(Dan Tri) - ByteDance, TikTok's parent company, surprised everyone when it introduced an AI tool capable of creating videos with realistic content.
ByteDance, the technology company behind the social network TikTok, has just introduced a new artificial intelligence (AI) model called OmniHuman-1, with the ability to create lifelike, animated videos.
ByteDance says OmniHuman-1 is based on deepfake technology, which allows it to create videos with extremely smooth movements and is very difficult to recognize as a product of AI. ByteDance confidently claims that its AI tool can create products with superior quality compared to other existing AI video creation tools.
Video of Albert Einstein's talk created by OmniHuman-1 from the scientist's image (Video: ByteDance).
According to ByteDance researchers, OmniHuman-1 only needs a single image and sound as a sample, such as a random person's voice or singing voice. Based on this data, the AI engine can generate videos of any length, with the voice or singing voice identical to the sample audio.
OmniHuman-1 can also edit and modify existing videos to change the content of the original video.
In addition, OmniHuman-1 also allows users to create new videos with a variety of visual and audio styles, such as animated videos or cinematic styles or lifelike videos.
Video created by OmniHuman-1 on user request (Video: ByteDance).
Users only need to give requirements on aspect ratio, body ratio (portrait, half body, full body or all in one), this AI tool can create videos with the highest authenticity, including movements, lighting, surrounding details...
ByteDance has yet to officially release OmniHuman-1 to the public, but the company has shared sample videos created by the tool.
A 23-second video was created from an image of Albert Einstein, in which the famous scientist gave a talk, with lifelike facial movements and lip movements.
Several other short videos created by OmniHuman-1 show people presenting with lifelike gestures and facial expressions.
Video of characters giving presentations, created by OmniHuman-1 from still images (Video: ByteDance).
Demonstration videos of OmniHuman-1's capabilities have left the online community and tech community in awe, with many admitting that this is the most realistic fake video they have ever seen.
ByteDance researchers said they have built an AI model capable of mixing data from images, videos and audio to create realistic videos, with facial proportions, body, facial expressions, actions... as human-like as possible.
The research team said that the videos created by AI have natural faces, head movements, gestures... that match the surrounding sounds and scenes, helping to increase the "realism" in the videos.
ByteDance has yet to reveal when OmniHuman-1 will be available to users.
Video created by OmniHuman-1 from images, with the scenery outside the train window changing realistically (Video: ByteDance).
The emergence of OmniHuman-1 shows the incredible capabilities of AI tools coming from China, but many are concerned that OmniHuman-1 could be used for malicious purposes, such as creating fake celebrity videos to share false information, scams, or even creating fake porn videos for blackmail.
The AI development race is becoming more intense between the US and China, as these two superpowers continuously introduce AI tools with superior features to defeat their opponents.
A few days ago, OpenAI also integrated a new feature into ChatGPT, allowing the tool to complete a report from a huge amount of data in just a few minutes, a task that would take humans hours to complete.
Deepfake is an artificial intelligence technology that uses image synthesis techniques of someone to combine and superimpose the face image onto a video or image of another person.
Deepfake will go through a "learning process", based on input data to create models and select appropriate algorithms to continuously process and learn from those models. After a period of "learning", deepfake can merge faces with high accuracy that is difficult to recognize with the naked eye.
Source: https://dantri.com.vn/suc-manh-so/cong-ty-trung-quoc-ra-mat-ai-tao-video-gia-mao-giong-het-nhu-that-20250206125927906.htm
Comment (0)