Haiper 2.0 is a generative video model developed by Haiper, a London-based AI startup founded by alumni from Google DeepMind and TikTok. It was released as a significant advancement over the platform's initial version, aiming to provide hyper-realistic video generation with improved temporal consistency and faster processing times. The model serves as a foundation for various visual content tools, including text-to-video, image-to-video, and video-to-video transformations.
Technical Details
The model is built on a Diffusion Transformer (DiT) architecture, a design that combines the scaling properties of transformers with the generative capabilities of diffusion models. This architecture allows Haiper 2.0 to handle complex spatial and temporal data more effectively than standard diffusion-only models. It is reported to feature over 10 billion parameters, enabling high-fidelity rendering of textures, lighting, and motion.
Core Features
Key features of Haiper 2.0 include the ability to generate high-definition video at 1080p resolution, with support for 4K. It introduced a Video Templates feature, which allows users to animate static images into specific pre-defined motion patterns like dancing or hugging. The model also supports precise control via keyframe conditioning and offers an integrated HD upscaler to improve the clarity of generated clips.