OpenAI caused quite a stir in the technology community a few months ago with their groundbreaking generative AI model, Sora. This model, named after the protagonist in the popular video game franchise Kingdom Hearts, has the ability to turn written descriptions into completely original videos without needing any traditional camera equipment or film crews. However, access to Sora has been limited, and it seems that OpenAI is primarily targeting the model towards well-funded creatives such as Hollywood directors. This leaves little room for hobbyists or small-time marketers to explore and utilize this powerful tool.
Realizing the potential for a more accessible AI video creation platform, Alex Mashrabov, former head of generative AI at Snap, took matters into his own hands and launched Higgsfield AI. Built upon advanced AI technology, Higgsfield is designed for more personalized and tailored applications.
“Our goal with Higgsfield is to cater to creators of all types,” Mashrabov explained in an interview with TechCrunch. “We want to make video creation and editing accessible to everyone – from regular users looking to create fun content with friends, to social content creators interested in trying out new content formats, to social media marketers who are looking to make their brand stand out.”
Mashrabov joined Snap after his previous startup, AI Factory, was acquired by the company for a whopping $166 million in 2020. During his time at Snap, Mashrabov played a key role in developing popular products such as AR effects and filters for Snapchat, including the controversial MyAI chatbot.
Higgsfield, co-founded by Mashrabov and AI researcher Yerzat Dulat, offers a unique combination of pre-generated clips, a tool to upload reference media (like images and videos), and a prompt editor that allows users to describe the characters, actions, and scenes they want to see in their videos. With Higgsfield’s first app, Diffuse, users can create videos from scratch or even insert themselves (or their digital likeness) into AI-generated scenes.
Mashrabov promises that Higgsfield’s custom text-to-video model is able to create highly realistic movements and expressions, making it stand out among other generative video startups like Runway and Haiper.
One of Higgsfield’s unique selling points is its mobile-first and social media-forward approach. By prioritizing the development of iOS and Android apps over traditional desktop workflows, the platform allows creators to produce compelling social media content anytime, anywhere.
“Building on a mobile platform has allowed us to prioritize user-friendly features and ease of use from the very beginning,” Mashrabov said.
Despite its innovative approach, Higgsfield is running on a lean budget. The platform’s generative models were developed by a team of just 16 people in less than nine months, and were trained on a modest cluster of only 32 GPUs. In comparison, OpenAI uses tens of thousands of GPUs to train their models. So far, Higgsfield has raised $8 million in funding, with a large portion of it coming from a recent round of seed funding led by Menlo Ventures.
To stay ahead of its competitors, Higgsfield plans to allocate the seed money towards developing an improved video editor that will allow users to modify characters and objects in videos, as well as train more powerful video generation models specifically for social media use. Mashrabov sees social media, particularly social media marketing, as Higgsfield’s main revenue stream.
While Diffuse is currently free to use, Mashrabov envisions a future in which marketers can pay for premium features, or a subscription for access to volume or large-scale campaigns. He believes that Higgsfield’s realistic and versatile video generation abilities make it a valuable tool for social media marketers who are looking to optimize production budgets and timelines while still delivering impactful content.
However, like any generative AI platform, Higgsfield is not immune to certain challenges and concerns. One major concern is the potential for copyright infringement, as generative AI models can unintentionally replicate copyrighted material used in their training data. Mashrabov did not reveal the source of Higgsfield’s training data, but he did mention that it comes from multiple publicly available sources. Additionally, Higgsfield relies on user data to train its models, which may not sit well with some business customers. To address these concerns, Mashrabov ensured that users have the option to delete their data at any time through the app.
Another concern surrounding generative AI is its potential for misuse, as seen with the recent rise of deepfakes on social media. Higgsfield, with its digital “cloning” capabilities, could potentially make it easier to steal creators’ content. Mashrabov acknowledges this and claims that the platform employs a combination of automated and manual moderation to prevent abuse. Higgsfield has also chosen to gradually roll out the product and monitor its use in select markets before expanding to a wider audience.
Despite these challenges, Mashrabov remains confident in Higgsfield’s capabilities and believes that it has the potential to revolutionize content production for social media marketers.
“We are excited to pave the way for consumer-friendly video generative AI solutions,” Mashrabov stated. “With Higgsfield, we hope to enable creators to produce high-quality, realistic videos in a cost-effective and efficient manner, ultimately making it easier for them to stand out in the crowded social media landscape.”
