In a constantly evolving market, AI-powered video generation has emerged as a highly sought-after technology, especially after OpenAI’s recent release of the Sora model. But the spotlight now shines on Haiper, a video generation tool developed by former Deepmind employees Yishu Miao and Ziyu Wang, equipped with its own AI model beneath.
Hailing from a background in machine learning, Miao and Wang initially focused on using neural networks for 3D reconstruction. However, after delving into video data, Miao revealed in an interview with TechCrunch that they were drawn to the captivating challenge of video generation, leading them to establish Haiper about six months ago. By this year, the company had officially incorporated.
After securing $13.8 million in funding from a seed round led by Octopus Ventures with 5Y Capital also contributing, Haiper has garnered attention from prominent angels such as Geoffrey Hinton and Nando de Freitas, who helped raise a pre-seed round of $5.4 million in April 2022.
Video generation service
Now live and accessible on Haiper’s website, users can experiment with generating videos for free by simply typing in text prompts. However, certain restrictions apply, as only videos in HD and no longer than two seconds in duration, or slightly lower-quality videos of up to four seconds, can be created.
The website also offers additional features, including bringing images to life and transforming videos with unique styles. And as the company continues to expand, they plan to introduce capabilities such as video extensions.
Miao expressed that the primary goal is to foster a community, hence why these features are free and a subscription-based business model is not yet in the works. Nevertheless, Haiper has already collaborated with major companies like JD.com to explore potential commercial uses.
We decided to use one of the original prompts from Sora to generate a sample video: “Several giant wooly mammoths approach treading through a snowy meadow, their long wooly fur lightly blows in the wind as they walk, snow-covered trees and dramatic snow-capped mountains in the distance, mid-afternoon light with wispy clouds and a sun high in the distance creates a warm glow, the low camera view is stunning capturing the large furry mammal with beautiful photography, depth of field.”
Building a core video model
While Haiper currently focuses on their consumer-facing website, they are also developing a strong video generation model that they could eventually offer to others. However, the company has not disclosed any details about the model to the public.
Miao affirmed that they have privately reached out to several developers to test their closed API, and that feedback from these developers is crucial in the company’s rapid iteration of the model. They have also contemplated open-sourcing their model in the future, allowing others to explore potential use cases.
The CEO recognizes that the “uncanny valley” problem – where viewers feel uneasy when faced with AI-generated human-like figures – is a critical hurdle to cross in video generation.
“We are not focused on solving issues related to content and style,” he explained, “but rather fundamental problems such as the appearance of AI-generated individuals while walking or snow falling.”
Having grown to a team of 20 employees, Haiper is actively recruiting for various roles in engineering and marketing.
Competition ahead
Currently, Haiper’s biggest competitor in the market is likely OpenAI’s recent release, Sora. However, the market also houses other players like Google and Nvidia-backed Runway, which have raised over $230 million in funding. On the other hand, Google and Meta each have their own video generation models, while Stability AI announced the research preview of their Stable Diffusion Video model last year.
Rebecca Hunt, a partner at Octopus Venture, believes that Haiper needs to develop a powerful video generation model in the next three years to stand out in this market.
“Realistically, only a select few companies have the potential to achieve this; and we are confident that the Haiper team is one of them,” she shared with TechCrunch via email. “Once the models have reached a level that surpasses the uncanny valley and accurately represent the real world and its physics, the possibilities will be endless.”
Despite investors’ eagerness to support AI-powered video generation startups, they also recognize that the technology still has a lot of room for growth.
“Currently, AI video technology is at a GPT-2 level. While significant progress has been made in the past year, it will likely be some time before everyday consumers integrate these products into their daily lives,” a16z’s Justine Moore wrote last year.
