ByteDance launches Goku AI to challenge OpenAI Sora

ByteDance has launched Goku AI, a model that has made breakthroughs in image and video generation technology. It adopts innovative modified rheology deformation design, challenging OpenAI's Sora, showing strong developments in China's AI technology.

The method used by Goku AI, unlike traditional diffusion technology, provides smoother content generation. It can generate images and videos from text to create realistic scenes including complex motion and multi-objective dynamic lighting.

Training relies on huge data sets (160 million image text pairs and 36 million video text pairs) to be screened and optimized through a variety of techniques. The training is carried out in stages, gradually improving from low resolution to high resolution.

Its Transformers architecture allows Goku AI to generate content faster and more efficiently, achieving better results in rare training steps than other models.

To support this model, ByteDance provides powerful computing resources and technologies, including efficient data processing and state management.

Although Goku AI has performed well in generating content, it has also raised concerns about deep formats and information security. As an open source model, it increases the complexity of AI technology control and regulation and becomes the focus of discussion on AI leadership.