Competitors in AI video technology heats up as DeepMind alums unveil Haiper


AI-powered video technology is a sizzling market on the again of OpenAI’s releasing the Sora mannequin final month. Two DeepMind alums, Yishu Miao and Ziyu Wang, have publicly launched their video-generation device Haiper with its personal AI mannequin beneath.

Miao, who was beforehand working at TikTok within the International Belief & Security crew, and Wang, who has labored as a analysis scientist for each DeepMind and Google, began engaged on the corporate in 2021 and formally integrated it in 2022.

The pair has experience in machine studying and began engaged on the issue of 3D reconstruction utilizing neural networks. After coaching on video information, Miao talked about to TechCrunch on a name that they came upon that video technology was a extra fascinating downside than 3D reconstruction. That’s why Haiper ended up specializing in video technology roughly six months in the past.

Haiper has raised $13.8 million in a seed spherical led by Octopus Ventures with participation from 5Y Capital. Earlier than that, angels like Phil Blunsom and Nando de Freitas helped the corporate increase a $5.4 million pre-seed spherical in April 2022.

Video-generation service

Customers can go to Haiper’s website and begin producing movies at no cost by typing in textual content prompts. Nonetheless, there are particular limitations. You may solely generate a two-second HD video and barely lower-quality video of as much as 4 seconds.

Haiper's consumer facing website.

Picture Credit: Haiper

The positioning additionally has options like animating your picture and repainting your video in a distinct type. Plus, the corporate is working to introduce capabilities like the power to increase a video.

Miao mentioned that the corporate goals to maintain these options free to be able to construct a group. He famous that it’s “too early” within the startup’s journey to consider constructing a subscription product round video technology. Nonetheless, it has collaborated with firms like to discover business use instances.

We used one of many unique Sora prompts to generate a pattern video: “A number of big wooly mammoths method treading by a snowy meadow, their lengthy wooly fur evenly blows within the wind as they stroll, snow-covered bushes and dramatic snow-capped mountains within the distance, mid-afternoon gentle with wispy clouds and a solar excessive within the distance creates a heat glow, the low digicam view is gorgeous capturing the massive furry mammal with lovely pictures, depth of subject.”

Constructing a core video mannequin

Whereas Haiper is at present specializing in its consumer-facing web site, it desires to construct a core video-generation mannequin that might be provided to others. The corporate hasn’t made public any particulars concerning the mannequin.

Miao mentioned that it has privately reached out to a bunch of builders to strive its closed API. He expects that developer suggestions is essential with the corporate iterating on the mannequin quickly. Haiper has additionally thought of open sourcing its fashions down the road to let folks discover completely different use instances.

The CEO believes that at present, it’s necessary to resolve the uncanny valley downside — a phenomenon that evokes eerie emotions when folks see AI-generated human-like figures — in video technology.

“We aren’t working in fixing issues in content material and magnificence space, however we try work on elementary points like how AI-generated people look whereas strolling or snow falling,” he mentioned.

The corporate at present has round 20 staff and is actively hiring for a number of roles throughout engineering and advertising and marketing.

Competitors forward

OpenAI’s just lately launched Sora might be the most well-liked competitor for Haiper in the mean time. Nonetheless, there are different gamers like Google and Nvidia-backed Runway, which has raised greater than $230 million in funding. Google and Meta even have their very own video-generation fashions. Final 12 months, Stability AI introduced Secure Diffusion Video mannequin in analysis preview.

Rebecca Hunt, a accomplice at Octopus Ventures, believes that within the subsequent three years, Haiper must construct a robust video-generation mannequin to realize differentiation on this market.

“There are realistically solely a handful of individuals positioned to realize this; this is among the causes we wished to again the Haiper crew. As soon as the fashions get to a degree that transcends the uncanny valley and displays the true world and all its physics there shall be a interval the place the functions are infinite,” she instructed TechCrunch over e-mail.

Whereas traders need to spend money on AI-powered video-generation startups, in addition they suppose the expertise nonetheless has a whole lot of room for enchancment.

“It appears like AI video is at GPT-2 degree. We’ve made huge strides within the final 12 months, however there’s nonetheless a strategy to go earlier than on a regular basis shoppers are utilizing these merchandise each day. When will the ‘ChatGPT second’ arrive for video?” a16z’s Justine Moore wrote final 12 months.

The article beforehand acknowledged Geoffrey Hinton as an angel investor. Whereas Hinton has labored with the startup’s founders previous to the corporate’s inception, shouldn’t be concerned as an investor. 

Supply hyperlink


Please enter your comment!
Please enter your name here