Reface and Prisma Co-Founders Launch Mirai to Revolutionize On-Device AI Inference

Two prominent figures in the mobile AI space, Alex Shevts and Dmitri Moiseenkov — co-founders of the viral photo-editing apps Reface and Prisma — have united once again to tackle a critical bottleneck in artificial intelligence: on-device model inference. Their new venture, Mirai, has secured $10 million in seed funding to develop software that allows sophisticated AI models to run efficiently directly on consumer devices such as smartphones and laptops, eliminating the need for constant cloud connectivity.

According to TechCrunch, Shevts and Moiseenkov’s motivation stems from their firsthand experience with the latency, privacy concerns, and bandwidth limitations inherent in cloud-dependent AI applications. While Reface and Prisma gained global popularity for their real-time face-swap and artistic filter capabilities, both founders recognized that the underlying AI models were constrained by the need to transmit user data to remote servers. "We wanted to use AI to create a pipeline that would allow us to enable complex tasks on the phone," Moiseenkov explained in a recent interview, underscoring their vision for truly private, responsive, and offline-capable AI experiences.

Mirai’s technology focuses on optimizing neural network inference through a combination of model compression, quantization, and hardware-aware compilation techniques. Unlike traditional approaches that rely on cloud-based APIs, Mirai’s stack enables large language models (LLMs) and diffusion models to execute locally with minimal latency and power consumption. This is achieved by dynamically adapting model architecture to the specific capabilities of the device’s processor — whether it’s an Apple A-series chip, Qualcomm Snapdragon, or Intel Core processor.

The funding round, led by a consortium of AI-focused venture capital firms, includes participation from investors who have previously backed successful edge-AI startups. Industry analysts note that Mirai enters a growing but still nascent market. While companies like Apple and Google have integrated on-device AI into their ecosystems (e.g., Apple Intelligence, Gemini Nano), few startups have focused exclusively on creating universal, cross-platform inference engines that work across diverse hardware configurations.

"The holy grail of consumer AI is not just better models — it’s models that work everywhere, instantly, without compromising privacy," said Dr. Elena Torres, a senior researcher at the MIT Media Lab, who was not involved with Mirai but has studied edge AI deployment. "Shevts and Moiseenkov bring a rare combination of product intuition and systems engineering expertise that could accelerate adoption beyond niche use cases."

Mirai’s initial applications are expected to target creative professionals and developers seeking to build AI-powered mobile apps without relying on third-party APIs. The company plans to release a developer SDK in early 2025, allowing app creators to integrate Mirai’s inference engine into their own products. Early tests show up to 40% faster inference times and 60% lower energy consumption compared to cloud-reliant alternatives on comparable hardware.

Privacy advocates have welcomed the move. With increasing regulatory scrutiny on data collection — particularly under the EU’s AI Act and California’s privacy laws — on-device processing offers a legally compliant pathway for AI innovation. "No data leaves the device. That’s not just a feature; it’s a fundamental shift in trust architecture," said Mirai’s Chief Privacy Officer, former Google AI ethics lead Maria Chen.

As the AI industry grapples with rising energy costs and server infrastructure demands, Mirai’s approach could represent a scalable, sustainable alternative. With its founding team’s track record of building consumer-facing AI tools and now their focus on the underlying infrastructure, Mirai is poised to become a key enabler of the next generation of private, powerful, and portable artificial intelligence.