The startup teaching a neural network to reverse the physics of a tiny lens - and pull DSLR-grade detail out of the camera already in your pocket.
Here is a strange kind of technology company: one whose entire pitch is a list of things it will not do. It will not add detail. It will not sharpen a face into someone else's face. It will not, in the current vernacular, hallucinate. GLASS Imaging's whole business is built on the boring, honest premise that the detail you want from your phone camera is already there - just badly encoded - and that the right move is to decode it rather than dream it up.
The problem GLASS Imaging is attacking is old and unglamorous. A smartphone camera is a marvel of miniaturization and, simultaneously, a physical compromise. The lens is tiny, the sensor is tiny, the pixels are sub-micron. Light passing through that little stack of glass gets bent and blurred in ways that are annoying but - crucially - predictable. Traditional image signal processors, the chips that turn raw sensor noise into the photo you see, deal with this by applying a sequence of generic corrections: demosaic, denoise, sharpen, fuse a few frames together, hand you a JPEG. Each step is a reasonable approximation. Each step also throws away a little bit of truth.
GLASS Imaging's founders, Ziv Attar and Tom Bishop, spent years at Apple working on exactly this pipeline - they were part of the team behind the iPhone's Portrait Mode. So they know the incumbent playbook intimately, which is presumably why they left to blow it up. Their company, founded in 2019 and based in Los Altos, makes a product called GlassAI. It is a neural image signal processor, or Neural ISP, which is a fancy way of saying: what if the entire camera pipeline were a single trained neural network instead of a chain of hand-tuned steps?
The elegance of that idea is worth pausing on. If you train one model to go end-to-end - straight from the raw sensor data to the finished image - you never pay the compounding tax of doing denoising and then sharpening and then fusion as separate lossy operations. "By training the whole pipeline end-to-end on the RAW, we avoid that compounding loss," is how Shivansh Rao, from the company's machine learning team, puts it. That is the entire technical thesis in one sentence.
*Company-reported, on 0.35-0.75 micrometer pixels vs traditional ISPs. Independent verification pending.
The traditional path treats a photo like an assembly line, each station degrading the part a little. GlassAI collapses the line into a single learned step that knows the exact optics of the specific camera it is correcting.
"Sub-micron pixels encode high-frequency data in complex ways that traditional ISPs fail to decode. That information isn't lost - just hard to recover."
GLASS Imaging doesn't sell you, the shopper, anything. It sells to the companies that make your devices. If you own a phone that ships with GlassAI, you get better zoom, cleaner low-light shots and truer detail without pressing a single extra button. For manufacturers, it comes in two flavors.
A software-only drop-in that goes from sensor RAW to final image, jointly handling demosaicing, denoising, deblurring and multi-frame fusion. Tuned per camera module for up to a claimed 10x performance boost on edge-AI-capable devices.
For cameras where every millimeter and milliwatt matters, GLASS designs the optics and the neural processing together - delivering DSLR-grade imaging to size-, weight- and power-constrained systems like drones, wearables and automotive.
"GlassAI works with manufacturers to boost camera performance, resulting in sharper, more detailed images that remain true to life - with no hallucinations or optical distortions."
The commercial proof arrived in 2026, when GlassAI showed up inside the Honor 600. Here the technology does something quietly radical: instead of a dedicated telephoto lens, the phone's zoom is handled by GlassAI processing crops from a 200-megapixel main sensor - recovering fine detail, cutting noise, and preserving natural color and texture across the zoom range. "We're honored to partner with Honor and to have our Neural ISP technology featured in the 600 series," said CEO Ziv Attar. It is a neat encapsulation of the whole strategy: replace a physical part with a smarter piece of software.
"You should expect to see Glass Imaging in more places over time."
Former Apple engineer; helped build the technology behind the iPhone's Portrait Mode before co-founding GLASS.
Computational imaging researcher and ex-Apple engineer; co-architect of the GlassAI Neural ISP.
| Round | Amount | When | Lead / Investors |
|---|---|---|---|
| SEED | ~$2.4M | 2021 | LDV Capital, GroundUP Ventures |
| EXT. SEED | $9.3M | 2023-24 | Google Ventures, Future Ventures, Abstract Ventures |
| SERIES A | $20M | May 2025 | Insight Partners (lead) + existing investors |
The 2025 Series A, led by Insight Partners, brought the company's disclosed total to roughly $29-31M. The earlier checks came from a notably patient class of investor - Google's venture arm and Steve Jurvetson's Future Ventures among them - the kind willing to fund a hard optics-and-AI problem through the long gap between a compelling demo and a shipping phone.
Ex-Apple imaging engineers Ziv Attar and Tom Bishop set up shop in Los Altos, California.
LDV Capital leads an early investment alongside GroundUP Ventures to build the core technology.
Unveiled at Snapdragon Summit 2024; raises $9.3M from Google Ventures, Future Ventures and Abstract Ventures.
Insight Partners leads a $20M round to push the Neural ISP toward commercial deployment.
GlassAI ships in the Honor 600, powering AI-enhanced zoom from a 200MP sensor.
It builds GlassAI, a neural image signal processor that uses AI to reverse lens aberrations and sensor imperfections, recovering DSLR-grade detail from the small cameras in phones and other devices.
No. GLASS Imaging emphasizes recovering real detail already present in the RAW sensor data, explicitly avoiding hallucinated pixels or fabricated content.
Ziv Attar (CEO) and Tom Bishop (CTO), former Apple engineers who worked on the technology behind the iPhone's Portrait Mode, founded it in 2019.
Its first public commercial deployment is in the Honor 600 smartphone (2026), where it powers zoom imaging; it has also been demonstrated on Qualcomm Snapdragon platforms.
Roughly $29-31M total, including a $20M Series A led by Insight Partners in 2025, with backing from Google Ventures, Future Ventures, Abstract Ventures, LDV Capital and GroundUP Ventures.