Haiper Takes On AI-Generated Hyperrealistic Video
Haiper represents a promising new player in the market for AI-generated video, with a unique focus on quality and ease of use. Could it outcompete OpenAI's Sora?
Haiper, a startup AI company, has recently introduced its own model for video generation, putting it in direct competition with existing players such as Pika Labs, Runway, and OpenAI's Sora in AI-driven storytelling. The company places great emphasis on quality and currently produces short two-second video clips. However, Haiper is aiming high, with ambitions to achieve Sora-like realism in the future.
Toward AGI With a Strong Perceptual Foundation Model
Like OpenAI and Anthropic, Haiper is focused on developing artificial general intelligence (AGI), by building what they call a "powerful perceptual foundation model." Behind Haiper are former engineers from Google DeepMind and TikTok, who combine top-tier machine learning expertise to make creativity more democratic.
Unique Traits Set Haiper Apart From the Crowd
Dr. Yishu Miao, co-founder and CEO of Haiper, emphasizes their unique focus on video and the ability to create impressive generative motion without extensive prompt engineering. The goal is to develop a simple product anyone can use, with a focus on improving training data and labeling to simplify usage.
How Does Haiper Perform?
Although Haiper's clips are short, Dr. Miao points out that their underlying model can intuitively determine correct motion for a video. A demonstration showed that when given shorter prompts and left to control motion on its own, the model can create more natural and realistic scenes.
Challenges and Future Potential
Although Haiper impresses with its ability to create realistic visual representations, there are challenges, particularly when it comes to the movement of larger objects. Compared to its competitors, Haiper hasn't yet achieved the same level of motion consistency, but it's very close.
With a recently secured funding round of $13.8 million, the team behind Haiper is optimistic about being able to quickly address these challenges. The main limitation lies in computing power and training time, but with increased resources, they're expected to soon be able to meet demand.