DeepMind alums launch Haiper with free AI video generation

Just as the world is still picking its jaw up off the floor from OpenAI’s Sora previews, two DeepMind alums are stepping into the ring with their own video generation engine.

Haiper, an AI video startup founded by former DeepMind researchers Yishu Miao and Ziyu Wang, just went live with its consumer platform alongside a fresh $13.8M seed round led by Octopus Ventures, with backing from 5Y Capital.

Free video generation, but with short clips

The two founders originally started out in 2021 trying to tackle 3D reconstruction. But after training on video data, they realized generating video was a far more fascinating puzzle and pivoted roughly six months ago.

Anyone can now jump onto Haiper’s site and generate clips from text prompts for free. There is a catch: you are currently limited to either 2-second HD clips or slightly lower-res 4-second videos.

Beyond text-to-video, the tool lets you animate still images and repaint existing videos into fresh visual styles. An extension tool to make clips longer is also in the pipeline.

Miao says it is way too early to think about subscription paywalls. The team wants to keep things free to build a loyal community, though they are already testing commercial use cases behind the scenes with retail giant JD.com.

Beating the uncanny valley

Haiper is not just building a fun web app. The long-term plan is to develop an underlying foundation model that can be distributed via API and potentially open-sourced down the road.

Right now, the 20-person startup is selectively granting developers access to a closed API to gather feedback.

Miao emphasized that Haiper is not trying to obsess over novelty art styles. Instead, they are laser-focused on fundamental physics, like whether an AI-generated person actually walks naturally or if snowflakes drift accurately.

They face monster competition from OpenAI’s Sora, Runway, Google, and Meta. But investors are betting that raw technical depth will bridge the gap before the video space hits its true commercial breakthrough.

Why it matters

For everyday creators, Haiper is an immediate win simply because it is free to try. While Sora remains locked behind closed doors and rivals charge credits by the second, having a no-cost sandbox to experiment with 2-second visual ideas is genuinely fun.

That said, you won't be directing short films here yet. Two to four seconds is barely enough for a transition, and AI video is clearly still in its GPT-2 era, as a16z’s Justine Moore previously pointed out.

Industry-wide, Haiper represents a crucial trend: frontier researchers breaking off from Big Tech to build open alternatives. Competing head-on with Runway's $230M war chest or OpenAI's computing clusters is a brutal battle.

We think Haiper’s real trump card is its potential to open-source its base models. If they can solve the messy physical dynamics of human motion and hand the weights to developers, they could become the go-to open backbone for video, much like Stable Diffusion did for still images.