Video-generative artificial intelligence startup Haiper said today it has raised $13.8 million in seed funding led by Octopus Ventures to help the company build its foundation model designed for animated visual content.

Founded in late 2021, Haiper is led by two former researchers from DeepMind, Chief Executive Dr. Yishu Miao and Chief Technology Officer Dr. Ziyu Wang, both of whom hold Ph.D.s in machine learning from Oxford University.

Haiper has been working on building an AI foundation model that can take a text prompt written by a user and transform it into a short, animated video. Users can also upload a static image alongside a written prompt and have it generate a creative animation.

Similar developments in visual generative AI include OpenAI’s Sora foundation model, which can generate videos up to a minute long while maintaining stunning visual quality. Other companies that have released similar models in the past include Meta Platforms Inc., Google LLC and Runway AI Inc. At issue for these models has always been quality and the ability to adhere to the text prompt.

According to the team at Haiper, video-generative AI is reaching an inflection point where the AI models need to expand in quality and accuracy. This will require models to understand more than just a vast array of videos and images fed to them, but the context of physics and interaction of information within them such as light, motion, texture and interactions between objects. For example, models need to understand how raindrops splash when they strike surfaces or how hair moves when wind flits across a person’s face.

Miao says the company aims to build a model capable of scaling to a form of artificial general intelligence, the holy grail of AI research, where AI would be capable of internalizing and understand using a human-like comprehension of the world.

“Our end goal is to build an AGI with full perceptual abilities, which has boundless potential to assist with creativity,” said Miao. “Our visual foundation model will be a leap forward in AI’s ability to deeply understand the physics of the world and replicate the essence of reality in the videos it generates. Such advancements lay the groundwork for AI that can understand, embrace, and enhance human storytelling.”

Using Haiper’s app, currently in beta test mode, available for free on the company’s website, users can access the company’s AI models through several different tools to generate videos. The company’s current tools include text-to-video that allows users to write their own imaginative descriptions of what they want to see and bring it to life or upload an image and ask the model to animate it for them with a description.

Additionally, with video repainting, a short video can be modified with different colors, textures and elements using a text description. For example, a bowl of soup could be filled with colorful fish or an animated video of clothes hanging on a line could be changed to display glowing flickering aliens hanging onto the clothesline.

“As the barrier between ideation and implementation lowers, Haiper’s powerful foundation model will give many in the industry the opportunity to develop stunning content and visualize their ideas in ways that were previously impossible,” said Miao.

Image: Haiper