Chips&Media and Visionary.ai have announced a major advancement in digital imaging by developing the world’s first fully AI-based image signal processor (ISP). This innovative collaboration aims to replace traditional hardware-based ISPs, which have been integral to digital imaging for decades, with a software-driven approach utilizing neural processing units (NPUs).
The new system seeks to enhance the entire image formation process by shifting it entirely into software. According to Oren Debbi, co-founder and CEO of Visionary.ai, “This is the first full end-to-end ISP pipeline that runs entirely on an NPU, without relying on a hardware ISP at all.” This development allows for real-time tuning, retraining, and updating of video processing capabilities, making it particularly beneficial for low-light video scenarios where traditional ISPs often struggle.
Traditional ISPs have remained largely unchanged in their hardware architecture, relying on fixed-function components that limit flexibility and adaptability. Chips&Media and Visionary.ai view this rigidity as a barrier to meeting the evolving demands of modern imaging applications, which now extend from smartphones to autonomous vehicles and virtual reality devices. With their new approach, the companies hope to overcome these limitations and provide a more scalable solution.
By processing RAW sensor data directly on an NPU or GPU, the AI-based ISP enables manufacturers to adjust settings and optimize performance through over-the-air updates, eliminating the need for manual tuning of hardware. Visionary.ai has developed a custom neural network for each image sensor, and their automated training platform can create new models in mere hours using only a few short video clips. This innovation significantly reduces integration overhead, allowing for faster adaptation across different sensors and platforms.
While AI-enhanced ISPs are already present in smartphones and cameras, Debbi argues that these systems remain overly hardware-centric. Existing pipelines typically add neural blocks to fixed-function ISPs, which do not process core RAW data. He emphasizes that the new system represents a shift to a “neural-first” image formation pipeline, where the core image processing is no longer reliant on traditional hardware.
The implications of this shift are most pronounced in challenging lighting conditions. Standard ISP pipelines often compromise detail to suppress noise, leading to artificial-looking images with artifacts. Debbi notes, “You see the biggest difference in the hard cases where classic ISPs have to trade off detail, noise, and artifacts — very low light, high dynamic range, and mixed lighting.” The new AI pipeline promises cleaner shadows, stable colors, and fewer temporal artifacts in video.
The focus on video processing does not exclude still photography. Debbi acknowledges that the underlying architecture can also enhance still images, although market demand has primarily driven their current focus on video applications. As phone cameras increasingly utilize techniques like HDR and low-light stacking, the potential for improved results through decoupled processing becomes evident.
This software-defined AI ISP can adapt to varying hardware capabilities, enabling manufacturers to implement advanced camera functions without the need for extensive silicon resources. Debbi explains that the AI system is designed to run efficiently, consuming power comparable to traditional ISPs while offering superior performance. “We’re able to run on a very small NPU and consume only slightly more than imaging with a traditional ISP, and that gap continues to shrink,” he adds.
Chips&Media’s WAVE-N NPU serves as a reference implementation for the AI ISP, showcasing a real-time neural imaging pipeline operating on video-focused AI hardware. The AI ISP is hardware-agnostic, allowing manufacturers to adapt the software pipeline to various NPUs or GPUs based on their system architecture and cost objectives. This flexibility can lead to significant advancements in imaging, including improved HDR, better exposure fusion, and enhanced segmentation for diverse applications.
Despite the ambitious goals of this collaboration, both companies recognize that traditional ISPs will not disappear immediately. However, they believe the trend is clearly shifting toward programmable AI computing. Visionary.ai can integrate its AI ISP into existing silicon within months through software updates, while also allowing for reductions in dedicated ISP silicon area in future chip generations.
Ultimately, Debbi argues that “software updates faster than silicon, adapts better to new sensors and use cases, and ultimately reduces cost and complexity.” As the imaging industry evolves, the collaboration between Chips&Media and Visionary.ai is poised to reshape how image quality is delivered, updated, and scaled. The full AI-based ISP is set to debut at CES 2026, marking a significant milestone in the transition towards AI-driven imaging solutions.