The rise of generative AI has taught us all to be wary of the text and images we come across on the internet, but video isn't far behind. On the heels of OpenAI's Sora video creation AI, a startup founded by former Google DeepMind researchers has announced its own video AI. Unlike Sora, you can try Haiper right now, but it will only spit out a few seconds of video.
Yishu Miao and Ziyu Wang began working on Haiper in 2021, founding the company in 2022. Initially, the pair were interested in using AI for 3D reconstruction. However, the process of training the AI on video data was so fascinating they decided to pivot to video generation about six months ago.
The company's progress is on display for all to see. The Haiper video snippets aren't as impressive as the samples posted by OpenAI, but very few people have had a chance to try Sora. Haiper will let you create a two-second HD video from a text prompt or generate a slightly longer clip at a lower resolution. Haiper can also take a static image and create an animated version or "repainting" a video in a different style. There are also plans to add the ability to extend a video with AI.
Currently, it takes about a minute to create each sample video. That's a function of the computing power Haiper has at its disposal. All generative AI systems require expensive AI accelerators like the Nvidia H100, which can cost upwards of $30,000 per card. As the company progresses in development, you can expect to see speed increases.
Mao tells TechCrunch that he intends to keep these basic features available for free, but it's still too early even to consider creating a paid service tier. The company has started contacting developers to test its closed API—this feedback will be vital to improve the model before it is released widely. Company leadership is even considering making the model open source so other groups can explore different uses for the video generator.
Tweet (opens in a new tab)
Sora was not the first AI to create videos, but its impressive results and the OpenAI name pushed it into the public consciousness. We can expect an AI video goldrush not unlike what we've seen with text and image-generating models over the last year. This is a much more difficult problem to solve, but there is no shortage of companies raising money to tackle it.
