Following Pika, Another Chinese Entrepreneur Enters AI Video Generation, Securing $10 Million in Funding
-
On March 5th, two Chinese entrepreneurs, Yishu Miao and Ziyu Wang, launched an AI startup featuring their self-developed AI model for video generation, named Haiper.
Both founders of Haiper hold PhDs in Machine Learning from the University of Oxford, bringing extensive knowledge and practical experience in the field. Yishu Miao previously worked at TikTok, while Ziyu Wang served as a researcher at Google and DeepMind. Haiper was officially established in 2022, initially focusing on AI 3D modeling. According to the founders, the team decided to shift their entrepreneurial direction to AI video generation half a year ago. Within six months, they launched the beta version of Haiper.
Currently, Haiper has completed two rounds of financing, with a total funding amount of $19.2 million. The most recent round raised $13.8 million, led by Octopus Ventures with participation from 5Y Capital.
A completely free video generation tool Unlike other video generation products that adopt subscription payment models, Haiper is currently completely free. Users only need to visit Haiper's official website to use it.
However, currently Haiper can only generate HD videos up to 2 seconds long and SD videos up to 4 seconds long. Currently, mainstream products like Pika, Runway, and Stable Video can directly generate videos up to 4 seconds long (extendable to 16 seconds), while Sora can generate videos up to 60 seconds long, but it's currently only available as a demo.
In addition to video generation, Haiper also provides features like video cartoonization, photo cartoonization, and video repainting. Future updates will introduce video extension functionality. Regarding the payment model, founder Miao Yishu mentioned that Haiper's video generation feature is currently not mature enough, making it "too early" to implement a paid subscription model. The company's plan is to maintain a free model and focus on building a community ecosystem. In terms of ToB (business-to-business) applications, Haiper has already partnered with companies like JD.com to explore commercial scenarios.
Haiper (top) and Sora (bottom) videos generated with the same prompt
Ultimate goal: Building a general-purpose AI model Currently, Haiper is concentrating on consumer-facing website video generation products, but the founders revealed their ambition to develop a more widely applicable video generation model. The company has privately contacted some developers to test Haiper's API and is refining its AI model based on their feedback. Additionally, Haiper is considering open-sourcing its model to allow more users to optimize it and explore richer use cases.
For the next phase, the Haiper team stated that AI video generation is reaching an inflection point, where models need to improve quality and accuracy to address the Uncanny Valley problem.
The Uncanny Valley refers to the sense of aversion people feel when viewing highly realistic human-like figures generated by artificial intelligence. This requires the model not only to comprehend vast amounts of training materials in the form of videos and images but also to understand the physical laws of the real world and the ways information interacts within it, such as light, motion, texture, and interactions between different objects. For example, the AI needs to understand how raindrops splash when hitting a surface or how hair moves when wind blows across a face. This is currently a key focus of Haiper's technological development.
In the future, founder Miao Yishu stated that Haiper's goal is to build a general AI model with full perceptual capabilities. The current development of a foundational vision model will enable AI to deeply understand the world and better replicate reality in generated videos, laying the groundwork for a general model.
While the vision is ambitious, after a year of development, the AI video generation field has become highly competitive. OpenAI's Sora leads the pack technologically, and Runway, which has raised over $230 million, already has a significant user base and first-mover advantage. Additionally, tech giants like Google, Meta, and Stability AI have all launched their own video generation models. As a new product, Haiper faces significant challenges in entering the market and achieving future commercialization. However, according to a16z's commentary article, current AI video tools are roughly at a development level similar to GPT-2 at that time. This means that while AI video tools have achieved technological breakthroughs, they still lack application scenarios suitable for ordinary users.
If Haiper can make breakthroughs in video realism technology and truly find practical application scenarios that can be implemented, achieving differentiation, then entering the market and even surpassing other similar products like Sora is not impossible.