What a digital twin actually is (and when you don't need one)
The phrase gets used for three completely different things, which is why nobody can agree on what it costs. So before anything else, three definitions.
A video avatar is not a digital twin
Upload a selfie to a hosted platform and you get a talking head on a flat background. It is fast, it is cheap to start, and for a lot of jobs it is exactly right.
What it cannot do: gesture, stand up, walk into a room, turn to a second camera, or exist in a scene you designed. It is a head and shoulders, rendered on somebody else's server, billed by the minute, for as long as you keep paying.
A deepfake is not a digital twin either
A deepfake takes a real person's face and puts it somewhere they never were, usually without asking. Same technology family. Completely different ethics.
A digital twin is built with the person, for the person, with a scope for what the face is allowed to say and an off switch they control. If someone selling you one will not draw that line clearly and in writing, that tells you something about the rest of the engagement.
So what is it
A photoreal 3D version of a specific person — rigged, animated and driven — that can say new things without that person being in the room.
Which means the founder does not re-record the onboarding video when the pricing changes. The trainer does not fly to four cities. The face of the brand can be in six languages on Monday. The person stays the same; the words change.
The one question to answer first
What will it say next month that it cannot say today?
If the answer is nothing, you do not need a twin. You need one good video, filmed once, and you should go and film it — it will be better and it will cost less.
A twin earns its cost when the words keep changing and the face has to stay the same. Updates. Languages. Versions. Volume. If your content is static, a twin is an expensive way to make one video, and anyone who tells you otherwise is selling.
Where the money actually goes
Studios quote well into five figures, and a lot of that is real work: scanning, grooming, rigging, a pipeline. But a large part of the price is that clients cannot tell which parts are hard.
Here is what is hard: the face rig and the performance system — the thing that decides how a written sentence becomes a delivered one. Here is what is not hard any more: the fact that it talks at all. Unreal Engine is free. MetaHuman is free and licensed for commercial use. Video-to-motion turned body animation into an evening rather than a contract.
So the honest number sits well below the studio quote and well above a monthly subscription — and unlike the subscription, at the end of it you own the thing.
Own it or rent it
Every decision in this space reduces to that, and it is not really about quality.
Rented is fast to start and disappears the day you stop paying. Your brand's face lives on someone else's servers under someone else's terms. Owned takes longer to set up, runs on your hardware, and keeps working if the company that built it vanishes.
Neither is wrong. But most people rent by default, without ever being asked the question.
I build these. If you want a straight answer about whether one fits your case — including "no" — describe it in one sentence and I will tell you.