying goodbye difficult. A grief with a beginning and an end has its psychological function. A shadow you can talk to anytime could, in theory, keep that process from ever ending. I'm not a mental-health professional and won't draw conclusions, but I think this is a possibility worth treating with caution.

Third, and most important: that is not them.

I'm unwilling to write this line vaguely. No matter how complete the corpus, no matter how advanced the model, what conversation generates is a piece of probability-based text. It is not consciousness, not memory, not the will that person left. Treating it as them is disrespectful to them, and to yourself.

So the form I recommend is this:

By default build a memorial hall, not a stand-in.

Inside, place their likeness, their voice, their footage, their actual words, the room they loved in life. Use original audio, not synthesis; place original words, not generation. When you miss them, go in and walk around, listen to a real recording of them, see their real face.

If you're sure you're ready, then turn conversation on as an optional switch — and clearly know what you're talking to.

This is what I consider a responsible approach. Many products on the market sell gimmicks on this, but I don't want to do that.

Remembrance matters more than resurrection. Truth matters more than realism.

VI. If You Actually Want to Start, Here's Roughly the Process

I won't write "three steps and done," because this genuinely takes your time. I'll write each step honestly, including what you must prepare yourself.

To be clear: this system is a framework, not a finished product. What it gives you is a complete set of runnable foundation — asset management, scene building, model loading, multi-user access, data self-custody. What exactly to build on top is for you to do with your own hands.

This sounds like a shortcoming, but I think in this scenario it's actually right. A memorial hall should be built by hand. Every hour you spend on it is itself part of the matter.

VII. In Closing

I built this product because I believe one thing: some things should be kept in your own hands.

Photos, footage, memory, the traces a person left in this world — the value of these things doesn't depend on how realistic they're made, but on whether they can always stay in a place you can find anytime, that no one else can touch.

He is on no one's server. He is in your own hard drive, in the room you built with your own hands.

This is what I understand as "forever."

This article describes a personal-use scenario. Before using a deceased relative's portrait, voice, or other personal information, please confirm you have the consistent consent of close relatives; for content involving deep synthesis, please comply with the laws and regulations of your location. This article does not constitute any form of external service commitment.