DxO PhotoLab High-ISO Noise Reduction Workflow

The example I'm using here was photographed on a Sony RX100 VII at ISO 5000. That's deliberately useful for this demonstration because the RX100 VII uses a relatively small 1-inch sensor compared with the larger sensors found in many interchangeable-lens cameras.

Push a file like this into a demanding low-light situation and noise becomes particularly obvious in darker areas, smooth surfaces and areas of subtle colour.

The problem isn't simply that the photograph looks grainy. Heavy noise can begin interfering with the detail you're actually trying to preserve.

DeepPRIME is built around machine-learning processing of RAW data. DxO combines denoising and demosaicing — the stage where the sensor's RAW information is reconstructed into a usable colour image — rather than treating them as completely separate processes.

DeepPRIME XD3 is the more detail-focused version of that processing and is designed for particularly demanding RAW files.

With PhotoLab 9.6, XD3 supports both Bayer and X-Trans sensors, which means the Sony Bayer-sensor file used in this workflow can now use the same latest-generation XD3 processing.

Why this matters visually

I don't want noise reduction to give me a technically clean photograph that no longer looks photographic.

The value of the more sophisticated processing is finding the balance between suppressing the unwanted noise and retaining the information that makes the photograph feel detailed.

One of the main things I wanted to demonstrate in the video is how little work can actually be required.

Rather than immediately reaching for every available control, I select DeepPRIME XD3 and look carefully at what the default processing does to the photograph.

On this ISO 5000 file the transformation is already significant. A large amount of distracting luminance and colour noise disappears while important details remain visible.

Aggressive noise reduction can make almost any photograph look cleaner if you're prepared to sacrifice enough detail.

That's not the result I'm looking for.

In this photograph I pay attention to the structural detail in the train, the platform and the surrounding architecture. Those areas give me useful visual references for whether the processing is genuinely recovering the photograph or simply smoothing everything indiscriminately.

Look beyond 100% pixel peeping

Magnified inspection is useful when judging technical processing, but the finished photograph also has to work at normal viewing size.

I want to understand what happens at pixel level without losing sight of how the whole image actually feels.

I tend to judge this image by image rather than applying the same philosophy to every RAW file.

A clean low-ISO photograph may not need the same level of intervention as a small-sensor image photographed at ISO 5000 in difficult available light.

The Sony RX100 VII is an interesting example because its portability is one of the reasons you'd choose it in the first place.

But the relatively small sensor means high-ISO files can become more demanding than those from larger full-frame cameras.

Good noise reduction doesn't magically turn one sensor into another, but it can give you considerably more confidence in the usable files coming from a compact camera.

That's important to the way I think about workflow. Software isn't simply there to rescue mistakes. Sometimes it allows smaller, lighter equipment to remain useful in situations where you might otherwise have felt forced to carry something larger.

• Travel and street photography after dark.
• Available-light event photography.
• Indoor photography where flash isn't appropriate.
• Wildlife and action where shutter speed has to remain high.
• Concert and performance photography.
• Compact-camera photography in poor light.
• RAW files where shadow recovery reveals additional noise.

In all of those situations I would rather make the photograph at the ISO I need than miss the moment because I was frightened of noise.

The processing workflow happens afterwards. The photograph still has to exist first.