After Deepsek, another Chinese AI model is trending. This is the Omnihuman of bytedance. Deepfake videos are growing rapidly. There is a flood of AI videos on social media. It is becoming difficult to differentiate between real and fake videos. Meanwhile, Tiktok’s parent company Bytedance has done something that people around the world are surprised.
Some researchers of bytedance have designed the Omihuman-1 model. It can make a video with just one photo and it looks similar to the real. It is difficult to identify whether the video is fake or real.
Currently, a lot of data is required to make Deepfake Video. For example, if you want to make a Deepfake video of a person, then dozens of photos and audio samples are seen. But now only one photo will do this work.
Resurreaurs of bytedance have prepared an AI system that can transform only one photo into a video -like video. Whether it is a way to speak, body movement or voice everything looks exactly real.
Many people consider it a big breakthrough for digital entertainment and communication. But it is also very scary in a way. Because in India, many types of scams are being done due to Deepfake videos and people are also dying in it.
People are blackmailed by making Deepfake Video. Money is cheated, but at the same time the lives of people are also being ruined. Cyber ​​criminals continue to blackmail someone by sending porn videos from Deepfake.
Talking about the new Omihuman model of bytedance, it generates not just facial, but also generates full body video from just one photo. In full body video, from hand gesture to body movement and their voice looks similar and a normal person can be deceived by seeing it.
Let us tell you that all the models of Deepfake videos till now make videos of face and upper body. But the openhuman model of bytedance is capable in moving the entire body including face.
The team of Bytedacne Researchers has used 18,700 hours of human video data to make Omnihuman. Text, audio and body movement have been used to train this model. The company has trained it with OMI condition, that is, all the movements have been captured like every 3D model so that the video looks absolutely real.
It is worth noting that all the big AI companies around the world are now working on AI video generation tools very fast. Many tools like Sora have come where you can give a genna -like video by giving prompt. Although this model is different from that, but the work is doing the same.
Omnihuman developers have said that they have used multiple conditioning signals to train this model. This includes text, audio and pose. By doing this, data waste can be avoided in training.
This model may be beneficial for some industries, but difficult for common users. Because cyber criminals can also blackmail people by misunderstanding it. So now it has become very important to know how to differentiate between real and fake videos.