After Depseek, the next logical step was Tiktok. And it has arrived. Bytedance researchers (the company behind the social network) have developed an AI system that transforms individual photographs into realistic videos of people speaking, singing and moving naturallyan advance that could transform entertainment and digital communications. And confuse us.
The new system, called Omnihuman, generates whole body videos that show people making gestures and moving so that they coincide with their way of speaking, surpassing the previous AI models that could only encourage faces or upper parts of the body.
How Omnihuman Use 18,700 hours of training data to create a realistic movement. The example that has gone viral of this technology is one in which Albert Einstein can be seen talking about the importance of science and its relationship with emotions … something that obviously never happened.
https://www.youtube.com/watch?v=n6hkcs2pj0q
“The end of extreme human animation has experienced notable advances in recent years -those responsible in a study published in Arxiv -point out. However, Existing methods still have difficulty climbing as large general models of video generationwhich limits its potential in real applications. ”
The team trained Omnihuman with more than 18,700 hours of human video data (the equivalent of more than 2 years) using a novel approach that combines multiple types of entries: text, audio and body movements. This “OMNI-Conditions” training strategy allows AI to learn from much larger and diverse data sets than the above methods.
“Our key idea is that the incorporation of multiple conditioning signals, such as text, audio and pose, during training It can significantly reduce data waste, ”add the authors.
Technology marks a significant advance in the media generated by AI, demonstrating capabilities ranging from the creation of videos of people pronouncing speeches Until the representation of subjects playing musical instruments. In the tests, Omnihuman surpassed existing systems at multiple quality reference points.
Development arises amid an increasingly intense competition in the generation of videos with AI, with companies such as Google, Meta and Microsoft that seek similar technologies. The advance of Bytedance could give its Matrix Tiktok company an advantage in this rapid evolution field.
Industry experts say This technology could transform entertainment productionthe creation of educational content and digital communications. However, it also raises concerns about the possible improper use in the creation of synthetic media for deceptive purposes in different areas. Something for which, There is still no clear prevention measure.