🔥 Introducing a New Neural Network for Video Generation

posted  5 Mar 2024
Photo - Introducing a New Neural Network for Video Generation
Deepmind alumni Yishu Miao and Ziyu Wang have launched their own neural network, Haiper, designed for generating videos from text inputs. In a notable departure from OpenAI's competing Sora model, Haiper is already available for all users free of charge, albeit with limitations on the duration of the videos.

Specifically, users can generate HD videos up to two seconds long and lower-quality videos up to four seconds. Beyond this, the application offers the ability to animate images and transform existing videos into new styles. An upcoming update will include the ability to extend the duration of these video clips.

Wang has expressed a desire to maintain these features as free-to-use in an effort to cultivate a comprehensive content collection from the user community. The developers are currently hesitant to introduce a paid subscription model, uncertain of the product's viability for commercial use.

The team is currently engaged in beta testing with third-party developers, aiming to enhance the model's performance. Their long-term vision involves open-sourcing the technology to allow other entities to develop their products, leveraging Haiper's foundational model.