Exclusive: Introducing Haiper 1.5 – The New AI Video Generation Model Taking on Sora and Runway

As AI-generated content becomes increasingly prevalent, startups in the sector are enhancing their offerings. Recently, RunwayML launched a more realistic model for video generation. Now, London-based Haiper, founded by former Google DeepMind researchers Yishu Miao and Ziyu Wang, introduces its visual foundation model: Haiper 1.5.

Available on Haiper's web and mobile platforms, Haiper 1.5 is an incremental upgrade that enables users to create 8-second clips from text, image, and video prompts—doubling the length of output from the original model. Additionally, Haiper has introduced an upscaling feature that enhances content quality and plans to add image generation capabilities.

Just four months after emerging from stealth mode, Haiper has already attracted over 1.5 million users, despite being less funded than its competitors. With this new model, the company aims to expand its user base and compete with established players like Runway.

Miao, Haiper's CEO, stated, “The race in video generative AI isn't just about model power but also the complexity of what these models can recreate. Our distributed data processing and scaled training allow us to enhance our foundation model continually. This update signifies our commitment to producing not only beautiful videos but also realistic imagery.”

What Does Haiper AI Bring to Users?

Launched in March, Haiper is positioned alongside Runway and Pika, offering a streamlined platform for video generation. Users input a text prompt, and the model generates corresponding content, allowing adjustments for characters, objects, backgrounds, and artistic styles.

Previously, Haiper produced clips of 2-4 seconds; the new model increases this to 8 seconds in response to user feedback. Users can also extend existing 2 and 4-second clips to 8 seconds, mirroring features from other AI tools like Luma's Dream Machine.

Miao commented, “The enthusiastic response to our models has motivated us to extend video generation to eight seconds, pushing the boundaries of this technology.”

New Features and Enhancements

The latest update also permits users to generate clips of any length in standard or high definition. The integrated upscaler enhances video quality to 1080p with a single click, compatible with both newly generated and existing content.

Moreover, Haiper will add a new image model, allowing users to create images from text prompts and integrate them into the video generation process for seamless results. This will enable better content testing and refinement before animation.

“At Haiper, we prioritize user feedback and innovation. The launch of our upscaler and the upcoming Text2Image tools illustrates our commitment to the community,” Miao emphasized.

Advancing Towards AGI

While Haiper’s updates look promising, they are pending wider community testing. Some features, such as the eight-second videos and upscaler, are currently exclusive to users on the Pro plan, priced at $24/month when billed annually. Miao indicated plans to broaden access to these features, including a credit system for longer video generation and a free release of the image model soon.

Initial quality feedback suggests that while the two-second videos are reliable, longer videos are inconsistent. However, with ongoing updates, Haiper aims to enhance the quality of its outputs significantly. The goal is to improve the models’ understanding of the world, progressing towards artificial general intelligence (AGI) capable of generating lifelike content by accurately depicting light, motion, texture, and object interactions.

Miao concluded, “Each frame carries intricate visual information. To create content that truly reflects reality, AI must grasp the complexities of the physical world. Such advancements could revolutionize not just content creation but also sectors like robotics and transportation.”

As Haiper continues to evolve, it will be interesting to observe how it competes against established rivals such as Runway, Pika, and OpenAI in the rapidly developing field of AI video generation.