Dallas-based Topaz Labs has released what it calls a breakthrough in local AI processing: a proprietary technology called NeuroStream that the company says can reduce the memory demands of large AI models by up to 95%, bringing professional-grade image and video enhancement to the desktop without relying on the cloud.

Topaz says NeuroStream won’t be limited to its own models. The company is positioning it as foundational technology.

The company—an Emmy Award winner for its AI-powered TV restoration technology—has built a loyal following among creators and professionals since its founding in 2005. Its technology is used by 1.5 million customers, including 20 of the world’s top 50 companies, according to the company.

Now, Topaz says NeuroStream has “the power to change local AI model use across the entire image and video industry.”

Foundational technology—with NVIDIA’s backing

Topaz has collaborated with NVIDIA to optimize NeuroStream for its hardware, and with the technology implemented, models that previously couldn’t run on consumer systems can now operate on every NVIDIA GeForce RTX and RTX PRO GPU, according to the company.

“As the demand for local processing on RTX GPUs continues to grow, NeuroStream provides an opportunity to run complex AI models on nearly all hardware,” said Gerardo Delgado Cabrera, director of product for AI PCs at NVIDIA.

Cabrera said the collaboration with Topaz Labs is part the company’s ongoing efforts to help develop tech optimized for use with NVIDIA-powered devices.

Topaz’s NeuroStream technology also works on AMD GPUs and Apple Macs, according to Xiaoyu Wang, the company’s CTO and an IEEE Fellow, who wrote on LinkedIn that NeuroStream “automatically analyzes the underlying hardware and applies adaptive optimizations” without manual tuning.

Breaking the memory barrier

Topaz NeuroStream reduces VRAM use by 95%, according to Topaz Labs. In this example, a large image/video model’s memory demands drop from 56GB to 2.8GB. [Courtesy of Topaz Labs]

Running large AI models on standard desktops, laptops, and Macs has generally been prohibitive because the models demand far more video memory than consumer hardware provides—often 30GB or more of VRAM, the kind of capacity found only in expensive server-grade GPUs. That has forced users to rely on cloud processing, which means ongoing costs, potential privacy concerns, and dependence on remote infrastructure.

NeuroStream is Topaz’s attempt to break that tradeoff.

“We envision a world where AI models are simply on your device—no cloud needed, no additional usage costs, no specialized hardware, and no security gaps,” CEO Eric Yang said in the announcement. “Our pro customers have been asking for this since we launched our first large, generative model. And now, we’re very excited to make it a reality.”

Wonder 2: The first model to ship with NeuroStream

Wonder 2 is the first Topaz model to ship with NeuroStream optimization. First announced in January 2026, Wonder 2 is designed to denoise, sharpen, and upscale images simultaneously in a single step, and it’s now available in Topaz Photo.

To put the VRAM reduction in concrete terms: Yang wrote on LinkedIn that the Wonder 2 model ordinarily requires roughly 30GB of VRAM—the kind of capacity found in high-end data center GPUs like the NVIDIA H100. With NeuroStream, it runs on 6GB, according to Yang, with “identical results” and only about “2-8% slower” performance on internal benchmarks.

Yang teased the broader rollout on LinkedIn last month. “Cloud processing will always be available as an option, but it shouldn’t be the only way,” he wrote. “Creators shouldn’t need specialized hardware or complex workflows to achieve professional results.”

All of Topaz’s cloud models will be available for local deployment over the coming months, starting with Wonder 2, according to Yang.

Scaling up

Topaz Labs’ new headquarters at Building 900 in Village on the Parkway in Addison. The company nearly quadrupled its office space from 7,000 to about 28,000 square feet. [Courtesy of Topaz Labs]

In addition to winning an Emmy last year, the company was a finalist for the D CEO and Dallas Innovates’ 2026 Innovation Awards for AI and Machine Learning. Its founder, Feng “Albert” Yang, now a Topaz advisor, was named to the inaugural Dallas Innovates AI 75 list honoring top AI visionaries in North Texas.

The NeuroStream launch comes after a significant physical expansion for the company. In January 2025, Topaz announced it would nearly quadruple its Addison headquarters from 7,000 to about 28,000 square feet, becoming the sole office tenant at Building 900 in Village on the Parkway. The company said it expects to invest at least $2.5 million to build out innovation labs and collaborative workspaces.

The company, which was awarded a $200,000 grant by the Town of Addison to support its expansion, plans to add 130 high-tech jobs by the end of 2027.

Nick Lee, president of the office division and principal at NAI Robert Lynn, who represented Topaz Labs in securing the new headquarters, said the deal reflects the company’s “rapid growth in the AI industry.”

Don’t miss what’s next. Subscribe to Dallas Innovates.

Track Dallas-Fort Worth’s business and innovation landscape with our curated news in your inbox Tuesday-Thursday.