D-Human 数字人平台

What D-Human is

D-Human is a Chinese digital-human platform that clones a person's appearance and voice from a short video so the resulting avatar can present scripted content, offered as software, an interface, or a custom deployment. It targets enterprise and media production, and the cloning capability is the whole product rather than a feature within it.

What it does

You record roughly eight minutes of footage with coaching support, the vendor trains a model and deploys it, and you then type a script, select the digital person and generate video, with optional voice cloning from short audio samples. The loop is film, train, generate.

Ready templates cover knowledge explainers, finance and insurance, corporate promotion, live selling and news reading. The vendor claims high fidelity and fast delivery, and supports software, interface and original-equipment integration. Pricing was described as low-cost custom project work rather than a published plan, so a sales conversation is required before any comparison.

Who it is for

It is aimed at Chinese enterprises, media groups and agencies producing presenter-led video at volume, where replacing routine filming with training a model once is economically sensible. Organisations with a recurring scripted format fit far better than those making one-off films.

What to keep in mind

The consent position here needs to be handled with unusual care, and the page itself is the reason. It openly advertises cloning yourself or another person and shows celebrity voice examples. Using a real person's face or voice without written authorisation risks violating portrait and voice rights, and in the wrong hands the same capability enables fraud. That is not a hypothetical risk of the category; it is the capability being sold.

So treat written authorisation as non-negotiable, and treat synthetic presenters as requiring disclosure to viewers. An audience that believes it is watching a person will extend trust that a synthetic presenter has not earned, and in finance and news that difference matters.

Three contractual points should be settled before training anything. Who owns the cloned model, whether it can still be used after the contract ends, and how the source footage is deleted. These are easy to leave open and expensive to reopen, because a trained model is the asset and its ownership determines what happens when the relationship does.

Pricing is custom, so budget for a project rather than a subscription, and confirm what happens when you need a new script, a new look, or a change of spokesperson. There is one more thing to require in writing: a description of how the trained model is stored and who can invoke it. A cloned face and voice sitting in someone else's infrastructure is an asset with obvious misuse potential, and knowing who holds the ability to generate with it is as important as knowing who owns it.