The Creator of Snapchat’s Viral Face Filters Is Redefining How Short-Form Videos Are Made

In 2020, Alex Mashrabov joined Snapchat as director of generative A.I. after the messaging app’s parent company, Snap, acquired his startup. Mashrabov was the brain behind Snapchat’s popular face filters, a signature feature that helped Snap stay competitive during the COVID years, when TikTok was spending billions annually to cement its foothold in the U.S. The success of those filters earned Mashrabov the industry nickname “the godfather of consumer generative A.I.”

In late 2023, Mashrabov left Snap to launch his own venture, Higgsfield, which builds a suite of A.I. tools designed to dramatically shorten video production and editing time. The product taps into an ever-growing demand for video content among marketers and influencers, where relevance increasingly depends less on polish than on frequency. The app has gained traction quickly, attracting celebrity users including Madonna, Snoop Dogg and Will Smith—entirely organically, according to the company.

“Social media is the largest media in the world and the most important media in the world, but the rules of the game are very fierce,” Mashrabov told Observer. “If someone wants to get popular on social media, they need to post several videos a day. Traditional production methods don’t get you there, but generative A.I. gets you there by shortening the production cycle from weeks to basically hours.”

That pitch has resonated with investors. Higgsfield recently raised $130 million in a Series A round at a valuation of $1.3 billion, making it one of the highest-valued A.I. video startups to date. Investors in the round include Accel, Menlo Ventures and AI Capital Partners, the U.S.-based fund affiliated with Alpha Intelligence Capital.

The company is also scaling rapidly on the revenue side. Higgsfield says it has reached an annual recurring revenue run rate (ARR) of roughly $200 million, up from about $100 million just two months earlier. While such figures are self-reported and difficult to compare directly across companies and eras, Higgsfield has described its recent growth as outpacing the early ARR trajectories of companies such as OpenAI, Slack and Zoom.

Under the hood, Higgsfield combines diffusion-based A.I. video generation models with large language models trained to understand camera motion, scene composition and visual continuity. Users can describe a scene, choose a camera movement and set a mood, and the system generates short clips—typically three to ten seconds long—optimized for platforms such as TikTok, Instagram Reels and YouTube Shorts.

The interface is designed to feel more like a virtual studio than a traditional editing suite. Users start in a video workspace and then select from specialized modules depending on the task. The “Cinema Studio” tool, for example, focuses on directed camera movement, offering preset motions like dolly-ins, aerial pullbacks, pans and push-ins that mirror classical cinematography. Other tools include “LipSync” for dialogue-driven clips and “Click to Ad,” which turns static product images into short promotional videos.

Higgsfield is emerging amid a fast-evolving A.I. video market that has accelerated since 2023, driven by demand for short-form content and breakthroughs in underlying models. Advances in diffusion techniques, transformer-based video understanding and large-scale multimodal training have pushed A.I. video generation from experimental demos toward practical, production-ready tools.

Transformers, a type of neural network architecture introduced by Google in 2017, are central to that shift. Unlike earlier models that processed data step by step, transformers learn context by modeling relationships across an entire sequence at once, whether words in a sentence or frames in a video.

“Transformer models going mainstream is a major technology shift,” Mashrabov said. “Those transformer models can fully model the world. It’s exciting to see that we went from generating very small, 64-by-64 images to producing consistent, minutes-long videos at 4K resolution.”

Large players are now racing into the same space. OpenAI, for example, has unveiled Sora, a text-to-video model capable of generating longer, more coherent clips from written prompts. Higgsfield, however, is betting on a different workflow it calls “click-to-video,” where users can create clips with minimal input, such as a single image, rather than lengthy prompts.

“It’s very difficult for a regular social media marketer to write 100-word prompts,” he said. “It gets complex because you need to include all the details you want, plus negative prompts for what you don’t want. For brand advertisers especially, it’s important to avoid controversy, which means even longer, more careful prompting. Click-to-video removes a lot of that friction.”