ByteDance, TikTok's parent company, just launched an AI video generator
The AI race is heating up
2 min. read
Published on
Share this article
Improve this guide
Read our disclosure page to find out how can you help MSPoweruser sustain the editorial team Read more
Key notes
- ByteDance’s OmniHuman creates realistic deepfake videos from one photo, trained on 18,700 hours of data.
- It can generate and edit videos with natural speech and movement.
- The AI video race intensifies with competitors like OpenAI’s Sora and Google’s Veo 2.
It’s quite worrying how good deepfake AI videos are getting. ByteDance, TikTok’s parent company, has now launched its own OmniHuman that can create realistic full-body videos from a single photo.
ByteDance said that it developed its AI system and trained it on over 18,700 hours of video data, so that OmniHuman can produce videos of people speaking, singing, and performing with impressive realism. The system can also edit existing videos and adjust body proportions and movements.
Here are some of the examples that the Chinese company cherrypicked:
Oh, it even brought Albert Einstein back to life, complete with the voices.
ByteDance researchers said in the model’s papers that “End-to-end human animation has undergone notable advancements in recent years. However, existing methods still struggle to scale up as large general video generation models, limiting their potential in real applications.”
“Unlike existing methods that reduce data due to stringent filtering, our approach benefits from large-scale mixed conditioned data,” the researchers add on why OmniHuman excels against its competitors.
There has been a race to make the best, most realistic AI video generator in recent months.
OpenAI, the AI company behind AI wonder ChatGPT, has hyped up Sora for so long, its text-to-video model. But, as it launched last year during the company’s 12 days of shipmas, Google moved fast by launching its answer: Veo 2, which can even generate AI videos to up to 4K resolution.
User forum
0 messages