Tencent Cloud, the arm of the Chinese tech giant for online services in the computer clouds, has introduced a platform for the production of “digital people” in the form of comparatively natural-looking bots. The service allows users to create videos of any person based on a three-minute moving image and 100 sentences of speech material for a fee of around 130 euros, which could pass as deepfakes. The bot generator uses Tencent’s proprietary artificial intelligence (AI) system to generate the videos.
Deepfake videos of “digital people” in 24 hours
Chinese media report that the cloud service could within 24 hours be trained on the basis of the recordings provided so that it produces a more or less convincing deepfake video. The British online magazine The Register According to Tencent, it received confirmationthat the service is available in Chinese and English. Digital human creation is offered in five variants: 3D realistic, 3D semi-realistic, 3D cartoon, 2D real person and 2D cartoon. The digital figures are therefore available as half or full bodies. Some aspects like background and hue are said to be customizable.
According to initial assessments, the videos created avoid the flat intonation and uniform speech rhythm that conventional acoustic models struggle with. A Tencent method for adjusting the timbre of small samples will be used, which is based on acoustic models with deep learning and neural network for voice recordings. Chen Lei, general manager of Tencent Cloud Intelligent Digital Human Products, told The Register that it aims to build an automated, AI-driven “intelligent digital human factory” and rely on a self-service platform for production, sales and service.
Recommended Editorial Content
With your consent, an external survey (Opinary GmbH) will be loaded here.
Always load polls
Commercials market – popular form of e-commerce
Tencent reportedly sees a large market in using the creations for live-streamed commercials. This is a popular form of e-commerce in China. It is also about creating “doctors, lawyers and other professionals” digitally. Tailor-made questions and answers can be created for the creations, turning them into a kind of chatbot. But the risk of misuse is great: Realistic-looking photos, audios or videos in which people are placed in new contexts with the help of AI are increasingly considered deepfakes. Words are often put into the mouths of third parties that they have never said before.
AI-created images are becoming increasingly difficult to distinguish from real ones. Prime examples of this are the attention-grabbing alleged depictions of the arrest of former US President Donald Trump and the Pope wearing a white designer padded jacket. The Office for Technology Assessment of the Bundestag (TAB) refers in a current study on ChatGPT & Co. on risks related to multimodal AI models: Coupled with the possibility of generating images, videos or voices, completely new future application possibilities could be imagined, for example in the generation of deepfakes. Twitter CEO Elon Musk recently warned of the increasing number of fakes impersonating him to promote cryptocurrency scams. Europol fears that deepfakes could become a standard tool for criminals. Chinese regulations are intended to prevent such scenarios.
(tiw)
